shiqianyan commited on
Commit
e7eae64
·
verified ·
1 Parent(s): e777b30

Release TangYuGe GPT-SoVITS v2ProPlus: fine-tuned weights, pretrained models, profiles and scripts

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +3 -32
  2. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-hubert-base/config.json +72 -0
  3. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-hubert-base/preprocessor_config.json +9 -0
  4. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-hubert-base/pytorch_model.bin +3 -0
  5. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large/config.json +34 -0
  6. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large/pytorch_model.bin +3 -0
  7. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large/tokenizer.json +0 -0
  8. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/fast_langdetect/lid.176.bin +3 -0
  9. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/s1v3.ckpt +3 -0
  10. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/sv/pretrained_eres2netv2w24s4ep4.ckpt +3 -0
  11. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/v2Pro/s2Dv2ProPlus.pth +3 -0
  12. GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/v2Pro/s2Gv2ProPlus.pth +3 -0
  13. GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/MONOPHONIC_CHARS.txt +0 -0
  14. GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/POLYPHONIC_CHARS.txt +0 -0
  15. GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/bopomofo_to_pinyin_wo_tune_dict.json +1 -0
  16. GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/char_bopomofo_dict.json +0 -0
  17. GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/config.py +19 -0
  18. GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/g2pW.onnx +3 -0
  19. GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/record.log +1005 -0
  20. GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/version +1 -0
  21. LICENSE +21 -0
  22. README.md +220 -0
  23. THIRD_PARTY.md +37 -0
  24. configs/tts_infer_pack.yaml +57 -0
  25. manifests/checksums.sha256 +73 -0
  26. manifests/references.yaml +338 -0
  27. manifests/training_report.md +73 -0
  28. profiles/apology.yaml +29 -0
  29. profiles/bright_fireworks.yaml +29 -0
  30. profiles/cafeteria_ordering.yaml +29 -0
  31. profiles/casual_invite.yaml +29 -0
  32. profiles/command_gentle.yaml +29 -0
  33. profiles/exclaim.yaml +29 -0
  34. profiles/food_daily.yaml +29 -0
  35. profiles/high_brightness_exclaim.yaml +29 -0
  36. profiles/intimate_reassure.yaml +29 -0
  37. profiles/lively_daily.yaml +29 -0
  38. profiles/low_energy_discussion.yaml +29 -0
  39. profiles/main_hybrid_mid.yaml +30 -0
  40. profiles/morning.yaml +29 -0
  41. profiles/narration_calm.yaml +29 -0
  42. profiles/neutral_polite.yaml +29 -0
  43. profiles/question_clear.yaml +29 -0
  44. profiles/question_firm.yaml +29 -0
  45. profiles/romantic_soft.yaml +29 -0
  46. profiles/soft_daily.yaml +29 -0
  47. references/apology_script32_L0040.wav +3 -0
  48. references/bright_fireworks_script22_L0948.wav +3 -0
  49. references/cafeteria_ordering_script1_L0930.wav +3 -0
  50. references/casual_invite_script21_L0249.wav +3 -0
.gitattributes CHANGED
@@ -1,35 +1,6 @@
1
- *.7z filter=lfs diff=lfs merge=lfs -text
2
- *.arrow filter=lfs diff=lfs merge=lfs -text
3
- *.bin filter=lfs diff=lfs merge=lfs -text
4
- *.bz2 filter=lfs diff=lfs merge=lfs -text
5
  *.ckpt filter=lfs diff=lfs merge=lfs -text
6
- *.ftz filter=lfs diff=lfs merge=lfs -text
7
- *.gz filter=lfs diff=lfs merge=lfs -text
8
- *.h5 filter=lfs diff=lfs merge=lfs -text
9
- *.joblib filter=lfs diff=lfs merge=lfs -text
10
- *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
- *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
- *.model filter=lfs diff=lfs merge=lfs -text
13
- *.msgpack filter=lfs diff=lfs merge=lfs -text
14
- *.npy filter=lfs diff=lfs merge=lfs -text
15
- *.npz filter=lfs diff=lfs merge=lfs -text
16
- *.onnx filter=lfs diff=lfs merge=lfs -text
17
- *.ot filter=lfs diff=lfs merge=lfs -text
18
- *.parquet filter=lfs diff=lfs merge=lfs -text
19
- *.pb filter=lfs diff=lfs merge=lfs -text
20
- *.pickle filter=lfs diff=lfs merge=lfs -text
21
- *.pkl filter=lfs diff=lfs merge=lfs -text
22
- *.pt filter=lfs diff=lfs merge=lfs -text
23
  *.pth filter=lfs diff=lfs merge=lfs -text
24
- *.rar filter=lfs diff=lfs merge=lfs -text
 
 
25
  *.safetensors filter=lfs diff=lfs merge=lfs -text
26
- saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
- *.tar.* filter=lfs diff=lfs merge=lfs -text
28
- *.tar filter=lfs diff=lfs merge=lfs -text
29
- *.tflite filter=lfs diff=lfs merge=lfs -text
30
- *.tgz filter=lfs diff=lfs merge=lfs -text
31
- *.wasm filter=lfs diff=lfs merge=lfs -text
32
- *.xz filter=lfs diff=lfs merge=lfs -text
33
- *.zip filter=lfs diff=lfs merge=lfs -text
34
- *.zst filter=lfs diff=lfs merge=lfs -text
35
- *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
1
  *.ckpt filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  *.pth filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.onnx filter=lfs diff=lfs merge=lfs -text
5
+ *.wav filter=lfs diff=lfs merge=lfs -text
6
  *.safetensors filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-hubert-base/config.json ADDED
@@ -0,0 +1,72 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/data/docker/liujing04/gpt-vits/chinese-hubert-base",
3
+ "activation_dropout": 0.1,
4
+ "apply_spec_augment": true,
5
+ "architectures": [
6
+ "HubertModel"
7
+ ],
8
+ "attention_dropout": 0.1,
9
+ "bos_token_id": 1,
10
+ "classifier_proj_size": 256,
11
+ "conv_bias": false,
12
+ "conv_dim": [
13
+ 512,
14
+ 512,
15
+ 512,
16
+ 512,
17
+ 512,
18
+ 512,
19
+ 512
20
+ ],
21
+ "conv_kernel": [
22
+ 10,
23
+ 3,
24
+ 3,
25
+ 3,
26
+ 3,
27
+ 2,
28
+ 2
29
+ ],
30
+ "conv_stride": [
31
+ 5,
32
+ 2,
33
+ 2,
34
+ 2,
35
+ 2,
36
+ 2,
37
+ 2
38
+ ],
39
+ "ctc_loss_reduction": "sum",
40
+ "ctc_zero_infinity": false,
41
+ "do_stable_layer_norm": false,
42
+ "eos_token_id": 2,
43
+ "feat_extract_activation": "gelu",
44
+ "feat_extract_norm": "group",
45
+ "feat_proj_dropout": 0.0,
46
+ "feat_proj_layer_norm": true,
47
+ "final_dropout": 0.1,
48
+ "hidden_act": "gelu",
49
+ "hidden_dropout": 0.1,
50
+ "hidden_size": 768,
51
+ "initializer_range": 0.02,
52
+ "intermediate_size": 3072,
53
+ "layer_norm_eps": 1e-05,
54
+ "layerdrop": 0.1,
55
+ "mask_feature_length": 10,
56
+ "mask_feature_min_masks": 0,
57
+ "mask_feature_prob": 0.0,
58
+ "mask_time_length": 10,
59
+ "mask_time_min_masks": 2,
60
+ "mask_time_prob": 0.05,
61
+ "model_type": "hubert",
62
+ "num_attention_heads": 12,
63
+ "num_conv_pos_embedding_groups": 16,
64
+ "num_conv_pos_embeddings": 128,
65
+ "num_feat_extract_layers": 7,
66
+ "num_hidden_layers": 12,
67
+ "pad_token_id": 0,
68
+ "torch_dtype": "float16",
69
+ "transformers_version": "4.30.2",
70
+ "use_weighted_layer_sum": false,
71
+ "vocab_size": 32
72
+ }
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-hubert-base/preprocessor_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "do_normalize": true,
3
+ "feature_extractor_type": "Wav2Vec2FeatureExtractor",
4
+ "feature_size": 1,
5
+ "padding_side": "right",
6
+ "padding_value": 0,
7
+ "return_attention_mask": false,
8
+ "sampling_rate": 16000
9
+ }
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-hubert-base/pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:24164f129c66499d1346e2aa55f183250c223161ec2770c0da3d3b08cf432d3c
3
+ size 188811417
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large/config.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/data/docker/liujing04/bert-vits2/Bert-VITS2-master20231106/bert/chinese-roberta-wwm-ext-large",
3
+ "architectures": [
4
+ "BertForMaskedLM"
5
+ ],
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "bos_token_id": 0,
8
+ "classifier_dropout": null,
9
+ "directionality": "bidi",
10
+ "eos_token_id": 2,
11
+ "hidden_act": "gelu",
12
+ "hidden_dropout_prob": 0.1,
13
+ "hidden_size": 1024,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 4096,
16
+ "layer_norm_eps": 1e-12,
17
+ "max_position_embeddings": 512,
18
+ "model_type": "bert",
19
+ "num_attention_heads": 16,
20
+ "num_hidden_layers": 24,
21
+ "output_past": true,
22
+ "pad_token_id": 0,
23
+ "pooler_fc_size": 768,
24
+ "pooler_num_attention_heads": 12,
25
+ "pooler_num_fc_layers": 3,
26
+ "pooler_size_per_head": 128,
27
+ "pooler_type": "first_token_transform",
28
+ "position_embedding_type": "absolute",
29
+ "torch_dtype": "float16",
30
+ "transformers_version": "4.30.2",
31
+ "type_vocab_size": 2,
32
+ "use_cache": true,
33
+ "vocab_size": 21128
34
+ }
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large/pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e53a693acc59ace251d143d068096ae0d7b79e4b1b503fa84c9dcf576448c1d8
3
+ size 651225145
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large/tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/fast_langdetect/lid.176.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7e69ec5451bc261cc7844e49e4792a85d7f09c06789ec800fc4a44aec362764e
3
+ size 131266198
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/s1v3.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87133414860ea14ff6620c483a3db5ed07b44be42e2c3fcdad65523a729a745a
3
+ size 155284856
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/sv/pretrained_eres2netv2w24s4ep4.ckpt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4f5a0bf73c61eb41b174e1bb54e7ee3c83233892be8e0af1f187024e8e581a35
3
+ size 107528697
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/v2Pro/s2Dv2ProPlus.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:635cd84bf6f7f9b8d41c88c7106f81d782c794c61f931845214ea037b0c5bef2
3
+ size 126432639
GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/v2Pro/s2Gv2ProPlus.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d42a22bbbf65fb2bbdd45ad6a66841156977db45c7aabe0a6992ff378d9c7d3b
3
+ size 200125741
GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/MONOPHONIC_CHARS.txt ADDED
The diff for this file is too large to render. See raw diff
 
GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/POLYPHONIC_CHARS.txt ADDED
The diff for this file is too large to render. See raw diff
 
GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/bopomofo_to_pinyin_wo_tune_dict.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"ㄌㄧㄥ": "ling", "ㄩㄢ": "yuan", "ㄒㄧㄥ": "xing", "ㄑㄧㄡ": "qiu", "ㄊㄧㄢ": "tian", "ㄎㄨㄚ": "kua", "ㄨ": "wu", "ㄧㄣ": "yin", "ㄧ": "yi", "ㄒㄧㄝ": "xie", "ㄔㄡ": "chou", "ㄋㄨㄛ": "nuo", "ㄉㄢ": "dan", "ㄒㄩ": "xu", "ㄒㄩㄥ": "xiong", "ㄌㄧㄡ": "liu", "ㄌㄧㄣ": "lin", "ㄒㄧㄤ": "xiang", "ㄩㄥ": "yong", "ㄒㄧㄣ": "xin", "ㄓㄣ": "zhen", "ㄉㄞ": "dai", "ㄆㄢ": "pan", "ㄖㄨ": "ru", "ㄇㄚ": "ma", "ㄑㄧㄢ": "qian", "ㄘ": "ci", "ㄓㄨㄥ": "zhong", "ㄋㄟ": "nei", "ㄔㄥ": "cheng", "ㄈㄥ": "feng", "ㄓㄨㄛ": "zhuo", "ㄈㄤ": "fang", "ㄠ": "ao", "ㄗㄨㄛ": "zuo", "ㄓㄡ": "zhou", "ㄉㄨㄥ": "dong", "ㄙㄨ": "su", "ㄑㄩㄥ": "qiong", "ㄎㄨㄤ": "kuang", "ㄨㄤ": "wang", "ㄌㄟ": "lei", "ㄋㄠ": "nao", "ㄓㄨ": "zhu", "ㄕㄨ": "shu", "ㄕㄣ": "shen", "ㄐㄧㄝ": "jie", "ㄉㄧㄝ": "die", "ㄔ": "chi", "ㄌㄨㄥ": "long", "ㄧㄥ": "ying", "ㄅㄥ": "beng", "ㄌㄢ": "lan", "ㄇㄧㄠ": "miao", "ㄌㄧ": "li", "ㄐㄧ": "ji", "ㄩ": "yu", "ㄌㄨㄛ": "luo", "ㄔㄞ": "chai", "ㄏㄨㄣ": "hun", "ㄏㄨㄟ": "hui", "ㄖㄠ": "rao", "ㄏㄢ": "han", "ㄒㄧ": "xi", "ㄊㄞ": "tai", "ㄧㄠ": "yao", "ㄐㄩㄣ": "jun", "ㄌㄩㄝ": "lve", "ㄊㄤ": "tang", "ㄓㄠ": "zhao", "ㄓㄞ": "zhai", "ㄓㄚ": "zha", "ㄦ": "er", "ㄖㄢ": "ran", "ㄑㄧ": "qi", "ㄙㄜ": "se", "ㄙ": "si", "ㄙㄚ": "sa", "ㄎㄨㄟ": "kui", "ㄆㄨ": "pu", "ㄊㄚ": "ta", "ㄉㄨ": "du", "ㄊㄨ": "tu", "ㄧㄤ": "yang", "ㄡ": "ou", "ㄇㄧㄢ": "mian", "ㄨㄣ": "wen", "ㄉㄧㄠ": "diao", "ㄇㄧㄝ": "mie", "ㄨㄚ": "wa", "ㄋㄧㄠ": "niao", "ㄧㄡ": "you", "ㄔㄜ": "che", "ㄑㄩㄢ": "quan", "ㄘㄞ": "cai", "ㄌㄧㄤ": "liang", "ㄍㄨ": "gu", "ㄇㄠ": "mao", "ㄍㄨㄚ": "gua", "ㄙㄨㄟ": "sui", "ㄇㄢ": "man", "ㄕ": "shi", "ㄎㄡ": "kou", "ㄊㄧㄥ": "ting", "ㄅㄧㄥ": "bing", "ㄏㄨㄛ": "huo", "ㄍㄨㄥ": "gong", "ㄑㄧㄣ": "qin", "ㄐㄩㄥ": "jiong", "ㄌㄨ": "lu", "ㄋㄢ": "nan", "ㄅㄧ": "bi", "ㄑㄧㄚ": "qia", "ㄆㄧ": "pi", "ㄉㄧㄢ": "dian", "ㄈㄨ": "fu", "ㄍㄜ": "ge", "ㄅㄞ": "bai", "ㄍㄢ": "gan", "ㄒㄩㄢ": "xuan", "ㄌㄤ": "lang", "ㄕㄜ": "she", "ㄏㄨㄚ": "hua", "ㄊㄡ": "tou", "ㄆㄧㄢ": "pian", "ㄉㄧ": "di", "ㄖㄨㄢ": "ruan", "ㄜ": "e", "ㄑㄧㄝ": "qie", "ㄉㄡ": "dou", "ㄖㄨㄟ": "rui", "ㄘㄨㄟ": "cui", "ㄐㄧㄢ": "jian", "ㄔㄨㄥ": "chong", "ㄉㄥ": "deng", "ㄐㄩㄝ": "jue", "ㄒㄩㄝ": "xue", "ㄒㄧㄠ": "xiao", "ㄗㄢ": "zan", "ㄓㄢ": "zhan", "ㄗㄡ": "zou", "ㄘㄡ": "cou", "ㄔㄨㄚ": "chua", "ㄈㄟ": "fei", "ㄅㄟ": "bei", "ㄔㄨ": "chu", "ㄅㄚ": "ba", "ㄎㄨㄞ": "kuai", "ㄒㄧㄚ": "xia", "ㄏㄜ": "he", "ㄅㄧㄝ": "bie", "ㄌㄩ": "lv", "ㄙㄨㄢ": "suan", "ㄏㄥ": "heng", "ㄍㄨㄟ": "gui", "ㄌㄡ": "lou", "ㄊㄧ": "ti", "ㄌㄜ": "le", "ㄙㄨㄣ": "sun", "ㄒㄧㄢ": "xian", "ㄑㄩㄝ": "que", "ㄓ": "zhi", "ㄐㄧㄚ": "jia", "ㄏㄨ": "hu", "ㄌㄚ": "la", "ㄎㄜ": "ke", "ㄞ": "ai", "ㄨㄟ": "wei", "ㄏㄨㄢ": "huan", "ㄕㄨㄚ": "shua", "ㄕㄨㄤ": "shuang", "ㄍㄞ": "gai", "ㄏㄞ": "hai", "ㄧㄢ": "yan", "ㄈㄢ": "fan", "ㄆㄤ": "pang", "ㄙㄨㄥ": "song", "ㄋㄜ": "ne", "ㄔㄣ": "chen", "ㄍㄨㄛ": "guo", "ㄣ": "en", "ㄋㄍ": "ng", "ㄆㄚ": "pa", "ㄈㄚ": "fa", "ㄆㄡ": "pou", "ㄏㄡ": "hou", "ㄑㄩ": "qu", "ㄒㄩㄣ": "xun", "ㄋㄧㄝ": "nie", "ㄏㄨㄥ": "hong", "ㄊㄨㄣ": "tun", "ㄨㄞ": "wai", "ㄕㄡ": "shou", "ㄧㄝ": "ye", "ㄐㄩ": "ju", "ㄙㄡ": "sou", "ㄌㄨㄣ": "lun", "ㄋㄧㄚ": "nia", "ㄆㄣ": "pen", "ㄈㄣ": "fen", "ㄔㄨㄣ": "chun", "ㄋㄧㄡ": "niu", "ㄖㄡ": "rou", "ㄉㄨㄛ": "duo", "ㄗㄜ": "ze", "ㄕㄥ": "sheng", "ㄎㄨ": "ku", "ㄧㄚ": "ya", "ㄓㄨㄟ": "zhui", "ㄍㄡ": "gou", "ㄅㄛ": "bo", "ㄋㄚ": "na", "ㄒㄧㄡ": "xiu", "ㄘㄨ": "cu", "ㄎㄨㄛ": "kuo", "ㄌㄠ": "lao", "ㄘㄨㄥ": "cong", "ㄉㄚ": "da", "ㄆㄛ": "po", "ㄙㄞ": "sai", "ㄌㄥ": "leng", "ㄖㄨㄥ": "rong", "ㄋㄧ": "ni", "ㄆㄠ": "pao", "ㄎㄢ": "kan", "ㄨㄥ": "weng", "ㄨㄢ": "wan", "ㄏㄠ": "hao", "ㄐㄧㄥ": "jing", "ㄊㄢ": "tan", "ㄅㄨ": "bu", "ㄗㄤ": "zang", "ㄐㄧㄡ": "jiu", "ㄇㄟ": "mei", "ㄇㄨ": "mu", "ㄉㄨㄟ": "dui", "ㄅㄤ": "bang", "ㄅㄠ": "bao", "ㄔㄤ": "chang", "ㄓㄤ": "zhang", "ㄗㄨㄥ": "zong", "ㄍㄨㄣ": "gun", "ㄌㄧㄠ": "liao", "ㄔㄢ": "chan", "ㄓㄜ": "zhe", "ㄇㄥ": "meng", "ㄑㄧㄠ": "qiao", "ㄋㄤ": "nang", "ㄩㄣ": "yun", "ㄎㄞ": "kai", "ㄍㄠ": "gao", "ㄊㄠ": "tao", "ㄕㄢ": "shan", "ㄌㄞ": "lai", "ㄅㄢ": "ban", "ㄎㄨㄥ": "kong", "ㄔㄨㄛ": "chuo", "ㄋㄨ": "nu", "ㄆㄟ": "pei", "ㄆㄥ": "peng", "ㄘㄢ": "can", "ㄙㄨㄛ": "suo", "ㄊㄨㄥ": "tong", "ㄑㄧㄤ": "qiang", "ㄙㄠ": "sao", "ㄓㄨㄢ": "zhuan", "ㄢ": "an", "ㄔㄚ": "cha", "ㄕㄚ": "sha", "ㄌㄧㄢ": "lian", "ㄇㄧ": "mi", "ㄋㄡ": "nou", "ㄘㄠ": "cao", "ㄙㄣ": "sen", "ㄋㄣ": "nen", "ㄋㄧㄢ": "nian", "ㄇㄞ": "mai", "ㄩㄝ": "yue", "ㄋㄞ": "nai", "ㄏㄨㄞ": "huai", "ㄗ": "zi", "ㄌㄨㄢ": "luan", "ㄉ��ㄥ": "ding", "ㄇㄤ": "mang", "ㄋㄧㄥ": "ning", "ㄇㄧㄥ": "ming", "ㄗㄨㄟ": "zui", "ㄎㄤ": "kang", "ㄉㄜ": "de", "ㄅㄧㄢ": "bian", "ㄐㄧㄣ": "jin", "ㄔㄨㄟ": "chui", "ㄊㄨㄟ": "tui", "ㄗㄚ": "za", "ㄘㄣ": "cen", "ㄇㄧㄣ": "min", "ㄏㄨㄤ": "huang", "ㄗㄨ": "zu", "ㄘㄨㄛ": "cuo", "ㄊㄨㄛ": "tuo", "ㄑㄩㄣ": "qun", "ㄅㄧㄣ": "bin", "ㄊㄧㄠ": "tiao", "ㄍㄤ": "gang", "ㄉㄨㄢ": "duan", "ㄅㄧㄠ": "biao", "ㄉㄠ": "dao", "ㄖㄨㄣ": "run", "ㄐㄧㄠ": "jiao", "ㄨㄛ": "wo", "ㄘㄨㄢ": "cuan", "ㄖㄣ": "ren", "ㄇㄣ": "men", "ㄓㄨㄣ": "zhun", "ㄎㄨㄣ": "kun", "ㄔㄨㄤ": "chuang", "ㄗㄠ": "zao", "ㄓㄥ": "zheng", "ㄆㄧㄣ": "pin", "ㄅㄣ": "ben", "ㄐㄧㄤ": "jiang", "ㄐㄩㄢ": "juan", "ㄘㄥ": "ceng", "ㄏㄤ": "hang", "ㄋㄧㄣ": "nin", "ㄌㄧㄝ": "lie", "ㄍㄨㄤ": "guang", "ㄙㄢ": "san", "ㄊㄜ": "te", "ㄕㄨㄣ": "shun", "ㄕㄨㄟ": "shui", "ㄔㄠ": "chao", "ㄘㄜ": "ce", "ㄍㄨㄞ": "guai", "ㄎㄥ": "keng", "ㄕㄞ": "shai", "ㄉㄣ": "den", "ㄊㄨㄢ": "tuan", "ㄆㄧㄠ": "piao", "ㄑㄧㄥ": "qing", "ㄍㄥ": "geng", "ㄔㄨㄞ": "chuai", "ㄕㄠ": "shao", "ㄍㄣ": "gen", "ㄋㄨㄢ": "nuan", "ㄖㄥ": "reng", "ㄇㄡ": "mou", "ㄆㄞ": "pai", "ㄤ": "ang", "ㄎㄚ": "ka", "ㄍㄨㄢ": "guan", "ㄕㄨㄛ": "shuo", "ㄏㄣ": "hen", "ㄔㄨㄢ": "chuan", "ㄎㄨㄢ": "kuan", "ㄏㄟ": "hei", "ㄇㄛ": "mo", "ㄗㄞ": "zai", "ㄋㄥ": "neng", "ㄕㄨㄞ": "shuai", "ㄖㄜ": "re", "ㄋㄩ": "nv", "ㄆㄧㄥ": "ping", "ㄘㄤ": "cang", "ㄋㄨㄥ": "nong", "ㄎㄠ": "kao", "ㄗㄨㄢ": "zuan", "ㄎㄣ": "ken", "ㄍㄚ": "ga", "ㄗㄣ": "zen", "ㄉㄤ": "dang", "ㄗㄥ": "zeng", "ㄉㄨㄣ": "dun", "ㄘㄚ": "ca", "ㄖㄤ": "rang", "ㄘㄨㄣ": "cun", "ㄖㄨㄛ": "ruo", "ㄊㄧㄝ": "tie", "ㄊㄥ": "teng", "ㄙㄥ": "seng", "ㄖ": "ri", "ㄗㄨㄣ": "zun", "ㄋㄧㄤ": "niang", "ㄋㄩㄝ": "nve", "ㄙㄤ": "sang", "ㄓㄨㄤ": "zhuang", "ㄕㄤ": "shang", "ㄆㄧㄝ": "pie", "ㄕㄨㄢ": "shuan", "ㄈㄡ": "fou", "ㄉㄧㄡ": "diu", "ㄇㄜ": "me", "ㄈㄛ": "fo", "ㄌㄧㄚ": "lia", "ㄎㄟ": "kei", "ㄏㄚ": "ha", "ㄚ": "a", "ㄌㄛ": "lo", "ㄧㄛ": "yo", "ㄛ": "o", "ㄏㄋㄍ": "hng", "ㄋ": "n", "ㄌㄣ": "len", "ㄉㄧㄚ": "dia", "ㄇㄧㄡ": "miu", "ㄉㄟ": "dei", "ㄏㄇ": "hm", "ㄋㄨㄣ": "nun", "ㄓㄨㄞ": "zhuai", "ㄊㄟ": "tei", "ㄗㄟ": "zei", "ㄓㄨㄚ": "zhua", "ㄖㄨㄚ": "rua", "ê": "ê", "ㄟ": "ei", "ㄍㄟ": "gei", "ㄈㄧㄠ": "fiao", "ㄕㄟ": "shei", "ㄓㄟ": "zhei", "ㄥ": "eng", "ㄘㄟ": "cei", "ㄉㄧㄣ": "din", "ㄅㄧㄤ": "biang", "ㄧㄞ": "yai"}
GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/char_bopomofo_dict.json ADDED
The diff for this file is too large to render. See raw diff
 
GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/config.py ADDED
@@ -0,0 +1,19 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ manual_seed = 1313
2
+ model_source = 'bert-base-chinese'
3
+ window_size = 32
4
+ num_workers = 2
5
+ use_mask = True
6
+ use_conditional = True
7
+ param_conditional = {
8
+ 'bias': True,
9
+ 'char-linear': True,
10
+ 'pos-linear': False,
11
+ 'char+pos-second': True,
12
+ }
13
+
14
+ batch_size = 256
15
+ use_pos = True
16
+ param_pos = {
17
+ 'weight': 0.1,
18
+ 'pos_joint_training': True,
19
+ }
GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/g2pW.onnx ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2eb3c71fd95117b2e1abef8d2d0cd78aae894bbe7f0fac105ddc9c32ce63cbd0
3
+ size 635212732
GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/record.log ADDED
@@ -0,0 +1,1005 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ device: cuda
2
+ now: 2022-04-01 22:13:18.349604
3
+ [200] train_loss=0.289519 valid_loss=0.102661 valid_pos_acc=0.924619 valid_acc=0.97596 / 0.703958 / 0.586078 best_acc=0.97596
4
+ now: 2022-04-01 22:25:27.330080
5
+ [400] train_loss=0.089245 valid_loss=0.0703849 valid_pos_acc=0.942315 valid_acc=0.984227 / 0.747566 / 0.616754 best_acc=0.984227
6
+ now: 2022-04-01 22:37:16.857336
7
+ [600] train_loss=0.0663516 valid_loss=0.0597114 valid_pos_acc=0.946489 valid_acc=0.98734 / 0.77479 / 0.638442 best_acc=0.98734
8
+ now: 2022-04-01 22:49:06.182095
9
+ [800] train_loss=0.0559394 valid_loss=0.0535268 valid_pos_acc=0.948245 valid_acc=0.988928 / 0.774415 / 0.643435 best_acc=0.988928
10
+ now: 2022-04-01 23:00:55.371920
11
+ [1000] train_loss=0.0497098 valid_loss=0.0490104 valid_pos_acc=0.954161 valid_acc=0.989486 / 0.796356 / 0.664386 best_acc=0.989486
12
+ now: 2022-04-01 23:12:49.781716
13
+ [1200] train_loss=0.0462926 valid_loss=0.0466889 valid_pos_acc=0.954634 valid_acc=0.989913 / 0.802885 / 0.673908 best_acc=0.989913
14
+ now: 2022-04-01 23:24:43.685062
15
+ [1400] train_loss=0.0433836 valid_loss=0.0451725 valid_pos_acc=0.956761 valid_acc=0.99049 / 0.805024 / 0.674369 best_acc=0.99049
16
+ now: 2022-04-01 23:36:46.100963
17
+ [1600] train_loss=0.0404561 valid_loss=0.0436914 valid_pos_acc=0.957201 valid_acc=0.991022 / 0.811412 / 0.679481 best_acc=0.991022
18
+ now: 2022-04-01 23:48:48.583240
19
+ [1800] train_loss=0.040905 valid_loss=0.0412648 valid_pos_acc=0.958418 valid_acc=0.991332 / 0.815194 / 0.681627 best_acc=0.991332
20
+ now: 2022-04-02 00:00:42.282365
21
+ [2000] train_loss=0.0384612 valid_loss=0.0402427 valid_pos_acc=0.959796 valid_acc=0.991534 / 0.819666 / 0.689516 best_acc=0.991534
22
+ now: 2022-04-02 00:12:52.902834
23
+ [2200] train_loss=0.0373539 valid_loss=0.0410455 valid_pos_acc=0.961692 valid_acc=0.991425 / 0.828402 / 0.696595 best_acc=0.991534
24
+ now: 2022-04-02 00:25:06.851427
25
+ [2400] train_loss=0.0367612 valid_loss=0.039694 valid_pos_acc=0.960926 valid_acc=0.991823 / 0.830391 / 0.700222 best_acc=0.991823
26
+ now: 2022-04-02 00:37:24.156808
27
+ [2600] train_loss=0.0386493 valid_loss=0.0377683 valid_pos_acc=0.962183 valid_acc=0.992202 / 0.832219 / 0.707156 best_acc=0.992202
28
+ now: 2022-04-02 00:49:37.943513
29
+ [2800] train_loss=0.0356553 valid_loss=0.0381159 valid_pos_acc=0.962729 valid_acc=0.992061 / 0.835112 / 0.707941 best_acc=0.992202
30
+ now: 2022-04-02 01:01:43.672504
31
+ [3000] train_loss=0.0338178 valid_loss=0.0386144 valid_pos_acc=0.962419 valid_acc=0.992322 / 0.835546 / 0.710556 best_acc=0.992322
32
+ now: 2022-04-02 01:13:56.991606
33
+ [3200] train_loss=0.0335683 valid_loss=0.0381786 valid_pos_acc=0.962755 valid_acc=0.992233 / 0.838008 / 0.713975 best_acc=0.992322
34
+ now: 2022-04-02 01:26:13.830261
35
+ [3400] train_loss=0.0316981 valid_loss=0.0373759 valid_pos_acc=0.963524 valid_acc=0.992309 / 0.843253 / 0.718974 best_acc=0.992322
36
+ now: 2022-04-02 01:38:08.308362
37
+ [3600] train_loss=0.0350782 valid_loss=0.0376615 valid_pos_acc=0.96404 valid_acc=0.992259 / 0.84979 / 0.725183 best_acc=0.992322
38
+ now: 2022-04-02 01:49:59.416353
39
+ [3800] train_loss=0.0321498 valid_loss=0.0367548 valid_pos_acc=0.964441 valid_acc=0.992801 / 0.850152 / 0.722988 best_acc=0.992801
40
+ now: 2022-04-02 02:02:09.238893
41
+ [4000] train_loss=0.0331685 valid_loss=0.0369892 valid_pos_acc=0.963339 valid_acc=0.992777 / 0.859395 / 0.730708 best_acc=0.992801
42
+ now: 2022-04-02 02:14:27.957159
43
+ [4200] train_loss=0.0317164 valid_loss=0.0350153 valid_pos_acc=0.965784 valid_acc=0.992656 / 0.853549 / 0.727562 best_acc=0.992801
44
+ now: 2022-04-02 02:26:43.092476
45
+ [4400] train_loss=0.0324034 valid_loss=0.0346509 valid_pos_acc=0.965843 valid_acc=0.992981 / 0.853694 / 0.73043 best_acc=0.992981
46
+ now: 2022-04-02 02:39:22.465030
47
+ [4600] train_loss=0.0298959 valid_loss=0.0356152 valid_pos_acc=0.965606 valid_acc=0.993022 / 0.855494 / 0.728954 best_acc=0.993022
48
+ now: 2022-04-02 02:51:53.210107
49
+ [4800] train_loss=0.0310447 valid_loss=0.0355586 valid_pos_acc=0.965446 valid_acc=0.992597 / 0.851145 / 0.728595 best_acc=0.993022
50
+ now: 2022-04-02 03:04:21.463931
51
+ [5000] train_loss=0.031017 valid_loss=0.034331 valid_pos_acc=0.965695 valid_acc=0.992866 / 0.852123 / 0.728928 best_acc=0.993022
52
+ now: 2022-04-02 03:16:32.777183
53
+ [5200] train_loss=0.0312034 valid_loss=0.0349778 valid_pos_acc=0.966472 valid_acc=0.993105 / 0.855114 / 0.733248 best_acc=0.993105
54
+ now: 2022-04-02 03:28:51.440974
55
+ [5400] train_loss=0.0294329 valid_loss=0.0339991 valid_pos_acc=0.966307 valid_acc=0.993109 / 0.852872 / 0.727198 best_acc=0.993109
56
+ now: 2022-04-02 03:41:07.884688
57
+ [5600] train_loss=0.0285982 valid_loss=0.0341394 valid_pos_acc=0.966307 valid_acc=0.993183 / 0.858873 / 0.736458 best_acc=0.993183
58
+ now: 2022-04-02 03:53:43.422479
59
+ [5800] train_loss=0.0283985 valid_loss=0.0325766 valid_pos_acc=0.96683 valid_acc=0.993376 / 0.856761 / 0.738166 best_acc=0.993376
60
+ now: 2022-04-02 04:06:06.964628
61
+ [6000] train_loss=0.0302441 valid_loss=0.0344224 valid_pos_acc=0.966774 valid_acc=0.992838 / 0.85689 / 0.733677 best_acc=0.993376
62
+ now: 2022-04-02 04:18:20.312766
63
+ [6200] train_loss=0.0289215 valid_loss=0.0348225 valid_pos_acc=0.966589 valid_acc=0.993367 / 0.858202 / 0.736723 best_acc=0.993376
64
+ now: 2022-04-02 04:30:36.722397
65
+ [6400] train_loss=0.0294263 valid_loss=0.0329629 valid_pos_acc=0.966854 valid_acc=0.993081 / 0.856632 / 0.7381 best_acc=0.993376
66
+ now: 2022-04-02 04:42:53.493232
67
+ [6600] train_loss=0.0285769 valid_loss=0.0333396 valid_pos_acc=0.967153 valid_acc=0.993547 / 0.865742 / 0.743425 best_acc=0.993547
68
+ now: 2022-04-02 04:55:17.818463
69
+ [6800] train_loss=0.0265485 valid_loss=0.0330653 valid_pos_acc=0.967776 valid_acc=0.993222 / 0.865918 / 0.743298 best_acc=0.993547
70
+ now: 2022-04-02 05:07:36.630349
71
+ [7000] train_loss=0.0284473 valid_loss=0.0320964 valid_pos_acc=0.968023 valid_acc=0.99355 / 0.868261 / 0.748849 best_acc=0.99355
72
+ now: 2022-04-02 05:20:01.434422
73
+ [7200] train_loss=0.0274993 valid_loss=0.0326511 valid_pos_acc=0.9669 valid_acc=0.993816 / 0.868294 / 0.746817 best_acc=0.993816
74
+ now: 2022-04-02 05:32:29.662142
75
+ [7400] train_loss=0.02851 valid_loss=0.0308467 valid_pos_acc=0.968453 valid_acc=0.993858 / 0.863909 / 0.746068 best_acc=0.993858
76
+ now: 2022-04-02 05:44:43.967440
77
+ [7600] train_loss=0.0282732 valid_loss=0.03368 valid_pos_acc=0.967292 valid_acc=0.993014 / 0.86581 / 0.745753 best_acc=0.993858
78
+ now: 2022-04-02 05:56:45.436298
79
+ [7800] train_loss=0.0252737 valid_loss=0.0315786 valid_pos_acc=0.967611 valid_acc=0.993799 / 0.869773 / 0.749114 best_acc=0.993858
80
+ now: 2022-04-02 06:08:51.140922
81
+ [8000] train_loss=0.0280509 valid_loss=0.0328118 valid_pos_acc=0.96732 valid_acc=0.99363 / 0.86537 / 0.74611 best_acc=0.993858
82
+ now: 2022-04-02 06:20:43.247091
83
+ [8200] train_loss=0.028321 valid_loss=0.0308812 valid_pos_acc=0.968106 valid_acc=0.993758 / 0.869684 / 0.751653 best_acc=0.993858
84
+ now: 2022-04-02 06:32:38.603877
85
+ [8400] train_loss=0.0271253 valid_loss=0.0326289 valid_pos_acc=0.968232 valid_acc=0.993426 / 0.869263 / 0.748637 best_acc=0.993858
86
+ now: 2022-04-02 06:44:45.010090
87
+ [8600] train_loss=0.02778 valid_loss=0.0308819 valid_pos_acc=0.968731 valid_acc=0.993693 / 0.87573 / 0.75794 best_acc=0.993858
88
+ now: 2022-04-02 06:56:45.886905
89
+ [8800] train_loss=0.0287492 valid_loss=0.0310371 valid_pos_acc=0.968256 valid_acc=0.993563 / 0.877011 / 0.759391 best_acc=0.993858
90
+ now: 2022-04-02 07:08:52.584840
91
+ [9000] train_loss=0.0281025 valid_loss=0.0297675 valid_pos_acc=0.968566 valid_acc=0.993979 / 0.866884 / 0.750877 best_acc=0.993979
92
+ now: 2022-04-02 07:21:04.827592
93
+ [9200] train_loss=0.026893 valid_loss=0.0310813 valid_pos_acc=0.968965 valid_acc=0.993758 / 0.869433 / 0.752492 best_acc=0.993979
94
+ now: 2022-04-02 07:33:11.165254
95
+ [9400] train_loss=0.0253738 valid_loss=0.0307835 valid_pos_acc=0.969295 valid_acc=0.994046 / 0.878856 / 0.754636 best_acc=0.994046
96
+ now: 2022-04-02 07:45:16.521889
97
+ [9600] train_loss=0.0263703 valid_loss=0.0308493 valid_pos_acc=0.969039 valid_acc=0.993986 / 0.873759 / 0.753114 best_acc=0.994046
98
+ now: 2022-04-02 07:57:19.055032
99
+ [9800] train_loss=0.0258709 valid_loss=0.0304116 valid_pos_acc=0.967514 valid_acc=0.993751 / 0.87442 / 0.760402 best_acc=0.994046
100
+ now: 2022-04-02 08:09:21.455984
101
+ [10000] train_loss=0.0261966 valid_loss=0.0310479 valid_pos_acc=0.968954 valid_acc=0.993786 / 0.879653 / 0.76128 best_acc=0.994046
102
+ now: 2022-04-02 08:21:30.441155
103
+ [10200] train_loss=0.0272568 valid_loss=0.0306756 valid_pos_acc=0.969087 valid_acc=0.993777 / 0.879809 / 0.760943 best_acc=0.994046
104
+ now: 2022-04-02 08:33:36.839764
105
+ [10400] train_loss=0.027559 valid_loss=0.0308756 valid_pos_acc=0.969366 valid_acc=0.993636 / 0.874443 / 0.755992 best_acc=0.994046
106
+ now: 2022-04-02 08:45:39.747008
107
+ [10600] train_loss=0.027269 valid_loss=0.0329513 valid_pos_acc=0.96888 valid_acc=0.992499 / 0.833803 / 0.727518 best_acc=0.994046
108
+ now: 2022-04-02 08:57:40.273311
109
+ [10800] train_loss=0.0255775 valid_loss=0.0318773 valid_pos_acc=0.969314 valid_acc=0.993576 / 0.865286 / 0.745435 best_acc=0.994046
110
+ now: 2022-04-02 09:09:28.232166
111
+ [11000] train_loss=0.027821 valid_loss=0.0324836 valid_pos_acc=0.969581 valid_acc=0.993517 / 0.858519 / 0.741976 best_acc=0.994046
112
+ now: 2022-04-02 09:21:18.956995
113
+ [11200] train_loss=0.0268467 valid_loss=0.0320919 valid_pos_acc=0.968768 valid_acc=0.993515 / 0.859293 / 0.743348 best_acc=0.994046
114
+ now: 2022-04-02 09:33:14.899728
115
+ [11400] train_loss=0.0277983 valid_loss=0.0304641 valid_pos_acc=0.969013 valid_acc=0.993803 / 0.857242 / 0.744822 best_acc=0.994046
116
+ now: 2022-04-02 09:45:20.431378
117
+ [11600] train_loss=0.0278141 valid_loss=0.0303669 valid_pos_acc=0.969312 valid_acc=0.993491 / 0.861739 / 0.751563 best_acc=0.994046
118
+ now: 2022-04-02 09:57:29.453034
119
+ [11800] train_loss=0.0272102 valid_loss=0.030325 valid_pos_acc=0.969045 valid_acc=0.993871 / 0.865131 / 0.753686 best_acc=0.994046
120
+ now: 2022-04-02 10:09:41.097392
121
+ [12000] train_loss=0.0271826 valid_loss=0.0302743 valid_pos_acc=0.969701 valid_acc=0.993645 / 0.865918 / 0.753973 best_acc=0.994046
122
+ now: 2022-04-02 10:21:38.263361
123
+ [12200] train_loss=0.0266099 valid_loss=0.0288773 valid_pos_acc=0.969902 valid_acc=0.994183 / 0.874035 / 0.76018 best_acc=0.994183
124
+ now: 2022-04-02 10:33:44.432773
125
+ [12400] train_loss=0.0252403 valid_loss=0.029718 valid_pos_acc=0.969434 valid_acc=0.994127 / 0.87762 / 0.756971 best_acc=0.994183
126
+ now: 2022-04-02 10:45:51.265489
127
+ [12600] train_loss=0.0247018 valid_loss=0.0312226 valid_pos_acc=0.969588 valid_acc=0.993641 / 0.881627 / 0.759701 best_acc=0.994183
128
+ now: 2022-04-02 10:57:42.967866
129
+ [12800] train_loss=0.0269899 valid_loss=0.0291686 valid_pos_acc=0.969384 valid_acc=0.994157 / 0.883137 / 0.766173 best_acc=0.994183
130
+ now: 2022-04-02 11:09:44.697867
131
+ [13000] train_loss=0.026225 valid_loss=0.0300868 valid_pos_acc=0.969607 valid_acc=0.993756 / 0.881478 / 0.764432 best_acc=0.994183
132
+ now: 2022-04-02 11:21:47.303966
133
+ [13200] train_loss=0.0251707 valid_loss=0.0292528 valid_pos_acc=0.969937 valid_acc=0.994205 / 0.882159 / 0.764326 best_acc=0.994205
134
+ now: 2022-04-02 11:34:07.084506
135
+ [13400] train_loss=0.0256715 valid_loss=0.0294879 valid_pos_acc=0.969972 valid_acc=0.994331 / 0.879444 / 0.763116 best_acc=0.994331
136
+ now: 2022-04-02 11:46:17.499315
137
+ [13600] train_loss=0.0266713 valid_loss=0.0307474 valid_pos_acc=0.968488 valid_acc=0.994092 / 0.88213 / 0.764678 best_acc=0.994331
138
+ now: 2022-04-02 11:58:29.053919
139
+ [13800] train_loss=0.0263307 valid_loss=0.0299171 valid_pos_acc=0.969182 valid_acc=0.994146 / 0.886646 / 0.766393 best_acc=0.994331
140
+ now: 2022-04-02 12:10:42.628035
141
+ [14000] train_loss=0.0254249 valid_loss=0.0291546 valid_pos_acc=0.970087 valid_acc=0.994083 / 0.884958 / 0.766853 best_acc=0.994331
142
+ now: 2022-04-02 12:22:47.459592
143
+ [14200] train_loss=0.0271003 valid_loss=0.0289376 valid_pos_acc=0.969803 valid_acc=0.994263 / 0.882713 / 0.765841 best_acc=0.994331
144
+ now: 2022-04-02 12:34:51.459159
145
+ [14400] train_loss=0.0253207 valid_loss=0.0284943 valid_pos_acc=0.97041 valid_acc=0.994483 / 0.883065 / 0.768558 best_acc=0.994483
146
+ now: 2022-04-02 12:47:03.082143
147
+ [14600] train_loss=0.0256933 valid_loss=0.0275894 valid_pos_acc=0.970781 valid_acc=0.994426 / 0.882073 / 0.768093 best_acc=0.994483
148
+ now: 2022-04-02 12:59:01.736374
149
+ [14800] train_loss=0.025288 valid_loss=0.0290729 valid_pos_acc=0.969898 valid_acc=0.994029 / 0.884465 / 0.764771 best_acc=0.994483
150
+ now: 2022-04-02 13:11:07.383148
151
+ [15000] train_loss=0.0254068 valid_loss=0.0292592 valid_pos_acc=0.970833 valid_acc=0.994096 / 0.888942 / 0.769842 best_acc=0.994483
152
+ now: 2022-04-02 13:23:22.842378
153
+ [15200] train_loss=0.0236412 valid_loss=0.0282784 valid_pos_acc=0.970644 valid_acc=0.994229 / 0.889311 / 0.768718 best_acc=0.994483
154
+ now: 2022-04-02 13:35:29.034906
155
+ [15400] train_loss=0.0243784 valid_loss=0.0292398 valid_pos_acc=0.970883 valid_acc=0.994187 / 0.889313 / 0.771062 best_acc=0.994483
156
+ now: 2022-04-02 13:47:31.205294
157
+ [15600] train_loss=0.0240879 valid_loss=0.0298062 valid_pos_acc=0.97036 valid_acc=0.994057 / 0.885163 / 0.767167 best_acc=0.994483
158
+ now: 2022-04-02 13:59:37.091807
159
+ [15800] train_loss=0.0241428 valid_loss=0.0298697 valid_pos_acc=0.97079 valid_acc=0.994118 / 0.886573 / 0.768139 best_acc=0.994483
160
+ now: 2022-04-02 14:11:45.813137
161
+ [16000] train_loss=0.0245795 valid_loss=0.028206 valid_pos_acc=0.970714 valid_acc=0.994365 / 0.895285 / 0.778247 best_acc=0.994483
162
+ now: 2022-04-02 14:23:31.259816
163
+ [16200] train_loss=0.0259529 valid_loss=0.0295037 valid_pos_acc=0.971532 valid_acc=0.994166 / 0.892761 / 0.773792 best_acc=0.994483
164
+ now: 2022-04-02 14:35:29.710419
165
+ [16400] train_loss=0.0245774 valid_loss=0.0282059 valid_pos_acc=0.970677 valid_acc=0.994159 / 0.892281 / 0.77366 best_acc=0.994483
166
+ now: 2022-04-02 14:47:28.883504
167
+ [16600] train_loss=0.0249353 valid_loss=0.0287864 valid_pos_acc=0.970805 valid_acc=0.994255 / 0.891979 / 0.776495 best_acc=0.994483
168
+ now: 2022-04-02 14:59:36.374751
169
+ [16800] train_loss=0.0266362 valid_loss=0.0283276 valid_pos_acc=0.970768 valid_acc=0.994281 / 0.898043 / 0.780269 best_acc=0.994483
170
+ now: 2022-04-02 15:11:40.341586
171
+ [17000] train_loss=0.0248526 valid_loss=0.0279962 valid_pos_acc=0.970482 valid_acc=0.994411 / 0.897109 / 0.780885 best_acc=0.994483
172
+ now: 2022-04-02 15:23:39.987145
173
+ [17200] train_loss=0.0237728 valid_loss=0.028023 valid_pos_acc=0.971417 valid_acc=0.994322 / 0.888697 / 0.776213 best_acc=0.994483
174
+ now: 2022-04-02 15:35:38.801398
175
+ [17400] train_loss=0.0249057 valid_loss=0.027339 valid_pos_acc=0.971389 valid_acc=0.994159 / 0.881219 / 0.768915 best_acc=0.994483
176
+ now: 2022-04-02 15:47:39.875724
177
+ [17600] train_loss=0.0246854 valid_loss=0.028386 valid_pos_acc=0.970968 valid_acc=0.994448 / 0.891969 / 0.775594 best_acc=0.994483
178
+ now: 2022-04-02 15:59:43.932068
179
+ [17800] train_loss=0.0264608 valid_loss=0.0281136 valid_pos_acc=0.971109 valid_acc=0.994465 / 0.896316 / 0.78033 best_acc=0.994483
180
+ now: 2022-04-02 16:11:42.780407
181
+ [18000] train_loss=0.0226492 valid_loss=0.0282867 valid_pos_acc=0.970959 valid_acc=0.994574 / 0.898661 / 0.782303 best_acc=0.994574
182
+ now: 2022-04-02 16:23:45.328393
183
+ [18200] train_loss=0.0253564 valid_loss=0.0272226 valid_pos_acc=0.971202 valid_acc=0.994485 / 0.894385 / 0.781905 best_acc=0.994574
184
+ now: 2022-04-02 16:35:43.743594
185
+ [18400] train_loss=0.0237427 valid_loss=0.0273525 valid_pos_acc=0.971284 valid_acc=0.994598 / 0.893183 / 0.778385 best_acc=0.994598
186
+ now: 2022-04-02 16:47:51.962569
187
+ [18600] train_loss=0.0226361 valid_loss=0.0275174 valid_pos_acc=0.971801 valid_acc=0.994608 / 0.897236 / 0.783469 best_acc=0.994608
188
+ now: 2022-04-02 17:00:05.072496
189
+ [18800] train_loss=0.0247811 valid_loss=0.0276029 valid_pos_acc=0.971766 valid_acc=0.994591 / 0.898223 / 0.780288 best_acc=0.994608
190
+ now: 2022-04-02 17:12:14.066971
191
+ [19000] train_loss=0.0249346 valid_loss=0.0269959 valid_pos_acc=0.9713 valid_acc=0.994433 / 0.889386 / 0.778011 best_acc=0.994608
192
+ now: 2022-04-02 17:24:25.443387
193
+ [19200] train_loss=0.024029 valid_loss=0.0273701 valid_pos_acc=0.971777 valid_acc=0.994565 / 0.89385 / 0.781098 best_acc=0.994608
194
+ now: 2022-04-02 17:36:21.119407
195
+ [19400] train_loss=0.0221598 valid_loss=0.028189 valid_pos_acc=0.971337 valid_acc=0.99447 / 0.892931 / 0.778729 best_acc=0.994608
196
+ now: 2022-04-02 17:48:26.051306
197
+ [19600] train_loss=0.0232854 valid_loss=0.027458 valid_pos_acc=0.97138 valid_acc=0.994535 / 0.892143 / 0.778963 best_acc=0.994608
198
+ now: 2022-04-02 18:00:41.153532
199
+ [19800] train_loss=0.0246367 valid_loss=0.0277884 valid_pos_acc=0.971415 valid_acc=0.994454 / 0.892544 / 0.777699 best_acc=0.994608
200
+ now: 2022-04-02 18:12:44.656831
201
+ [20000] train_loss=0.0193271 valid_loss=0.0288193 valid_pos_acc=0.97153 valid_acc=0.9945 / 0.89201 / 0.778198 best_acc=0.994608
202
+ now: 2022-04-02 18:24:47.237186
203
+ [20200] train_loss=0.0195292 valid_loss=0.0281468 valid_pos_acc=0.972115 valid_acc=0.99463 / 0.894395 / 0.782171 best_acc=0.99463
204
+ now: 2022-04-02 18:37:12.174319
205
+ [20400] train_loss=0.0194709 valid_loss=0.0272298 valid_pos_acc=0.971836 valid_acc=0.994686 / 0.901333 / 0.789041 best_acc=0.994686
206
+ now: 2022-04-02 18:49:37.572502
207
+ [20600] train_loss=0.0191372 valid_loss=0.0279598 valid_pos_acc=0.971139 valid_acc=0.994476 / 0.898648 / 0.786924 best_acc=0.994686
208
+ now: 2022-04-02 19:01:57.223764
209
+ [20800] train_loss=0.0191731 valid_loss=0.0283044 valid_pos_acc=0.971853 valid_acc=0.994665 / 0.906199 / 0.791619 best_acc=0.994686
210
+ now: 2022-04-02 19:14:03.413253
211
+ [21000] train_loss=0.0206041 valid_loss=0.0264156 valid_pos_acc=0.972094 valid_acc=0.994773 / 0.903901 / 0.792328 best_acc=0.994773
212
+ now: 2022-04-02 19:26:27.975422
213
+ [21200] train_loss=0.016823 valid_loss=0.0271615 valid_pos_acc=0.97184 valid_acc=0.99471 / 0.904068 / 0.790123 best_acc=0.994773
214
+ now: 2022-04-02 19:38:24.145598
215
+ [21400] train_loss=0.0205676 valid_loss=0.0285307 valid_pos_acc=0.971365 valid_acc=0.994502 / 0.899852 / 0.786174 best_acc=0.994773
216
+ now: 2022-04-02 19:50:32.570024
217
+ [21600] train_loss=0.0193456 valid_loss=0.0265744 valid_pos_acc=0.971851 valid_acc=0.994726 / 0.902354 / 0.791664 best_acc=0.994773
218
+ now: 2022-04-02 20:02:43.793933
219
+ [21800] train_loss=0.0202321 valid_loss=0.0266519 valid_pos_acc=0.971712 valid_acc=0.994788 / 0.900679 / 0.791454 best_acc=0.994788
220
+ now: 2022-04-02 20:14:55.874949
221
+ [22000] train_loss=0.0209497 valid_loss=0.0274821 valid_pos_acc=0.970824 valid_acc=0.99463 / 0.890966 / 0.780236 best_acc=0.994788
222
+ now: 2022-04-02 20:26:46.865536
223
+ [22200] train_loss=0.0212923 valid_loss=0.0270931 valid_pos_acc=0.971727 valid_acc=0.994695 / 0.893237 / 0.779557 best_acc=0.994788
224
+ now: 2022-04-02 20:38:52.796788
225
+ [22400] train_loss=0.021577 valid_loss=0.027899 valid_pos_acc=0.971914 valid_acc=0.994771 / 0.90721 / 0.790755 best_acc=0.994788
226
+ now: 2022-04-02 20:50:45.585864
227
+ [22600] train_loss=0.020184 valid_loss=0.0270084 valid_pos_acc=0.971606 valid_acc=0.994723 / 0.901765 / 0.793151 best_acc=0.994788
228
+ now: 2022-04-02 21:02:35.179860
229
+ [22800] train_loss=0.0210601 valid_loss=0.0255871 valid_pos_acc=0.972658 valid_acc=0.994784 / 0.89429 / 0.783486 best_acc=0.994788
230
+ now: 2022-04-02 21:14:32.385194
231
+ [23000] train_loss=0.0184889 valid_loss=0.0267728 valid_pos_acc=0.971968 valid_acc=0.994912 / 0.898744 / 0.787857 best_acc=0.994912
232
+ now: 2022-04-02 21:26:30.460311
233
+ [23200] train_loss=0.0196876 valid_loss=0.027992 valid_pos_acc=0.971883 valid_acc=0.99448 / 0.902701 / 0.789254 best_acc=0.994912
234
+ now: 2022-04-02 21:38:18.919183
235
+ [23400] train_loss=0.0192092 valid_loss=0.027083 valid_pos_acc=0.972339 valid_acc=0.994817 / 0.898885 / 0.789367 best_acc=0.994912
236
+ now: 2022-04-02 21:50:24.318095
237
+ [23600] train_loss=0.0205622 valid_loss=0.0268123 valid_pos_acc=0.972528 valid_acc=0.994867 / 0.898603 / 0.786304 best_acc=0.994912
238
+ now: 2022-04-02 22:02:15.239367
239
+ [23800] train_loss=0.0199384 valid_loss=0.0272734 valid_pos_acc=0.972411 valid_acc=0.99471 / 0.901925 / 0.78745 best_acc=0.994912
240
+ now: 2022-04-02 22:14:18.216068
241
+ [24000] train_loss=0.0178143 valid_loss=0.0279724 valid_pos_acc=0.971799 valid_acc=0.994817 / 0.902843 / 0.788122 best_acc=0.994912
242
+ now: 2022-04-02 22:26:08.869937
243
+ [24200] train_loss=0.0204505 valid_loss=0.0271799 valid_pos_acc=0.971603 valid_acc=0.994628 / 0.901783 / 0.790354 best_acc=0.994912
244
+ now: 2022-04-02 22:38:05.762235
245
+ [24400] train_loss=0.0196403 valid_loss=0.0279875 valid_pos_acc=0.971712 valid_acc=0.994673 / 0.90053 / 0.786413 best_acc=0.994912
246
+ now: 2022-04-02 22:49:58.498158
247
+ [24600] train_loss=0.0196382 valid_loss=0.0282003 valid_pos_acc=0.971235 valid_acc=0.994626 / 0.902809 / 0.787394 best_acc=0.994912
248
+ now: 2022-04-02 23:02:07.775949
249
+ [24800] train_loss=0.0193936 valid_loss=0.0267504 valid_pos_acc=0.972562 valid_acc=0.994665 / 0.902442 / 0.787082 best_acc=0.994912
250
+ now: 2022-04-02 23:14:14.202685
251
+ [25000] train_loss=0.0193612 valid_loss=0.0258062 valid_pos_acc=0.972654 valid_acc=0.994873 / 0.905545 / 0.794271 best_acc=0.994912
252
+ now: 2022-04-02 23:26:09.859008
253
+ [25200] train_loss=0.0190576 valid_loss=0.0271653 valid_pos_acc=0.972484 valid_acc=0.994906 / 0.909854 / 0.793673 best_acc=0.994912
254
+ now: 2022-04-02 23:38:05.448510
255
+ [25400] train_loss=0.0206686 valid_loss=0.0264603 valid_pos_acc=0.972343 valid_acc=0.994934 / 0.908615 / 0.793492 best_acc=0.994934
256
+ now: 2022-04-02 23:50:23.423237
257
+ [25600] train_loss=0.019207 valid_loss=0.0264518 valid_pos_acc=0.972285 valid_acc=0.994947 / 0.909035 / 0.793275 best_acc=0.994947
258
+ now: 2022-04-03 00:02:25.949756
259
+ [25800] train_loss=0.0204011 valid_loss=0.027376 valid_pos_acc=0.972029 valid_acc=0.994463 / 0.903614 / 0.787668 best_acc=0.994947
260
+ now: 2022-04-03 00:14:32.470694
261
+ [26000] train_loss=0.0198477 valid_loss=0.0271728 valid_pos_acc=0.972293 valid_acc=0.994845 / 0.909642 / 0.792003 best_acc=0.994947
262
+ now: 2022-04-03 00:26:30.440395
263
+ [26200] train_loss=0.0182446 valid_loss=0.0269998 valid_pos_acc=0.972797 valid_acc=0.994936 / 0.90766 / 0.791791 best_acc=0.994947
264
+ now: 2022-04-03 00:38:11.226201
265
+ [26400] train_loss=0.0188574 valid_loss=0.0277104 valid_pos_acc=0.971779 valid_acc=0.994897 / 0.91142 / 0.795515 best_acc=0.994947
266
+ now: 2022-04-03 00:50:19.930552
267
+ [26600] train_loss=0.0195086 valid_loss=0.0266913 valid_pos_acc=0.972246 valid_acc=0.994884 / 0.904172 / 0.791556 best_acc=0.994947
268
+ now: 2022-04-03 01:02:40.190107
269
+ [26800] train_loss=0.0204701 valid_loss=0.0262269 valid_pos_acc=0.972677 valid_acc=0.994895 / 0.901927 / 0.792913 best_acc=0.994947
270
+ now: 2022-04-03 01:14:37.242724
271
+ [27000] train_loss=0.0227353 valid_loss=0.0265531 valid_pos_acc=0.972556 valid_acc=0.994843 / 0.907322 / 0.797663 best_acc=0.994947
272
+ now: 2022-04-03 01:26:32.001134
273
+ [27200] train_loss=0.0199549 valid_loss=0.0256493 valid_pos_acc=0.972434 valid_acc=0.994821 / 0.907325 / 0.799249 best_acc=0.994947
274
+ now: 2022-04-03 01:38:35.194802
275
+ [27400] train_loss=0.0178435 valid_loss=0.0273656 valid_pos_acc=0.972109 valid_acc=0.994851 / 0.911329 / 0.795818 best_acc=0.994947
276
+ now: 2022-04-03 01:50:47.426096
277
+ [27600] train_loss=0.0196421 valid_loss=0.0268288 valid_pos_acc=0.972094 valid_acc=0.994532 / 0.90122 / 0.788515 best_acc=0.994947
278
+ now: 2022-04-03 02:02:43.219077
279
+ [27800] train_loss=0.0218845 valid_loss=0.0267693 valid_pos_acc=0.972343 valid_acc=0.994884 / 0.904811 / 0.790976 best_acc=0.994947
280
+ now: 2022-04-03 02:14:42.299297
281
+ [28000] train_loss=0.0201817 valid_loss=0.0273861 valid_pos_acc=0.972345 valid_acc=0.994552 / 0.904774 / 0.792625 best_acc=0.994947
282
+ now: 2022-04-03 02:26:51.221363
283
+ [28200] train_loss=0.0200979 valid_loss=0.026012 valid_pos_acc=0.972452 valid_acc=0.994995 / 0.908827 / 0.797558 best_acc=0.994995
284
+ now: 2022-04-03 02:39:03.091258
285
+ [28400] train_loss=0.0201783 valid_loss=0.0256883 valid_pos_acc=0.97261 valid_acc=0.994864 / 0.907562 / 0.798298 best_acc=0.994995
286
+ now: 2022-04-03 02:51:04.300548
287
+ [28600] train_loss=0.0194111 valid_loss=0.0273932 valid_pos_acc=0.971562 valid_acc=0.994906 / 0.911251 / 0.795228 best_acc=0.994995
288
+ now: 2022-04-03 03:03:10.729523
289
+ [28800] train_loss=0.0223215 valid_loss=0.0264798 valid_pos_acc=0.972521 valid_acc=0.994823 / 0.91054 / 0.798156 best_acc=0.994995
290
+ now: 2022-04-03 03:15:16.733585
291
+ [29000] train_loss=0.0203798 valid_loss=0.026267 valid_pos_acc=0.972539 valid_acc=0.994979 / 0.910358 / 0.797495 best_acc=0.994995
292
+ now: 2022-04-03 03:27:15.662423
293
+ [29200] train_loss=0.0198116 valid_loss=0.0271517 valid_pos_acc=0.972395 valid_acc=0.994804 / 0.90796 / 0.792402 best_acc=0.994995
294
+ now: 2022-04-03 03:39:25.302705
295
+ [29400] train_loss=0.0214404 valid_loss=0.0256727 valid_pos_acc=0.973261 valid_acc=0.994984 / 0.910775 / 0.799215 best_acc=0.994995
296
+ now: 2022-04-03 03:51:35.147315
297
+ [29600] train_loss=0.0187954 valid_loss=0.0264936 valid_pos_acc=0.97286 valid_acc=0.99499 / 0.91241 / 0.795951 best_acc=0.994995
298
+ now: 2022-04-03 04:03:37.804465
299
+ [29800] train_loss=0.0205343 valid_loss=0.0262582 valid_pos_acc=0.972979 valid_acc=0.994997 / 0.910193 / 0.797294 best_acc=0.994997
300
+ now: 2022-04-03 04:15:40.869840
301
+ [30000] train_loss=0.0212577 valid_loss=0.0263549 valid_pos_acc=0.972243 valid_acc=0.994684 / 0.909313 / 0.798057 best_acc=0.994997
302
+ now: 2022-04-03 04:27:42.822407
303
+ [30200] train_loss=0.0183862 valid_loss=0.0254776 valid_pos_acc=0.972784 valid_acc=0.994877 / 0.903545 / 0.791495 best_acc=0.994997
304
+ now: 2022-04-03 04:40:02.431413
305
+ [30400] train_loss=0.0197688 valid_loss=0.0264948 valid_pos_acc=0.973224 valid_acc=0.994997 / 0.911333 / 0.793087 best_acc=0.994997
306
+ now: 2022-04-03 04:52:02.140132
307
+ [30600] train_loss=0.0203299 valid_loss=0.0260468 valid_pos_acc=0.972966 valid_acc=0.994799 / 0.902308 / 0.791264 best_acc=0.994997
308
+ now: 2022-04-03 05:04:02.910278
309
+ [30800] train_loss=0.0216859 valid_loss=0.0270245 valid_pos_acc=0.972929 valid_acc=0.994676 / 0.910333 / 0.796835 best_acc=0.994997
310
+ now: 2022-04-03 05:16:05.153874
311
+ [31000] train_loss=0.0211773 valid_loss=0.0258814 valid_pos_acc=0.972768 valid_acc=0.994916 / 0.907256 / 0.793468 best_acc=0.994997
312
+ now: 2022-04-03 05:28:07.493337
313
+ [31200] train_loss=0.0202802 valid_loss=0.0254245 valid_pos_acc=0.972955 valid_acc=0.995092 / 0.914897 / 0.805234 best_acc=0.995092
314
+ now: 2022-04-03 05:40:12.780431
315
+ [31400] train_loss=0.0194152 valid_loss=0.0259566 valid_pos_acc=0.972664 valid_acc=0.99496 / 0.919699 / 0.808312 best_acc=0.995092
316
+ now: 2022-04-03 05:52:16.611374
317
+ [31600] train_loss=0.019009 valid_loss=0.0260382 valid_pos_acc=0.97263 valid_acc=0.995129 / 0.917311 / 0.804833 best_acc=0.995129
318
+ now: 2022-04-03 06:04:15.437060
319
+ [31800] train_loss=0.0205566 valid_loss=0.0260822 valid_pos_acc=0.973148 valid_acc=0.994979 / 0.919627 / 0.808927 best_acc=0.995129
320
+ now: 2022-04-03 06:16:10.198962
321
+ [32000] train_loss=0.0192955 valid_loss=0.0259151 valid_pos_acc=0.973142 valid_acc=0.995034 / 0.9158 / 0.804333 best_acc=0.995129
322
+ now: 2022-04-03 06:28:05.111366
323
+ [32200] train_loss=0.0204335 valid_loss=0.0255095 valid_pos_acc=0.972736 valid_acc=0.995073 / 0.916156 / 0.810316 best_acc=0.995129
324
+ now: 2022-04-03 06:40:09.460506
325
+ [32400] train_loss=0.0201043 valid_loss=0.0261998 valid_pos_acc=0.972545 valid_acc=0.994929 / 0.91379 / 0.802841 best_acc=0.995129
326
+ now: 2022-04-03 06:52:23.127760
327
+ [32600] train_loss=0.0180875 valid_loss=0.0249739 valid_pos_acc=0.973335 valid_acc=0.994997 / 0.912169 / 0.802927 best_acc=0.995129
328
+ now: 2022-04-03 07:04:31.009479
329
+ [32800] train_loss=0.0198901 valid_loss=0.0254487 valid_pos_acc=0.972621 valid_acc=0.995029 / 0.920603 / 0.80979 best_acc=0.995129
330
+ now: 2022-04-03 07:16:32.128110
331
+ [33000] train_loss=0.0208962 valid_loss=0.0254032 valid_pos_acc=0.972883 valid_acc=0.994979 / 0.909972 / 0.799621 best_acc=0.995129
332
+ now: 2022-04-03 07:28:44.400824
333
+ [33200] train_loss=0.0201999 valid_loss=0.0258948 valid_pos_acc=0.972847 valid_acc=0.994801 / 0.911759 / 0.799804 best_acc=0.995129
334
+ now: 2022-04-03 07:40:49.361680
335
+ [33400] train_loss=0.0217783 valid_loss=0.0256737 valid_pos_acc=0.973255 valid_acc=0.994951 / 0.914217 / 0.800689 best_acc=0.995129
336
+ now: 2022-04-03 07:52:50.822397
337
+ [33600] train_loss=0.0198491 valid_loss=0.0264241 valid_pos_acc=0.972823 valid_acc=0.994986 / 0.912473 / 0.801322 best_acc=0.995129
338
+ now: 2022-04-03 08:04:54.092732
339
+ [33800] train_loss=0.0221377 valid_loss=0.02493 valid_pos_acc=0.972877 valid_acc=0.994938 / 0.915101 / 0.804716 best_acc=0.995129
340
+ now: 2022-04-03 08:16:54.243602
341
+ [34000] train_loss=0.0213205 valid_loss=0.025545 valid_pos_acc=0.972677 valid_acc=0.994979 / 0.915158 / 0.805311 best_acc=0.995129
342
+ now: 2022-04-03 08:29:03.784710
343
+ [34200] train_loss=0.0192532 valid_loss=0.0251619 valid_pos_acc=0.97335 valid_acc=0.995099 / 0.916618 / 0.804815 best_acc=0.995129
344
+ now: 2022-04-03 08:41:15.345717
345
+ [34400] train_loss=0.0219833 valid_loss=0.0255126 valid_pos_acc=0.97335 valid_acc=0.995068 / 0.91454 / 0.801902 best_acc=0.995129
346
+ now: 2022-04-03 08:53:18.026172
347
+ [34600] train_loss=0.02057 valid_loss=0.0257689 valid_pos_acc=0.973476 valid_acc=0.995138 / 0.923171 / 0.810884 best_acc=0.995138
348
+ now: 2022-04-03 09:05:29.405654
349
+ [34800] train_loss=0.0212472 valid_loss=0.0260386 valid_pos_acc=0.973087 valid_acc=0.995238 / 0.919353 / 0.805834 best_acc=0.995238
350
+ now: 2022-04-03 09:17:41.688908
351
+ [35000] train_loss=0.0193925 valid_loss=0.02788 valid_pos_acc=0.972441 valid_acc=0.994587 / 0.915334 / 0.800469 best_acc=0.995238
352
+ now: 2022-04-03 09:29:44.816243
353
+ [35200] train_loss=0.0190577 valid_loss=0.0251073 valid_pos_acc=0.972968 valid_acc=0.995166 / 0.916023 / 0.808521 best_acc=0.995238
354
+ now: 2022-04-03 09:41:43.856892
355
+ [35400] train_loss=0.0225248 valid_loss=0.0244108 valid_pos_acc=0.973092 valid_acc=0.994988 / 0.911331 / 0.806775 best_acc=0.995238
356
+ now: 2022-04-03 09:53:53.841427
357
+ [35600] train_loss=0.0204164 valid_loss=0.0257028 valid_pos_acc=0.972951 valid_acc=0.994999 / 0.913099 / 0.803078 best_acc=0.995238
358
+ now: 2022-04-03 10:05:54.756481
359
+ [35800] train_loss=0.0207206 valid_loss=0.0250318 valid_pos_acc=0.973441 valid_acc=0.995112 / 0.915014 / 0.809688 best_acc=0.995238
360
+ now: 2022-04-03 10:17:52.373071
361
+ [36000] train_loss=0.0210285 valid_loss=0.0264345 valid_pos_acc=0.972836 valid_acc=0.994726 / 0.912637 / 0.802366 best_acc=0.995238
362
+ now: 2022-04-03 10:29:48.836136
363
+ [36200] train_loss=0.0210124 valid_loss=0.0253509 valid_pos_acc=0.972771 valid_acc=0.994962 / 0.908097 / 0.799097 best_acc=0.995238
364
+ now: 2022-04-03 10:41:53.650854
365
+ [36400] train_loss=0.0206329 valid_loss=0.0255921 valid_pos_acc=0.973576 valid_acc=0.995097 / 0.915445 / 0.807018 best_acc=0.995238
366
+ now: 2022-04-03 10:54:04.782468
367
+ [36600] train_loss=0.0190987 valid_loss=0.025047 valid_pos_acc=0.973278 valid_acc=0.99504 / 0.911809 / 0.804772 best_acc=0.995238
368
+ now: 2022-04-03 11:06:14.982105
369
+ [36800] train_loss=0.0193329 valid_loss=0.0255344 valid_pos_acc=0.973205 valid_acc=0.994995 / 0.914842 / 0.810892 best_acc=0.995238
370
+ now: 2022-04-03 11:18:21.542298
371
+ [37000] train_loss=0.019776 valid_loss=0.0257551 valid_pos_acc=0.973228 valid_acc=0.995025 / 0.911626 / 0.801857 best_acc=0.995238
372
+ now: 2022-04-03 11:30:14.909051
373
+ [37200] train_loss=0.0203762 valid_loss=0.0253398 valid_pos_acc=0.973005 valid_acc=0.995255 / 0.91017 / 0.804603 best_acc=0.995255
374
+ now: 2022-04-03 11:42:31.753627
375
+ [37400] train_loss=0.0188329 valid_loss=0.0251868 valid_pos_acc=0.973304 valid_acc=0.995248 / 0.915374 / 0.809239 best_acc=0.995255
376
+ now: 2022-04-03 11:54:33.700994
377
+ [37600] train_loss=0.0183661 valid_loss=0.0254057 valid_pos_acc=0.973443 valid_acc=0.995225 / 0.914564 / 0.805779 best_acc=0.995255
378
+ now: 2022-04-03 12:06:21.313158
379
+ [37800] train_loss=0.0211401 valid_loss=0.0246414 valid_pos_acc=0.973185 valid_acc=0.995164 / 0.914028 / 0.805807 best_acc=0.995255
380
+ now: 2022-04-03 12:18:25.200242
381
+ [38000] train_loss=0.021069 valid_loss=0.0244758 valid_pos_acc=0.973411 valid_acc=0.995285 / 0.914013 / 0.806843 best_acc=0.995285
382
+ now: 2022-04-03 12:30:27.765559
383
+ [38200] train_loss=0.0213957 valid_loss=0.0233822 valid_pos_acc=0.973997 valid_acc=0.995231 / 0.918159 / 0.809317 best_acc=0.995285
384
+ now: 2022-04-03 12:42:34.527640
385
+ [38400] train_loss=0.01998 valid_loss=0.0246681 valid_pos_acc=0.973157 valid_acc=0.995144 / 0.911465 / 0.802431 best_acc=0.995285
386
+ now: 2022-04-03 12:54:35.718367
387
+ [38600] train_loss=0.0219669 valid_loss=0.0265307 valid_pos_acc=0.973142 valid_acc=0.994856 / 0.913512 / 0.802637 best_acc=0.995285
388
+ now: 2022-04-03 13:06:31.081322
389
+ [38800] train_loss=0.0200181 valid_loss=0.0254716 valid_pos_acc=0.972988 valid_acc=0.995194 / 0.918644 / 0.809186 best_acc=0.995285
390
+ now: 2022-04-03 13:18:35.624914
391
+ [39000] train_loss=0.0210553 valid_loss=0.0242162 valid_pos_acc=0.973558 valid_acc=0.995114 / 0.914893 / 0.807304 best_acc=0.995285
392
+ now: 2022-04-03 13:30:40.043785
393
+ [39200] train_loss=0.0183049 valid_loss=0.0254598 valid_pos_acc=0.973745 valid_acc=0.995105 / 0.921543 / 0.812539 best_acc=0.995285
394
+ now: 2022-04-03 13:42:42.189633
395
+ [39400] train_loss=0.0204597 valid_loss=0.0242165 valid_pos_acc=0.973517 valid_acc=0.995151 / 0.916498 / 0.808433 best_acc=0.995285
396
+ now: 2022-04-03 13:54:41.379549
397
+ [39600] train_loss=0.0182456 valid_loss=0.0259895 valid_pos_acc=0.97315 valid_acc=0.995231 / 0.922271 / 0.812708 best_acc=0.995285
398
+ now: 2022-04-03 14:06:51.554192
399
+ [39800] train_loss=0.0163934 valid_loss=0.0258782 valid_pos_acc=0.973615 valid_acc=0.995144 / 0.916559 / 0.809627 best_acc=0.995285
400
+ now: 2022-04-03 14:18:44.881390
401
+ [40000] train_loss=0.015392 valid_loss=0.0261389 valid_pos_acc=0.974283 valid_acc=0.995225 / 0.920042 / 0.812848 best_acc=0.995285
402
+ now: 2022-04-03 14:30:42.900837
403
+ [40200] train_loss=0.0151938 valid_loss=0.0266225 valid_pos_acc=0.973879 valid_acc=0.995105 / 0.921502 / 0.815619 best_acc=0.995285
404
+ now: 2022-04-03 14:42:37.709297
405
+ [40400] train_loss=0.0156664 valid_loss=0.0252791 valid_pos_acc=0.973784 valid_acc=0.99524 / 0.920868 / 0.819319 best_acc=0.995285
406
+ now: 2022-04-03 14:54:43.815772
407
+ [40600] train_loss=0.0156026 valid_loss=0.0251239 valid_pos_acc=0.973782 valid_acc=0.995305 / 0.916191 / 0.813418 best_acc=0.995305
408
+ now: 2022-04-03 15:06:48.756040
409
+ [40800] train_loss=0.015617 valid_loss=0.0250889 valid_pos_acc=0.973797 valid_acc=0.995248 / 0.913875 / 0.809349 best_acc=0.995305
410
+ now: 2022-04-03 15:18:48.093498
411
+ [41000] train_loss=0.0160921 valid_loss=0.0254156 valid_pos_acc=0.974016 valid_acc=0.995324 / 0.918261 / 0.811825 best_acc=0.995324
412
+ now: 2022-04-03 15:30:49.330639
413
+ [41200] train_loss=0.0153382 valid_loss=0.0253904 valid_pos_acc=0.973777 valid_acc=0.995157 / 0.915275 / 0.813835 best_acc=0.995324
414
+ now: 2022-04-03 15:42:50.366939
415
+ [41400] train_loss=0.0149767 valid_loss=0.0262346 valid_pos_acc=0.97348 valid_acc=0.995216 / 0.916934 / 0.809716 best_acc=0.995324
416
+ now: 2022-04-03 15:54:50.211349
417
+ [41600] train_loss=0.0163188 valid_loss=0.0256865 valid_pos_acc=0.974062 valid_acc=0.995279 / 0.917293 / 0.810614 best_acc=0.995324
418
+ now: 2022-04-03 16:06:51.678378
419
+ [41800] train_loss=0.0152591 valid_loss=0.0257784 valid_pos_acc=0.973934 valid_acc=0.995374 / 0.918737 / 0.814463 best_acc=0.995374
420
+ now: 2022-04-03 16:19:01.215393
421
+ [42000] train_loss=0.0153742 valid_loss=0.0256425 valid_pos_acc=0.973743 valid_acc=0.995279 / 0.921392 / 0.819395 best_acc=0.995374
422
+ now: 2022-04-03 16:31:02.965955
423
+ [42200] train_loss=0.0170421 valid_loss=0.0256818 valid_pos_acc=0.973704 valid_acc=0.995233 / 0.921882 / 0.816565 best_acc=0.995374
424
+ now: 2022-04-03 16:42:58.346109
425
+ [42400] train_loss=0.0173119 valid_loss=0.0262359 valid_pos_acc=0.973567 valid_acc=0.995142 / 0.927474 / 0.824186 best_acc=0.995374
426
+ now: 2022-04-03 16:54:51.654450
427
+ [42600] train_loss=0.0151309 valid_loss=0.0263674 valid_pos_acc=0.974088 valid_acc=0.995246 / 0.928593 / 0.819351 best_acc=0.995374
428
+ now: 2022-04-03 17:06:53.863013
429
+ [42800] train_loss=0.0146644 valid_loss=0.0256878 valid_pos_acc=0.973406 valid_acc=0.995248 / 0.926729 / 0.822369 best_acc=0.995374
430
+ now: 2022-04-03 17:18:52.584371
431
+ [43000] train_loss=0.0165593 valid_loss=0.0256607 valid_pos_acc=0.973534 valid_acc=0.995077 / 0.92122 / 0.820859 best_acc=0.995374
432
+ now: 2022-04-03 17:30:49.691185
433
+ [43200] train_loss=0.0159887 valid_loss=0.0257545 valid_pos_acc=0.973704 valid_acc=0.995084 / 0.91855 / 0.813994 best_acc=0.995374
434
+ now: 2022-04-03 17:42:50.184875
435
+ [43400] train_loss=0.0176695 valid_loss=0.0257385 valid_pos_acc=0.973474 valid_acc=0.995123 / 0.92218 / 0.818823 best_acc=0.995374
436
+ now: 2022-04-03 17:54:44.709886
437
+ [43600] train_loss=0.017015 valid_loss=0.0253947 valid_pos_acc=0.973645 valid_acc=0.995281 / 0.918643 / 0.814191 best_acc=0.995374
438
+ now: 2022-04-03 18:06:47.227964
439
+ [43800] train_loss=0.0166192 valid_loss=0.0250654 valid_pos_acc=0.973309 valid_acc=0.995097 / 0.917002 / 0.81292 best_acc=0.995374
440
+ now: 2022-04-03 18:18:48.239180
441
+ [44000] train_loss=0.0179612 valid_loss=0.0247244 valid_pos_acc=0.974474 valid_acc=0.995259 / 0.922084 / 0.818199 best_acc=0.995374
442
+ now: 2022-04-03 18:30:58.780112
443
+ [44200] train_loss=0.0170823 valid_loss=0.0253463 valid_pos_acc=0.973673 valid_acc=0.995277 / 0.920828 / 0.819238 best_acc=0.995374
444
+ now: 2022-04-03 18:43:02.816823
445
+ [44400] train_loss=0.0180965 valid_loss=0.0255119 valid_pos_acc=0.973356 valid_acc=0.994997 / 0.920482 / 0.821315 best_acc=0.995374
446
+ now: 2022-04-03 18:55:04.839467
447
+ [44600] train_loss=0.0169638 valid_loss=0.0253875 valid_pos_acc=0.973916 valid_acc=0.995218 / 0.923398 / 0.819969 best_acc=0.995374
448
+ now: 2022-04-03 19:07:07.807908
449
+ [44800] train_loss=0.0175315 valid_loss=0.0250045 valid_pos_acc=0.973931 valid_acc=0.995225 / 0.919188 / 0.816001 best_acc=0.995374
450
+ now: 2022-04-03 19:18:53.383956
451
+ [45000] train_loss=0.0159776 valid_loss=0.0247916 valid_pos_acc=0.973947 valid_acc=0.99527 / 0.922533 / 0.819024 best_acc=0.995374
452
+ now: 2022-04-03 19:30:53.526194
453
+ [45200] train_loss=0.0171182 valid_loss=0.0256464 valid_pos_acc=0.97397 valid_acc=0.995225 / 0.923638 / 0.816987 best_acc=0.995374
454
+ now: 2022-04-03 19:42:50.236892
455
+ [45400] train_loss=0.0156089 valid_loss=0.0251274 valid_pos_acc=0.97379 valid_acc=0.995353 / 0.920763 / 0.815259 best_acc=0.995374
456
+ now: 2022-04-03 19:54:46.588740
457
+ [45600] train_loss=0.0163814 valid_loss=0.0255403 valid_pos_acc=0.973895 valid_acc=0.995298 / 0.923606 / 0.816856 best_acc=0.995374
458
+ now: 2022-04-03 20:06:53.199451
459
+ [45800] train_loss=0.017835 valid_loss=0.0246602 valid_pos_acc=0.973981 valid_acc=0.995372 / 0.923306 / 0.821397 best_acc=0.995374
460
+ now: 2022-04-03 20:19:00.767866
461
+ [46000] train_loss=0.0178728 valid_loss=0.025165 valid_pos_acc=0.973771 valid_acc=0.995149 / 0.922218 / 0.817612 best_acc=0.995374
462
+ now: 2022-04-03 20:30:58.988089
463
+ [46200] train_loss=0.0168901 valid_loss=0.0256853 valid_pos_acc=0.974276 valid_acc=0.995216 / 0.923973 / 0.820527 best_acc=0.995374
464
+ now: 2022-04-03 20:42:51.449300
465
+ [46400] train_loss=0.0167886 valid_loss=0.0253529 valid_pos_acc=0.973704 valid_acc=0.995159 / 0.92335 / 0.820036 best_acc=0.995374
466
+ now: 2022-04-03 20:55:00.160971
467
+ [46600] train_loss=0.0176656 valid_loss=0.0256036 valid_pos_acc=0.973929 valid_acc=0.995366 / 0.922949 / 0.818223 best_acc=0.995374
468
+ now: 2022-04-03 21:07:02.579327
469
+ [46800] train_loss=0.0168645 valid_loss=0.0251908 valid_pos_acc=0.974203 valid_acc=0.995385 / 0.92207 / 0.81494 best_acc=0.995385
470
+ now: 2022-04-03 21:18:57.696871
471
+ [47000] train_loss=0.0172549 valid_loss=0.0256528 valid_pos_acc=0.974391 valid_acc=0.995118 / 0.918959 / 0.811756 best_acc=0.995385
472
+ now: 2022-04-03 21:30:53.620916
473
+ [47200] train_loss=0.0177735 valid_loss=0.0247787 valid_pos_acc=0.97404 valid_acc=0.995071 / 0.922916 / 0.818151 best_acc=0.995385
474
+ now: 2022-04-03 21:42:47.443922
475
+ [47400] train_loss=0.0168849 valid_loss=0.0250654 valid_pos_acc=0.973947 valid_acc=0.995368 / 0.919669 / 0.817169 best_acc=0.995385
476
+ now: 2022-04-03 21:54:53.531320
477
+ [47600] train_loss=0.0162995 valid_loss=0.0245945 valid_pos_acc=0.973951 valid_acc=0.995285 / 0.922128 / 0.819555 best_acc=0.995385
478
+ now: 2022-04-03 22:06:54.960049
479
+ [47800] train_loss=0.0166094 valid_loss=0.0254666 valid_pos_acc=0.974647 valid_acc=0.995314 / 0.925199 / 0.818337 best_acc=0.995385
480
+ now: 2022-04-03 22:18:44.813398
481
+ [48000] train_loss=0.018357 valid_loss=0.0258162 valid_pos_acc=0.974018 valid_acc=0.99527 / 0.924671 / 0.820527 best_acc=0.995385
482
+ now: 2022-04-03 22:30:33.755723
483
+ [48200] train_loss=0.0168674 valid_loss=0.025839 valid_pos_acc=0.973747 valid_acc=0.995309 / 0.921703 / 0.813788 best_acc=0.995385
484
+ now: 2022-04-03 22:42:23.398005
485
+ [48400] train_loss=0.016813 valid_loss=0.0248057 valid_pos_acc=0.973541 valid_acc=0.995385 / 0.92499 / 0.824521 best_acc=0.995385
486
+ now: 2022-04-03 22:54:25.099880
487
+ [48600] train_loss=0.016574 valid_loss=0.0255942 valid_pos_acc=0.973608 valid_acc=0.995177 / 0.925379 / 0.821092 best_acc=0.995385
488
+ now: 2022-04-03 23:06:29.088401
489
+ [48800] train_loss=0.0164469 valid_loss=0.025258 valid_pos_acc=0.973758 valid_acc=0.995409 / 0.924966 / 0.819685 best_acc=0.995409
490
+ now: 2022-04-03 23:18:37.642881
491
+ [49000] train_loss=0.0179612 valid_loss=0.0246981 valid_pos_acc=0.974266 valid_acc=0.995318 / 0.920644 / 0.816254 best_acc=0.995409
492
+ now: 2022-04-03 23:30:33.455618
493
+ [49200] train_loss=0.0163115 valid_loss=0.025103 valid_pos_acc=0.973717 valid_acc=0.995405 / 0.917957 / 0.814416 best_acc=0.995409
494
+ now: 2022-04-03 23:42:28.866307
495
+ [49400] train_loss=0.0171099 valid_loss=0.0263086 valid_pos_acc=0.973556 valid_acc=0.995151 / 0.91743 / 0.810825 best_acc=0.995409
496
+ now: 2022-04-03 23:54:18.959772
497
+ [49600] train_loss=0.0189349 valid_loss=0.0246903 valid_pos_acc=0.974151 valid_acc=0.995272 / 0.92322 / 0.818726 best_acc=0.995409
498
+ now: 2022-04-04 00:06:15.909786
499
+ [49800] train_loss=0.0167492 valid_loss=0.025506 valid_pos_acc=0.974257 valid_acc=0.995439 / 0.925636 / 0.815986 best_acc=0.995439
500
+ now: 2022-04-04 00:18:24.239516
501
+ [50000] train_loss=0.0176046 valid_loss=0.024858 valid_pos_acc=0.974309 valid_acc=0.995346 / 0.930482 / 0.820187 best_acc=0.995439
502
+ now: 2022-04-04 00:30:15.506831
503
+ [50200] train_loss=0.0163029 valid_loss=0.0252902 valid_pos_acc=0.974383 valid_acc=0.995381 / 0.923226 / 0.818356 best_acc=0.995439
504
+ now: 2022-04-04 00:42:18.897896
505
+ [50400] train_loss=0.0171214 valid_loss=0.0246193 valid_pos_acc=0.974333 valid_acc=0.995368 / 0.921255 / 0.81669 best_acc=0.995439
506
+ now: 2022-04-04 00:54:13.841028
507
+ [50600] train_loss=0.0161805 valid_loss=0.0250435 valid_pos_acc=0.974437 valid_acc=0.995186 / 0.918271 / 0.813367 best_acc=0.995439
508
+ now: 2022-04-04 01:06:16.083462
509
+ [50800] train_loss=0.0179548 valid_loss=0.0245154 valid_pos_acc=0.974691 valid_acc=0.995259 / 0.922457 / 0.81699 best_acc=0.995439
510
+ now: 2022-04-04 01:18:13.790064
511
+ [51000] train_loss=0.0164793 valid_loss=0.0248721 valid_pos_acc=0.974378 valid_acc=0.995322 / 0.924732 / 0.817262 best_acc=0.995439
512
+ now: 2022-04-04 01:30:17.861722
513
+ [51200] train_loss=0.016939 valid_loss=0.0265039 valid_pos_acc=0.974007 valid_acc=0.995044 / 0.922191 / 0.811527 best_acc=0.995439
514
+ now: 2022-04-04 01:42:23.079103
515
+ [51400] train_loss=0.015836 valid_loss=0.0262405 valid_pos_acc=0.973289 valid_acc=0.995235 / 0.922246 / 0.817143 best_acc=0.995439
516
+ now: 2022-04-04 01:54:20.740833
517
+ [51600] train_loss=0.0175937 valid_loss=0.0250272 valid_pos_acc=0.973819 valid_acc=0.995214 / 0.925015 / 0.824207 best_acc=0.995439
518
+ now: 2022-04-04 02:06:12.246740
519
+ [51800] train_loss=0.0194151 valid_loss=0.0250101 valid_pos_acc=0.973567 valid_acc=0.995253 / 0.921771 / 0.822537 best_acc=0.995439
520
+ now: 2022-04-04 02:18:00.425728
521
+ [52000] train_loss=0.0175319 valid_loss=0.0252638 valid_pos_acc=0.973914 valid_acc=0.995235 / 0.92165 / 0.81825 best_acc=0.995439
522
+ now: 2022-04-04 02:30:00.527309
523
+ [52200] train_loss=0.0177649 valid_loss=0.0251296 valid_pos_acc=0.974302 valid_acc=0.995218 / 0.916528 / 0.820442 best_acc=0.995439
524
+ now: 2022-04-04 02:42:15.444828
525
+ [52400] train_loss=0.0142718 valid_loss=0.0261507 valid_pos_acc=0.974112 valid_acc=0.995218 / 0.926747 / 0.823267 best_acc=0.995439
526
+ now: 2022-04-04 02:54:22.534812
527
+ [52600] train_loss=0.0181186 valid_loss=0.024454 valid_pos_acc=0.974678 valid_acc=0.995266 / 0.920649 / 0.821322 best_acc=0.995439
528
+ now: 2022-04-04 03:06:28.190695
529
+ [52800] train_loss=0.0186914 valid_loss=0.0248696 valid_pos_acc=0.97445 valid_acc=0.995272 / 0.925439 / 0.824017 best_acc=0.995439
530
+ now: 2022-04-04 03:18:37.506758
531
+ [53000] train_loss=0.0180586 valid_loss=0.0249947 valid_pos_acc=0.973493 valid_acc=0.995246 / 0.922071 / 0.820278 best_acc=0.995439
532
+ now: 2022-04-04 03:30:42.193804
533
+ [53200] train_loss=0.0174878 valid_loss=0.0238657 valid_pos_acc=0.974348 valid_acc=0.995357 / 0.928082 / 0.822476 best_acc=0.995439
534
+ now: 2022-04-04 03:42:35.505948
535
+ [53400] train_loss=0.0175929 valid_loss=0.0238299 valid_pos_acc=0.974042 valid_acc=0.995331 / 0.921639 / 0.819943 best_acc=0.995439
536
+ now: 2022-04-04 03:54:28.949782
537
+ [53600] train_loss=0.0177671 valid_loss=0.0252258 valid_pos_acc=0.974318 valid_acc=0.995259 / 0.923376 / 0.819476 best_acc=0.995439
538
+ now: 2022-04-04 04:06:31.080946
539
+ [53800] train_loss=0.0181123 valid_loss=0.0245157 valid_pos_acc=0.974474 valid_acc=0.995344 / 0.926992 / 0.822282 best_acc=0.995439
540
+ now: 2022-04-04 04:18:31.824945
541
+ [54000] train_loss=0.0163909 valid_loss=0.0247618 valid_pos_acc=0.973955 valid_acc=0.995322 / 0.926681 / 0.819102 best_acc=0.995439
542
+ now: 2022-04-04 04:30:40.751754
543
+ [54200] train_loss=0.0182664 valid_loss=0.0249099 valid_pos_acc=0.973999 valid_acc=0.995418 / 0.928271 / 0.824931 best_acc=0.995439
544
+ now: 2022-04-04 04:42:35.399533
545
+ [54400] train_loss=0.0186873 valid_loss=0.0251008 valid_pos_acc=0.974205 valid_acc=0.995405 / 0.932078 / 0.826293 best_acc=0.995439
546
+ now: 2022-04-04 04:54:44.051651
547
+ [54600] train_loss=0.0176127 valid_loss=0.0242345 valid_pos_acc=0.97437 valid_acc=0.995394 / 0.926177 / 0.823775 best_acc=0.995439
548
+ now: 2022-04-04 05:06:49.804392
549
+ [54800] train_loss=0.0163823 valid_loss=0.0258135 valid_pos_acc=0.974289 valid_acc=0.995075 / 0.92924 / 0.822328 best_acc=0.995439
550
+ now: 2022-04-04 05:18:59.599641
551
+ [55000] train_loss=0.0179242 valid_loss=0.024379 valid_pos_acc=0.973957 valid_acc=0.995426 / 0.928179 / 0.828975 best_acc=0.995439
552
+ now: 2022-04-04 05:31:03.121724
553
+ [55200] train_loss=0.0184118 valid_loss=0.0241673 valid_pos_acc=0.974077 valid_acc=0.995316 / 0.929769 / 0.826971 best_acc=0.995439
554
+ now: 2022-04-04 05:43:02.794339
555
+ [55400] train_loss=0.0165821 valid_loss=0.0247912 valid_pos_acc=0.974302 valid_acc=0.995285 / 0.925138 / 0.817615 best_acc=0.995439
556
+ now: 2022-04-04 05:55:02.273212
557
+ [55600] train_loss=0.0186432 valid_loss=0.0251953 valid_pos_acc=0.974374 valid_acc=0.995231 / 0.926073 / 0.822135 best_acc=0.995439
558
+ now: 2022-04-04 06:07:03.296464
559
+ [55800] train_loss=0.0171283 valid_loss=0.0251252 valid_pos_acc=0.974962 valid_acc=0.995283 / 0.92335 / 0.820246 best_acc=0.995439
560
+ now: 2022-04-04 06:19:11.613771
561
+ [56000] train_loss=0.0186047 valid_loss=0.0247604 valid_pos_acc=0.97448 valid_acc=0.995277 / 0.927931 / 0.823571 best_acc=0.995439
562
+ now: 2022-04-04 06:31:02.274835
563
+ [56200] train_loss=0.0167043 valid_loss=0.0252192 valid_pos_acc=0.974166 valid_acc=0.995279 / 0.92564 / 0.822268 best_acc=0.995439
564
+ now: 2022-04-04 06:43:00.961416
565
+ [56400] train_loss=0.0193165 valid_loss=0.0252836 valid_pos_acc=0.974016 valid_acc=0.995268 / 0.9252 / 0.817183 best_acc=0.995439
566
+ now: 2022-04-04 06:55:01.506312
567
+ [56600] train_loss=0.0185221 valid_loss=0.0242256 valid_pos_acc=0.974535 valid_acc=0.995424 / 0.925529 / 0.821226 best_acc=0.995439
568
+ now: 2022-04-04 07:06:52.135175
569
+ [56800] train_loss=0.0171885 valid_loss=0.0252071 valid_pos_acc=0.974706 valid_acc=0.995227 / 0.924077 / 0.817878 best_acc=0.995439
570
+ now: 2022-04-04 07:18:50.316902
571
+ [57000] train_loss=0.0175959 valid_loss=0.0248842 valid_pos_acc=0.974235 valid_acc=0.995218 / 0.92808 / 0.823735 best_acc=0.995439
572
+ now: 2022-04-04 07:30:56.648830
573
+ [57200] train_loss=0.0171702 valid_loss=0.0248792 valid_pos_acc=0.974953 valid_acc=0.995392 / 0.927917 / 0.824835 best_acc=0.995439
574
+ now: 2022-04-04 07:43:00.234541
575
+ [57400] train_loss=0.0164997 valid_loss=0.0247215 valid_pos_acc=0.97473 valid_acc=0.995363 / 0.927725 / 0.822496 best_acc=0.995439
576
+ now: 2022-04-04 07:55:01.561685
577
+ [57600] train_loss=0.0175738 valid_loss=0.0246078 valid_pos_acc=0.974686 valid_acc=0.995413 / 0.9303 / 0.822759 best_acc=0.995439
578
+ now: 2022-04-04 08:07:03.437555
579
+ [57800] train_loss=0.0182974 valid_loss=0.0242409 valid_pos_acc=0.974461 valid_acc=0.995472 / 0.929586 / 0.827576 best_acc=0.995472
580
+ now: 2022-04-04 08:19:09.447609
581
+ [58000] train_loss=0.0175345 valid_loss=0.0243995 valid_pos_acc=0.974287 valid_acc=0.995515 / 0.93219 / 0.826722 best_acc=0.995515
582
+ now: 2022-04-04 08:31:08.680812
583
+ [58200] train_loss=0.0183634 valid_loss=0.0238503 valid_pos_acc=0.974051 valid_acc=0.995335 / 0.922815 / 0.821793 best_acc=0.995515
584
+ now: 2022-04-04 08:43:10.046244
585
+ [58400] train_loss=0.0163139 valid_loss=0.024504 valid_pos_acc=0.975008 valid_acc=0.995541 / 0.930931 / 0.823118 best_acc=0.995541
586
+ now: 2022-04-04 08:55:18.433955
587
+ [58600] train_loss=0.017843 valid_loss=0.0241767 valid_pos_acc=0.9746 valid_acc=0.995476 / 0.930497 / 0.832621 best_acc=0.995541
588
+ now: 2022-04-04 09:07:27.795974
589
+ [58800] train_loss=0.0176362 valid_loss=0.0243888 valid_pos_acc=0.974819 valid_acc=0.995457 / 0.929875 / 0.82877 best_acc=0.995541
590
+ now: 2022-04-04 09:19:29.918584
591
+ [59000] train_loss=0.017933 valid_loss=0.0242775 valid_pos_acc=0.974493 valid_acc=0.995507 / 0.930605 / 0.8321 best_acc=0.995541
592
+ now: 2022-04-04 09:31:33.387747
593
+ [59200] train_loss=0.0166545 valid_loss=0.0248671 valid_pos_acc=0.974669 valid_acc=0.995372 / 0.934492 / 0.832004 best_acc=0.995541
594
+ now: 2022-04-04 09:43:38.498786
595
+ [59400] train_loss=0.0144546 valid_loss=0.0250983 valid_pos_acc=0.974652 valid_acc=0.995494 / 0.93082 / 0.829281 best_acc=0.995541
596
+ now: 2022-04-04 09:55:40.006405
597
+ [59600] train_loss=0.0139316 valid_loss=0.02474 valid_pos_acc=0.974144 valid_acc=0.99545 / 0.932469 / 0.831644 best_acc=0.995541
598
+ now: 2022-04-04 10:07:46.456322
599
+ [59800] train_loss=0.0134991 valid_loss=0.0268217 valid_pos_acc=0.974183 valid_acc=0.995357 / 0.927956 / 0.825977 best_acc=0.995541
600
+ now: 2022-04-04 10:19:47.170813
601
+ [60000] train_loss=0.0152101 valid_loss=0.0253673 valid_pos_acc=0.974528 valid_acc=0.995463 / 0.932345 / 0.830733 best_acc=0.995541
602
+ now: 2022-04-04 10:31:54.967223
603
+ [60200] train_loss=0.0138335 valid_loss=0.0252645 valid_pos_acc=0.974877 valid_acc=0.995359 / 0.928942 / 0.827531 best_acc=0.995541
604
+ now: 2022-04-04 10:43:57.300159
605
+ [60400] train_loss=0.0136089 valid_loss=0.0252041 valid_pos_acc=0.974778 valid_acc=0.995353 / 0.92988 / 0.824825 best_acc=0.995541
606
+ now: 2022-04-04 10:55:50.855788
607
+ [60600] train_loss=0.013999 valid_loss=0.0255136 valid_pos_acc=0.975068 valid_acc=0.995389 / 0.93112 / 0.827874 best_acc=0.995541
608
+ now: 2022-04-04 11:07:45.181290
609
+ [60800] train_loss=0.0147645 valid_loss=0.024898 valid_pos_acc=0.974433 valid_acc=0.9954 / 0.932707 / 0.830214 best_acc=0.995541
610
+ now: 2022-04-04 11:19:31.882106
611
+ [61000] train_loss=0.0144584 valid_loss=0.0256239 valid_pos_acc=0.974613 valid_acc=0.995318 / 0.931253 / 0.826158 best_acc=0.995541
612
+ now: 2022-04-04 11:31:20.954761
613
+ [61200] train_loss=0.0148378 valid_loss=0.025611 valid_pos_acc=0.974391 valid_acc=0.995316 / 0.933647 / 0.831387 best_acc=0.995541
614
+ now: 2022-04-04 11:43:16.513341
615
+ [61400] train_loss=0.0164332 valid_loss=0.0245625 valid_pos_acc=0.974274 valid_acc=0.995533 / 0.933621 / 0.832726 best_acc=0.995541
616
+ now: 2022-04-04 11:55:19.484783
617
+ [61600] train_loss=0.0128714 valid_loss=0.0261392 valid_pos_acc=0.974322 valid_acc=0.995478 / 0.929096 / 0.826284 best_acc=0.995541
618
+ now: 2022-04-04 12:07:11.369487
619
+ [61800] train_loss=0.0133303 valid_loss=0.0250031 valid_pos_acc=0.974207 valid_acc=0.995429 / 0.929365 / 0.828611 best_acc=0.995541
620
+ now: 2022-04-04 12:19:18.413032
621
+ [62000] train_loss=0.015107 valid_loss=0.0254061 valid_pos_acc=0.974804 valid_acc=0.995526 / 0.9288 / 0.82914 best_acc=0.995541
622
+ now: 2022-04-04 12:31:20.055266
623
+ [62200] train_loss=0.0138232 valid_loss=0.0257889 valid_pos_acc=0.974927 valid_acc=0.99542 / 0.931311 / 0.830115 best_acc=0.995541
624
+ now: 2022-04-04 12:43:13.349393
625
+ [62400] train_loss=0.0141178 valid_loss=0.0256085 valid_pos_acc=0.97463 valid_acc=0.995348 / 0.926439 / 0.829561 best_acc=0.995541
626
+ now: 2022-04-04 12:55:00.906269
627
+ [62600] train_loss=0.0142263 valid_loss=0.0262747 valid_pos_acc=0.974962 valid_acc=0.995402 / 0.925902 / 0.82448 best_acc=0.995541
628
+ now: 2022-04-04 13:06:57.609555
629
+ [62800] train_loss=0.0147258 valid_loss=0.0259502 valid_pos_acc=0.974404 valid_acc=0.995504 / 0.928971 / 0.826749 best_acc=0.995541
630
+ now: 2022-04-04 13:18:51.020957
631
+ [63000] train_loss=0.0158923 valid_loss=0.0249411 valid_pos_acc=0.974795 valid_acc=0.995535 / 0.924258 / 0.825275 best_acc=0.995541
632
+ now: 2022-04-04 13:30:41.825005
633
+ [63200] train_loss=0.0129023 valid_loss=0.0256269 valid_pos_acc=0.974296 valid_acc=0.99553 / 0.924678 / 0.824117 best_acc=0.995541
634
+ now: 2022-04-04 13:42:34.233889
635
+ [63400] train_loss=0.0155354 valid_loss=0.0239082 valid_pos_acc=0.974923 valid_acc=0.995526 / 0.923666 / 0.830834 best_acc=0.995541
636
+ now: 2022-04-04 13:54:30.059726
637
+ [63600] train_loss=0.0151712 valid_loss=0.0252058 valid_pos_acc=0.97448 valid_acc=0.995481 / 0.926182 / 0.827018 best_acc=0.995541
638
+ now: 2022-04-04 14:06:25.923276
639
+ [63800] train_loss=0.0154875 valid_loss=0.0249389 valid_pos_acc=0.974504 valid_acc=0.995502 / 0.92361 / 0.826288 best_acc=0.995541
640
+ now: 2022-04-04 14:18:29.316170
641
+ [64000] train_loss=0.0151137 valid_loss=0.0256968 valid_pos_acc=0.974333 valid_acc=0.995459 / 0.924967 / 0.829635 best_acc=0.995541
642
+ now: 2022-04-04 14:30:34.246579
643
+ [64200] train_loss=0.0152567 valid_loss=0.0251615 valid_pos_acc=0.974465 valid_acc=0.995544 / 0.930084 / 0.832375 best_acc=0.995544
644
+ now: 2022-04-04 14:42:51.084382
645
+ [64400] train_loss=0.0145794 valid_loss=0.0253407 valid_pos_acc=0.975203 valid_acc=0.995587 / 0.929932 / 0.82946 best_acc=0.995587
646
+ now: 2022-04-04 14:54:58.723482
647
+ [64600] train_loss=0.0145396 valid_loss=0.0245939 valid_pos_acc=0.974552 valid_acc=0.995468 / 0.926718 / 0.83152 best_acc=0.995587
648
+ now: 2022-04-04 15:06:51.962741
649
+ [64800] train_loss=0.0141351 valid_loss=0.0257409 valid_pos_acc=0.975149 valid_acc=0.995465 / 0.92718 / 0.826698 best_acc=0.995587
650
+ now: 2022-04-04 15:18:55.351985
651
+ [65000] train_loss=0.0150484 valid_loss=0.0243046 valid_pos_acc=0.974875 valid_acc=0.995609 / 0.928307 / 0.831687 best_acc=0.995609
652
+ now: 2022-04-04 15:31:02.499770
653
+ [65200] train_loss=0.0137273 valid_loss=0.0247366 valid_pos_acc=0.974598 valid_acc=0.995478 / 0.927726 / 0.828495 best_acc=0.995609
654
+ now: 2022-04-04 15:43:12.293967
655
+ [65400] train_loss=0.0146922 valid_loss=0.0248483 valid_pos_acc=0.97481 valid_acc=0.995433 / 0.926791 / 0.822958 best_acc=0.995609
656
+ now: 2022-04-04 15:55:16.151208
657
+ [65600] train_loss=0.0163436 valid_loss=0.0252635 valid_pos_acc=0.974797 valid_acc=0.995318 / 0.927537 / 0.824153 best_acc=0.995609
658
+ now: 2022-04-04 16:07:11.212065
659
+ [65800] train_loss=0.0148312 valid_loss=0.0246498 valid_pos_acc=0.975016 valid_acc=0.995429 / 0.927171 / 0.82399 best_acc=0.995609
660
+ now: 2022-04-04 16:19:08.685980
661
+ [66000] train_loss=0.015976 valid_loss=0.0249127 valid_pos_acc=0.974988 valid_acc=0.995481 / 0.928756 / 0.828428 best_acc=0.995609
662
+ now: 2022-04-04 16:31:02.374897
663
+ [66200] train_loss=0.014783 valid_loss=0.0250196 valid_pos_acc=0.974611 valid_acc=0.995452 / 0.925158 / 0.827485 best_acc=0.995609
664
+ now: 2022-04-04 16:42:47.960220
665
+ [66400] train_loss=0.0157549 valid_loss=0.0247606 valid_pos_acc=0.974945 valid_acc=0.995422 / 0.927323 / 0.829169 best_acc=0.995609
666
+ now: 2022-04-04 16:54:48.551560
667
+ [66600] train_loss=0.0150551 valid_loss=0.0252349 valid_pos_acc=0.975318 valid_acc=0.995544 / 0.929646 / 0.829167 best_acc=0.995609
668
+ now: 2022-04-04 17:06:51.181160
669
+ [66800] train_loss=0.014767 valid_loss=0.0253356 valid_pos_acc=0.9747 valid_acc=0.995528 / 0.930906 / 0.830688 best_acc=0.995609
670
+ now: 2022-04-04 17:18:56.196395
671
+ [67000] train_loss=0.0143435 valid_loss=0.0247641 valid_pos_acc=0.974734 valid_acc=0.99553 / 0.930212 / 0.832562 best_acc=0.995609
672
+ now: 2022-04-04 17:30:49.412781
673
+ [67200] train_loss=0.0143362 valid_loss=0.0262739 valid_pos_acc=0.974728 valid_acc=0.995355 / 0.92964 / 0.824912 best_acc=0.995609
674
+ now: 2022-04-04 17:42:55.107881
675
+ [67400] train_loss=0.0162339 valid_loss=0.0246594 valid_pos_acc=0.974643 valid_acc=0.995465 / 0.927617 / 0.827191 best_acc=0.995609
676
+ now: 2022-04-04 17:54:54.064581
677
+ [67600] train_loss=0.0160092 valid_loss=0.0252428 valid_pos_acc=0.974259 valid_acc=0.995446 / 0.930018 / 0.833761 best_acc=0.995609
678
+ now: 2022-04-04 18:06:53.764486
679
+ [67800] train_loss=0.0152992 valid_loss=0.0253698 valid_pos_acc=0.97448 valid_acc=0.995392 / 0.926869 / 0.829631 best_acc=0.995609
680
+ now: 2022-04-04 18:18:51.081973
681
+ [68000] train_loss=0.0147154 valid_loss=0.0257934 valid_pos_acc=0.974166 valid_acc=0.995533 / 0.932229 / 0.833046 best_acc=0.995609
682
+ now: 2022-04-04 18:30:58.562494
683
+ [68200] train_loss=0.0133411 valid_loss=0.0261108 valid_pos_acc=0.974467 valid_acc=0.995472 / 0.928011 / 0.826824 best_acc=0.995609
684
+ now: 2022-04-04 18:42:51.828880
685
+ [68400] train_loss=0.0153007 valid_loss=0.0262793 valid_pos_acc=0.974007 valid_acc=0.995255 / 0.92542 / 0.825899 best_acc=0.995609
686
+ now: 2022-04-04 18:54:48.421492
687
+ [68600] train_loss=0.0156991 valid_loss=0.0250134 valid_pos_acc=0.974808 valid_acc=0.99552 / 0.925519 / 0.827053 best_acc=0.995609
688
+ now: 2022-04-04 19:06:46.853231
689
+ [68800] train_loss=0.0148552 valid_loss=0.0246115 valid_pos_acc=0.974407 valid_acc=0.995626 / 0.92897 / 0.829723 best_acc=0.995626
690
+ now: 2022-04-04 19:18:49.907609
691
+ [69000] train_loss=0.0148742 valid_loss=0.0240215 valid_pos_acc=0.974463 valid_acc=0.995648 / 0.932707 / 0.837616 best_acc=0.995648
692
+ now: 2022-04-04 19:30:58.657495
693
+ [69200] train_loss=0.0154729 valid_loss=0.0243906 valid_pos_acc=0.974762 valid_acc=0.995609 / 0.932612 / 0.832795 best_acc=0.995648
694
+ now: 2022-04-04 19:43:05.888719
695
+ [69400] train_loss=0.016396 valid_loss=0.0258633 valid_pos_acc=0.974834 valid_acc=0.995557 / 0.930491 / 0.82893 best_acc=0.995648
696
+ now: 2022-04-04 19:55:14.202470
697
+ [69600] train_loss=0.0151303 valid_loss=0.0253337 valid_pos_acc=0.975164 valid_acc=0.995507 / 0.926279 / 0.828464 best_acc=0.995648
698
+ now: 2022-04-04 20:07:14.789065
699
+ [69800] train_loss=0.0160168 valid_loss=0.0261905 valid_pos_acc=0.974936 valid_acc=0.995244 / 0.919724 / 0.822949 best_acc=0.995648
700
+ now: 2022-04-04 20:19:07.585227
701
+ [70000] train_loss=0.0145597 valid_loss=0.0257852 valid_pos_acc=0.974459 valid_acc=0.995411 / 0.922878 / 0.822821 best_acc=0.995648
702
+ now: 2022-04-04 20:31:00.634042
703
+ [70200] train_loss=0.0158036 valid_loss=0.0254432 valid_pos_acc=0.974986 valid_acc=0.995442 / 0.927001 / 0.827555 best_acc=0.995648
704
+ now: 2022-04-04 20:42:51.615293
705
+ [70400] train_loss=0.0152405 valid_loss=0.0246695 valid_pos_acc=0.974684 valid_acc=0.995517 / 0.926662 / 0.828915 best_acc=0.995648
706
+ now: 2022-04-04 20:54:54.958272
707
+ [70600] train_loss=0.0142744 valid_loss=0.0256758 valid_pos_acc=0.974656 valid_acc=0.995446 / 0.924052 / 0.827786 best_acc=0.995648
708
+ now: 2022-04-04 21:06:51.313734
709
+ [70800] train_loss=0.0156632 valid_loss=0.0254473 valid_pos_acc=0.97496 valid_acc=0.995385 / 0.92814 / 0.832267 best_acc=0.995648
710
+ now: 2022-04-04 21:18:43.761819
711
+ [71000] train_loss=0.0164862 valid_loss=0.0255363 valid_pos_acc=0.974172 valid_acc=0.995342 / 0.92511 / 0.826733 best_acc=0.995648
712
+ now: 2022-04-04 21:30:44.417684
713
+ [71200] train_loss=0.0138593 valid_loss=0.0255767 valid_pos_acc=0.974632 valid_acc=0.995394 / 0.930841 / 0.830682 best_acc=0.995648
714
+ now: 2022-04-04 21:42:42.850833
715
+ [71400] train_loss=0.0157498 valid_loss=0.025416 valid_pos_acc=0.975127 valid_acc=0.995617 / 0.928939 / 0.826405 best_acc=0.995648
716
+ now: 2022-04-04 21:54:46.086563
717
+ [71600] train_loss=0.0152817 valid_loss=0.0257021 valid_pos_acc=0.974856 valid_acc=0.995578 / 0.927289 / 0.824619 best_acc=0.995648
718
+ now: 2022-04-04 22:06:47.189552
719
+ [71800] train_loss=0.0149438 valid_loss=0.0264182 valid_pos_acc=0.974916 valid_acc=0.995507 / 0.931164 / 0.830416 best_acc=0.995648
720
+ now: 2022-04-04 22:18:39.870279
721
+ [72000] train_loss=0.0163576 valid_loss=0.0259418 valid_pos_acc=0.974795 valid_acc=0.995374 / 0.924427 / 0.824195 best_acc=0.995648
722
+ now: 2022-04-04 22:30:33.184078
723
+ [72200] train_loss=0.016734 valid_loss=0.0250246 valid_pos_acc=0.974524 valid_acc=0.995552 / 0.927199 / 0.829767 best_acc=0.995648
724
+ now: 2022-04-04 22:42:36.180178
725
+ [72400] train_loss=0.0154278 valid_loss=0.025262 valid_pos_acc=0.974578 valid_acc=0.9955 / 0.930839 / 0.835087 best_acc=0.995648
726
+ now: 2022-04-04 22:54:42.093287
727
+ [72600] train_loss=0.0154066 valid_loss=0.0257742 valid_pos_acc=0.973793 valid_acc=0.995394 / 0.933008 / 0.828687 best_acc=0.995648
728
+ now: 2022-04-04 23:06:54.703488
729
+ [72800] train_loss=0.0164941 valid_loss=0.0249958 valid_pos_acc=0.974504 valid_acc=0.995418 / 0.931091 / 0.83236 best_acc=0.995648
730
+ now: 2022-04-04 23:19:01.661367
731
+ [73000] train_loss=0.016436 valid_loss=0.0265112 valid_pos_acc=0.974443 valid_acc=0.995385 / 0.927996 / 0.829395 best_acc=0.995648
732
+ now: 2022-04-04 23:31:00.152399
733
+ [73200] train_loss=0.0162149 valid_loss=0.0255855 valid_pos_acc=0.974825 valid_acc=0.995448 / 0.928238 / 0.829202 best_acc=0.995648
734
+ now: 2022-04-04 23:43:00.392230
735
+ [73400] train_loss=0.0144402 valid_loss=0.0252108 valid_pos_acc=0.9747 valid_acc=0.995587 / 0.929875 / 0.834653 best_acc=0.995648
736
+ now: 2022-04-04 23:55:08.297531
737
+ [73600] train_loss=0.0156009 valid_loss=0.0243919 valid_pos_acc=0.974626 valid_acc=0.995526 / 0.930281 / 0.835324 best_acc=0.995648
738
+ now: 2022-04-05 00:07:21.557640
739
+ [73800] train_loss=0.0162462 valid_loss=0.0250821 valid_pos_acc=0.975079 valid_acc=0.995567 / 0.927734 / 0.826903 best_acc=0.995648
740
+ now: 2022-04-05 00:19:20.964981
741
+ [74000] train_loss=0.0149793 valid_loss=0.0246167 valid_pos_acc=0.974808 valid_acc=0.995598 / 0.930624 / 0.833348 best_acc=0.995648
742
+ now: 2022-04-05 00:31:27.707459
743
+ [74200] train_loss=0.016266 valid_loss=0.0240239 valid_pos_acc=0.975188 valid_acc=0.995574 / 0.931921 / 0.831013 best_acc=0.995648
744
+ now: 2022-04-05 00:43:16.494193
745
+ [74400] train_loss=0.0133132 valid_loss=0.0248758 valid_pos_acc=0.97506 valid_acc=0.99553 / 0.935143 / 0.835638 best_acc=0.995648
746
+ now: 2022-04-05 00:55:14.868213
747
+ [74600] train_loss=0.0161027 valid_loss=0.0249716 valid_pos_acc=0.97488 valid_acc=0.9955 / 0.933607 / 0.832621 best_acc=0.995648
748
+ now: 2022-04-05 01:07:25.462651
749
+ [74800] train_loss=0.014376 valid_loss=0.0250494 valid_pos_acc=0.975396 valid_acc=0.995504 / 0.933943 / 0.833992 best_acc=0.995648
750
+ now: 2022-04-05 01:19:25.187529
751
+ [75000] train_loss=0.0161117 valid_loss=0.0241017 valid_pos_acc=0.974871 valid_acc=0.995626 / 0.930057 / 0.833235 best_acc=0.995648
752
+ now: 2022-04-05 01:31:27.005032
753
+ [75200] train_loss=0.0158166 valid_loss=0.0236026 valid_pos_acc=0.974979 valid_acc=0.995587 / 0.929956 / 0.831575 best_acc=0.995648
754
+ now: 2022-04-05 01:43:35.408468
755
+ [75400] train_loss=0.0149878 valid_loss=0.0256847 valid_pos_acc=0.974949 valid_acc=0.995667 / 0.932654 / 0.832065 best_acc=0.995667
756
+ now: 2022-04-05 01:55:54.653418
757
+ [75600] train_loss=0.0155141 valid_loss=0.024818 valid_pos_acc=0.974934 valid_acc=0.995609 / 0.933709 / 0.831186 best_acc=0.995667
758
+ now: 2022-04-05 02:07:43.823418
759
+ [75800] train_loss=0.0144699 valid_loss=0.0258603 valid_pos_acc=0.975235 valid_acc=0.995617 / 0.933296 / 0.832935 best_acc=0.995667
760
+ now: 2022-04-05 02:19:44.987772
761
+ [76000] train_loss=0.0152171 valid_loss=0.0252886 valid_pos_acc=0.97524 valid_acc=0.995693 / 0.932408 / 0.835173 best_acc=0.995693
762
+ now: 2022-04-05 02:31:42.885767
763
+ [76200] train_loss=0.0155771 valid_loss=0.0255284 valid_pos_acc=0.97537 valid_acc=0.995643 / 0.935199 / 0.834838 best_acc=0.995693
764
+ now: 2022-04-05 02:43:44.509209
765
+ [76400] train_loss=0.0153983 valid_loss=0.0248721 valid_pos_acc=0.975023 valid_acc=0.995574 / 0.93275 / 0.835416 best_acc=0.995693
766
+ now: 2022-04-05 02:55:49.966478
767
+ [76600] train_loss=0.0154212 valid_loss=0.0257234 valid_pos_acc=0.975149 valid_acc=0.995465 / 0.930268 / 0.829088 best_acc=0.995693
768
+ now: 2022-04-05 03:08:00.620954
769
+ [76800] train_loss=0.0151146 valid_loss=0.0246848 valid_pos_acc=0.975071 valid_acc=0.995398 / 0.930615 / 0.832294 best_acc=0.995693
770
+ now: 2022-04-05 03:20:08.341056
771
+ [77000] train_loss=0.0160863 valid_loss=0.0271371 valid_pos_acc=0.974689 valid_acc=0.995244 / 0.93075 / 0.823407 best_acc=0.995693
772
+ now: 2022-04-05 03:32:17.172790
773
+ [77200] train_loss=0.0157593 valid_loss=0.0253599 valid_pos_acc=0.974697 valid_acc=0.995528 / 0.929039 / 0.8295 best_acc=0.995693
774
+ now: 2022-04-05 03:44:21.046580
775
+ [77400] train_loss=0.0161056 valid_loss=0.0245615 valid_pos_acc=0.974919 valid_acc=0.995472 / 0.93189 / 0.832852 best_acc=0.995693
776
+ now: 2022-04-05 03:56:20.405982
777
+ [77600] train_loss=0.0149618 valid_loss=0.0246419 valid_pos_acc=0.975073 valid_acc=0.995513 / 0.932703 / 0.831197 best_acc=0.995693
778
+ now: 2022-04-05 04:08:28.594227
779
+ [77800] train_loss=0.016911 valid_loss=0.0236979 valid_pos_acc=0.974723 valid_acc=0.995596 / 0.931261 / 0.832728 best_acc=0.995693
780
+ now: 2022-04-05 04:20:31.662452
781
+ [78000] train_loss=0.0165653 valid_loss=0.0239289 valid_pos_acc=0.974678 valid_acc=0.995502 / 0.929082 / 0.827912 best_acc=0.995693
782
+ now: 2022-04-05 04:32:28.019168
783
+ [78200] train_loss=0.0167179 valid_loss=0.0249402 valid_pos_acc=0.9747 valid_acc=0.995502 / 0.923062 / 0.82385 best_acc=0.995693
784
+ now: 2022-04-05 04:44:26.471615
785
+ [78400] train_loss=0.0167333 valid_loss=0.0240457 valid_pos_acc=0.9746 valid_acc=0.995606 / 0.92763 / 0.83109 best_acc=0.995693
786
+ now: 2022-04-05 04:56:24.357334
787
+ [78600] train_loss=0.0145129 valid_loss=0.0249626 valid_pos_acc=0.974871 valid_acc=0.995366 / 0.927649 / 0.828631 best_acc=0.995693
788
+ now: 2022-04-05 05:08:26.198865
789
+ [78800] train_loss=0.0172761 valid_loss=0.0238513 valid_pos_acc=0.975407 valid_acc=0.995611 / 0.925864 / 0.831329 best_acc=0.995693
790
+ now: 2022-04-05 05:20:20.936271
791
+ [79000] train_loss=0.0153685 valid_loss=0.0233658 valid_pos_acc=0.97557 valid_acc=0.995645 / 0.931677 / 0.832522 best_acc=0.995693
792
+ now: 2022-04-05 05:32:17.131240
793
+ [79200] train_loss=0.0127379 valid_loss=0.0258231 valid_pos_acc=0.974604 valid_acc=0.995533 / 0.933338 / 0.832997 best_acc=0.995693
794
+ now: 2022-04-05 05:44:21.126943
795
+ [79400] train_loss=0.0127164 valid_loss=0.025304 valid_pos_acc=0.975057 valid_acc=0.995698 / 0.930098 / 0.832254 best_acc=0.995698
796
+ now: 2022-04-05 05:56:18.071160
797
+ [79600] train_loss=0.0110389 valid_loss=0.0258263 valid_pos_acc=0.975101 valid_acc=0.995643 / 0.930643 / 0.835766 best_acc=0.995698
798
+ now: 2022-04-05 06:08:12.503631
799
+ [79800] train_loss=0.0132959 valid_loss=0.0263015 valid_pos_acc=0.974487 valid_acc=0.995446 / 0.932765 / 0.83365 best_acc=0.995698
800
+ now: 2022-04-05 06:20:20.191496
801
+ [80000] train_loss=0.0138857 valid_loss=0.025657 valid_pos_acc=0.974903 valid_acc=0.995635 / 0.931425 / 0.834477 best_acc=0.995698
802
+ now: 2022-04-05 06:32:30.331123
803
+ [80200] train_loss=0.0129796 valid_loss=0.0252719 valid_pos_acc=0.974886 valid_acc=0.995539 / 0.930875 / 0.836388 best_acc=0.995698
804
+ now: 2022-04-05 06:44:36.817366
805
+ [80400] train_loss=0.0126967 valid_loss=0.0255622 valid_pos_acc=0.97511 valid_acc=0.995624 / 0.930096 / 0.832861 best_acc=0.995698
806
+ now: 2022-04-05 06:56:45.374554
807
+ [80600] train_loss=0.0142307 valid_loss=0.0258526 valid_pos_acc=0.97511 valid_acc=0.995489 / 0.932336 / 0.832256 best_acc=0.995698
808
+ now: 2022-04-05 07:08:47.240323
809
+ [80800] train_loss=0.0138569 valid_loss=0.0268916 valid_pos_acc=0.974849 valid_acc=0.995609 / 0.936109 / 0.834517 best_acc=0.995698
810
+ now: 2022-04-05 07:21:07.181733
811
+ [81000] train_loss=0.0126373 valid_loss=0.0253882 valid_pos_acc=0.975253 valid_acc=0.995587 / 0.932359 / 0.834985 best_acc=0.995698
812
+ now: 2022-04-05 07:33:38.912179
813
+ [81200] train_loss=0.0130838 valid_loss=0.0268405 valid_pos_acc=0.974689 valid_acc=0.995491 / 0.933348 / 0.837004 best_acc=0.995698
814
+ now: 2022-04-05 07:45:58.486868
815
+ [81400] train_loss=0.0134847 valid_loss=0.0259947 valid_pos_acc=0.974574 valid_acc=0.99542 / 0.934234 / 0.837761 best_acc=0.995698
816
+ now: 2022-04-05 07:58:01.699833
817
+ [81600] train_loss=0.0127687 valid_loss=0.0267179 valid_pos_acc=0.975088 valid_acc=0.995461 / 0.932898 / 0.833357 best_acc=0.995698
818
+ now: 2022-04-05 08:10:22.110075
819
+ [81800] train_loss=0.0125759 valid_loss=0.026263 valid_pos_acc=0.974645 valid_acc=0.995409 / 0.934211 / 0.83588 best_acc=0.995698
820
+ now: 2022-04-05 08:22:30.965145
821
+ [82000] train_loss=0.0138023 valid_loss=0.0262454 valid_pos_acc=0.97488 valid_acc=0.995481 / 0.933091 / 0.835412 best_acc=0.995698
822
+ now: 2022-04-05 08:34:55.047735
823
+ [82200] train_loss=0.0127197 valid_loss=0.0260373 valid_pos_acc=0.975346 valid_acc=0.995517 / 0.934354 / 0.834983 best_acc=0.995698
824
+ now: 2022-04-05 08:47:21.849999
825
+ [82400] train_loss=0.0133346 valid_loss=0.0259693 valid_pos_acc=0.975274 valid_acc=0.995563 / 0.932429 / 0.834782 best_acc=0.995698
826
+ now: 2022-04-05 08:59:46.077640
827
+ [82600] train_loss=0.0119814 valid_loss=0.0268453 valid_pos_acc=0.97447 valid_acc=0.995485 / 0.935298 / 0.838585 best_acc=0.995698
828
+ now: 2022-04-05 09:12:10.429291
829
+ [82800] train_loss=0.0138231 valid_loss=0.0256561 valid_pos_acc=0.975485 valid_acc=0.995604 / 0.932017 / 0.837154 best_acc=0.995698
830
+ now: 2022-04-05 09:24:25.503501
831
+ [83000] train_loss=0.0138415 valid_loss=0.0260283 valid_pos_acc=0.975131 valid_acc=0.99547 / 0.93434 / 0.835245 best_acc=0.995698
832
+ now: 2022-04-05 09:36:38.417154
833
+ [83200] train_loss=0.0124775 valid_loss=0.0260958 valid_pos_acc=0.975194 valid_acc=0.995672 / 0.935893 / 0.837375 best_acc=0.995698
834
+ now: 2022-04-05 09:48:54.047384
835
+ [83400] train_loss=0.0132698 valid_loss=0.0259098 valid_pos_acc=0.974975 valid_acc=0.995465 / 0.934838 / 0.8402 best_acc=0.995698
836
+ now: 2022-04-05 10:00:55.953481
837
+ [83600] train_loss=0.0134172 valid_loss=0.0258395 valid_pos_acc=0.975396 valid_acc=0.995544 / 0.934034 / 0.836583 best_acc=0.995698
838
+ now: 2022-04-05 10:12:56.731723
839
+ [83800] train_loss=0.0129522 valid_loss=0.0251187 valid_pos_acc=0.975463 valid_acc=0.995596 / 0.933734 / 0.833624 best_acc=0.995698
840
+ now: 2022-04-05 10:24:54.745844
841
+ [84000] train_loss=0.0141191 valid_loss=0.0254002 valid_pos_acc=0.975463 valid_acc=0.995585 / 0.935384 / 0.836613 best_acc=0.995698
842
+ now: 2022-04-05 10:37:00.190609
843
+ [84200] train_loss=0.013092 valid_loss=0.0266141 valid_pos_acc=0.97488 valid_acc=0.995517 / 0.934064 / 0.836598 best_acc=0.995698
844
+ now: 2022-04-05 10:49:11.600372
845
+ [84400] train_loss=0.0146677 valid_loss=0.025732 valid_pos_acc=0.975146 valid_acc=0.995476 / 0.934118 / 0.834405 best_acc=0.995698
846
+ now: 2022-04-05 11:01:26.620881
847
+ [84600] train_loss=0.0147797 valid_loss=0.0257641 valid_pos_acc=0.974908 valid_acc=0.9955 / 0.931502 / 0.830953 best_acc=0.995698
848
+ now: 2022-04-05 11:13:35.462462
849
+ [84800] train_loss=0.0130389 valid_loss=0.0259705 valid_pos_acc=0.97511 valid_acc=0.995654 / 0.933228 / 0.838112 best_acc=0.995698
850
+ now: 2022-04-05 11:25:41.335912
851
+ [85000] train_loss=0.0130362 valid_loss=0.0270572 valid_pos_acc=0.975012 valid_acc=0.995524 / 0.935052 / 0.83898 best_acc=0.995698
852
+ now: 2022-04-05 11:37:45.634992
853
+ [85200] train_loss=0.0136579 valid_loss=0.0259226 valid_pos_acc=0.975407 valid_acc=0.995561 / 0.932682 / 0.833727 best_acc=0.995698
854
+ now: 2022-04-05 11:49:49.322682
855
+ [85400] train_loss=0.0138471 valid_loss=0.0278054 valid_pos_acc=0.974313 valid_acc=0.995253 / 0.926638 / 0.826328 best_acc=0.995698
856
+ now: 2022-04-05 12:02:06.577902
857
+ [85600] train_loss=0.0141456 valid_loss=0.0270836 valid_pos_acc=0.97522 valid_acc=0.995439 / 0.931749 / 0.83259 best_acc=0.995698
858
+ now: 2022-04-05 12:14:09.865872
859
+ [85800] train_loss=0.0136233 valid_loss=0.0261673 valid_pos_acc=0.975724 valid_acc=0.995485 / 0.932979 / 0.832328 best_acc=0.995698
860
+ now: 2022-04-05 12:26:24.557886
861
+ [86000] train_loss=0.0129815 valid_loss=0.026333 valid_pos_acc=0.975672 valid_acc=0.995426 / 0.933565 / 0.836186 best_acc=0.995698
862
+ now: 2022-04-05 12:38:22.054353
863
+ [86200] train_loss=0.0138994 valid_loss=0.025931 valid_pos_acc=0.975602 valid_acc=0.995533 / 0.927751 / 0.828184 best_acc=0.995698
864
+ now: 2022-04-05 12:50:25.067251
865
+ [86400] train_loss=0.0136555 valid_loss=0.0258883 valid_pos_acc=0.975031 valid_acc=0.995409 / 0.927127 / 0.829497 best_acc=0.995698
866
+ now: 2022-04-05 13:02:21.412803
867
+ [86600] train_loss=0.0142481 valid_loss=0.0267677 valid_pos_acc=0.975535 valid_acc=0.995535 / 0.929604 / 0.830529 best_acc=0.995698
868
+ now: 2022-04-05 13:14:29.577497
869
+ [86800] train_loss=0.0142199 valid_loss=0.026345 valid_pos_acc=0.974986 valid_acc=0.995331 / 0.93041 / 0.828971 best_acc=0.995698
870
+ now: 2022-04-05 13:26:39.137708
871
+ [87000] train_loss=0.0132453 valid_loss=0.0277596 valid_pos_acc=0.975698 valid_acc=0.995459 / 0.929227 / 0.825919 best_acc=0.995698
872
+ now: 2022-04-05 13:38:41.105816
873
+ [87200] train_loss=0.0140319 valid_loss=0.0262992 valid_pos_acc=0.975387 valid_acc=0.995372 / 0.925482 / 0.828088 best_acc=0.995698
874
+ now: 2022-04-05 13:50:47.452684
875
+ [87400] train_loss=0.0143099 valid_loss=0.0264765 valid_pos_acc=0.974565 valid_acc=0.995402 / 0.930457 / 0.829678 best_acc=0.995698
876
+ now: 2022-04-05 14:02:54.386306
877
+ [87600] train_loss=0.0139416 valid_loss=0.0260449 valid_pos_acc=0.975598 valid_acc=0.99555 / 0.930947 / 0.830807 best_acc=0.995698
878
+ now: 2022-04-05 14:14:49.214068
879
+ [87800] train_loss=0.013045 valid_loss=0.0260407 valid_pos_acc=0.974786 valid_acc=0.995637 / 0.934269 / 0.839114 best_acc=0.995698
880
+ now: 2022-04-05 14:26:42.833491
881
+ [88000] train_loss=0.0141046 valid_loss=0.0263996 valid_pos_acc=0.975235 valid_acc=0.995537 / 0.933095 / 0.831651 best_acc=0.995698
882
+ now: 2022-04-05 14:38:40.768985
883
+ [88200] train_loss=0.0146935 valid_loss=0.0263447 valid_pos_acc=0.975728 valid_acc=0.995494 / 0.932254 / 0.829169 best_acc=0.995698
884
+ now: 2022-04-05 14:50:39.132010
885
+ [88400] train_loss=0.013988 valid_loss=0.0252771 valid_pos_acc=0.975452 valid_acc=0.995494 / 0.928645 / 0.829601 best_acc=0.995698
886
+ now: 2022-04-05 15:02:36.068409
887
+ [88600] train_loss=0.0131047 valid_loss=0.0262127 valid_pos_acc=0.975194 valid_acc=0.995539 / 0.933965 / 0.833988 best_acc=0.995698
888
+ now: 2022-04-05 15:14:39.273565
889
+ [88800] train_loss=0.0134752 valid_loss=0.0262747 valid_pos_acc=0.975752 valid_acc=0.995544 / 0.93222 / 0.832838 best_acc=0.995698
890
+ now: 2022-04-05 15:26:47.762768
891
+ [89000] train_loss=0.0140879 valid_loss=0.026511 valid_pos_acc=0.975494 valid_acc=0.995559 / 0.930583 / 0.832717 best_acc=0.995698
892
+ now: 2022-04-05 15:38:34.298601
893
+ [89200] train_loss=0.0145324 valid_loss=0.0263811 valid_pos_acc=0.975348 valid_acc=0.995639 / 0.933692 / 0.831782 best_acc=0.995698
894
+ now: 2022-04-05 15:50:34.391187
895
+ [89400] train_loss=0.0143614 valid_loss=0.0266067 valid_pos_acc=0.975589 valid_acc=0.99558 / 0.93295 / 0.832933 best_acc=0.995698
896
+ now: 2022-04-05 16:02:28.811246
897
+ [89600] train_loss=0.0139432 valid_loss=0.025967 valid_pos_acc=0.975483 valid_acc=0.995678 / 0.930272 / 0.833364 best_acc=0.995698
898
+ now: 2022-04-05 16:14:16.963671
899
+ [89800] train_loss=0.015888 valid_loss=0.025742 valid_pos_acc=0.975517 valid_acc=0.995591 / 0.930457 / 0.833334 best_acc=0.995698
900
+ now: 2022-04-05 16:26:24.078615
901
+ [90000] train_loss=0.0144244 valid_loss=0.0256335 valid_pos_acc=0.975693 valid_acc=0.995632 / 0.93239 / 0.830571 best_acc=0.995698
902
+ now: 2022-04-05 16:38:39.760929
903
+ [90200] train_loss=0.0143907 valid_loss=0.0264985 valid_pos_acc=0.975496 valid_acc=0.995324 / 0.930264 / 0.822809 best_acc=0.995698
904
+ now: 2022-04-05 16:50:47.722632
905
+ [90400] train_loss=0.0137474 valid_loss=0.0254164 valid_pos_acc=0.975394 valid_acc=0.995574 / 0.93623 / 0.834253 best_acc=0.995698
906
+ now: 2022-04-05 17:02:52.333896
907
+ [90600] train_loss=0.0142556 valid_loss=0.0260749 valid_pos_acc=0.975272 valid_acc=0.995496 / 0.932219 / 0.832236 best_acc=0.995698
908
+ now: 2022-04-05 17:14:37.154862
909
+ [90800] train_loss=0.0138562 valid_loss=0.0259641 valid_pos_acc=0.975235 valid_acc=0.99537 / 0.92701 / 0.827433 best_acc=0.995698
910
+ now: 2022-04-05 17:26:38.096136
911
+ [91000] train_loss=0.0128578 valid_loss=0.0257195 valid_pos_acc=0.975418 valid_acc=0.995511 / 0.927246 / 0.829352 best_acc=0.995698
912
+ now: 2022-04-05 17:38:33.834741
913
+ [91200] train_loss=0.0139655 valid_loss=0.0263024 valid_pos_acc=0.975672 valid_acc=0.995513 / 0.931341 / 0.827522 best_acc=0.995698
914
+ now: 2022-04-05 17:50:27.641391
915
+ [91400] train_loss=0.0143926 valid_loss=0.0255845 valid_pos_acc=0.975272 valid_acc=0.995307 / 0.931624 / 0.830926 best_acc=0.995698
916
+ now: 2022-04-05 18:02:38.786522
917
+ [91600] train_loss=0.0145323 valid_loss=0.0253511 valid_pos_acc=0.975374 valid_acc=0.995478 / 0.93136 / 0.832302 best_acc=0.995698
918
+ now: 2022-04-05 18:14:33.259268
919
+ [91800] train_loss=0.0145714 valid_loss=0.0256718 valid_pos_acc=0.975031 valid_acc=0.995468 / 0.929101 / 0.830769 best_acc=0.995698
920
+ now: 2022-04-05 18:26:31.723280
921
+ [92000] train_loss=0.0150823 valid_loss=0.025818 valid_pos_acc=0.975392 valid_acc=0.995515 / 0.933814 / 0.836218 best_acc=0.995698
922
+ now: 2022-04-05 18:38:37.282828
923
+ [92200] train_loss=0.0144193 valid_loss=0.0249235 valid_pos_acc=0.975411 valid_acc=0.995576 / 0.930254 / 0.831995 best_acc=0.995698
924
+ now: 2022-04-05 18:50:39.861549
925
+ [92400] train_loss=0.0145149 valid_loss=0.0255686 valid_pos_acc=0.97557 valid_acc=0.995656 / 0.93331 / 0.837843 best_acc=0.995698
926
+ now: 2022-04-05 19:02:40.114394
927
+ [92600] train_loss=0.0136907 valid_loss=0.0251938 valid_pos_acc=0.975494 valid_acc=0.995682 / 0.934727 / 0.838754 best_acc=0.995698
928
+ now: 2022-04-05 19:14:38.262458
929
+ [92800] train_loss=0.0145302 valid_loss=0.0252417 valid_pos_acc=0.97527 valid_acc=0.995661 / 0.932038 / 0.837297 best_acc=0.995698
930
+ now: 2022-04-05 19:26:38.439473
931
+ [93000] train_loss=0.0156955 valid_loss=0.0263089 valid_pos_acc=0.975235 valid_acc=0.995504 / 0.933166 / 0.828832 best_acc=0.995698
932
+ now: 2022-04-05 19:38:54.778835
933
+ [93200] train_loss=0.0159822 valid_loss=0.0254963 valid_pos_acc=0.975159 valid_acc=0.995413 / 0.93672 / 0.840871 best_acc=0.995698
934
+ now: 2022-04-05 19:50:52.991181
935
+ [93400] train_loss=0.0137888 valid_loss=0.0254194 valid_pos_acc=0.975212 valid_acc=0.995617 / 0.933914 / 0.832067 best_acc=0.995698
936
+ now: 2022-04-05 20:03:09.308012
937
+ [93600] train_loss=0.0159022 valid_loss=0.0244989 valid_pos_acc=0.97522 valid_acc=0.995565 / 0.933225 / 0.836939 best_acc=0.995698
938
+ now: 2022-04-05 20:15:19.273222
939
+ [93800] train_loss=0.0142521 valid_loss=0.0248734 valid_pos_acc=0.975294 valid_acc=0.995643 / 0.93422 / 0.840297 best_acc=0.995698
940
+ now: 2022-04-05 20:27:27.195510
941
+ [94000] train_loss=0.0150861 valid_loss=0.0239483 valid_pos_acc=0.975533 valid_acc=0.995619 / 0.936434 / 0.84042 best_acc=0.995698
942
+ now: 2022-04-05 20:39:32.119195
943
+ [94200] train_loss=0.0152762 valid_loss=0.0248546 valid_pos_acc=0.975739 valid_acc=0.99565 / 0.932887 / 0.835955 best_acc=0.995698
944
+ now: 2022-04-05 20:51:35.782856
945
+ [94400] train_loss=0.0141213 valid_loss=0.0250847 valid_pos_acc=0.974971 valid_acc=0.995682 / 0.935789 / 0.840368 best_acc=0.995698
946
+ now: 2022-04-05 21:03:36.489728
947
+ [94600] train_loss=0.0144385 valid_loss=0.0249789 valid_pos_acc=0.975439 valid_acc=0.995544 / 0.938386 / 0.840405 best_acc=0.995698
948
+ now: 2022-04-05 21:15:30.678169
949
+ [94800] train_loss=0.013553 valid_loss=0.0256229 valid_pos_acc=0.975455 valid_acc=0.995611 / 0.93238 / 0.837687 best_acc=0.995698
950
+ now: 2022-04-05 21:27:32.536531
951
+ [95000] train_loss=0.0158608 valid_loss=0.0256513 valid_pos_acc=0.97537 valid_acc=0.995528 / 0.935879 / 0.844516 best_acc=0.995698
952
+ now: 2022-04-05 21:39:32.357321
953
+ [95200] train_loss=0.0151035 valid_loss=0.0254666 valid_pos_acc=0.97575 valid_acc=0.995669 / 0.934501 / 0.837649 best_acc=0.995698
954
+ now: 2022-04-05 21:51:43.227092
955
+ [95400] train_loss=0.0151553 valid_loss=0.0249784 valid_pos_acc=0.975474 valid_acc=0.995526 / 0.934746 / 0.835569 best_acc=0.995698
956
+ now: 2022-04-05 22:03:48.247576
957
+ [95600] train_loss=0.0155748 valid_loss=0.0260542 valid_pos_acc=0.975218 valid_acc=0.995552 / 0.933312 / 0.835163 best_acc=0.995698
958
+ now: 2022-04-05 22:15:38.086255
959
+ [95800] train_loss=0.0146772 valid_loss=0.0247799 valid_pos_acc=0.975637 valid_acc=0.995626 / 0.930346 / 0.833518 best_acc=0.995698
960
+ now: 2022-04-05 22:27:43.689507
961
+ [96000] train_loss=0.01602 valid_loss=0.0267345 valid_pos_acc=0.975233 valid_acc=0.995433 / 0.931918 / 0.829208 best_acc=0.995698
962
+ now: 2022-04-05 22:39:43.113855
963
+ [96200] train_loss=0.0151231 valid_loss=0.0255382 valid_pos_acc=0.9755 valid_acc=0.995478 / 0.926487 / 0.826398 best_acc=0.995698
964
+ now: 2022-04-05 22:51:31.072458
965
+ [96400] train_loss=0.0149376 valid_loss=0.0258116 valid_pos_acc=0.97491 valid_acc=0.995498 / 0.928117 / 0.831135 best_acc=0.995698
966
+ now: 2022-04-05 23:03:31.843003
967
+ [96600] train_loss=0.0144005 valid_loss=0.0254002 valid_pos_acc=0.975784 valid_acc=0.99558 / 0.930447 / 0.829763 best_acc=0.995698
968
+ now: 2022-04-05 23:15:34.625356
969
+ [96800] train_loss=0.0146142 valid_loss=0.0268592 valid_pos_acc=0.974348 valid_acc=0.995389 / 0.930559 / 0.830846 best_acc=0.995698
970
+ now: 2022-04-05 23:27:30.008875
971
+ [97000] train_loss=0.0142077 valid_loss=0.025247 valid_pos_acc=0.975439 valid_acc=0.995654 / 0.92956 / 0.835155 best_acc=0.995698
972
+ now: 2022-04-05 23:39:35.358671
973
+ [97200] train_loss=0.0136525 valid_loss=0.025149 valid_pos_acc=0.975429 valid_acc=0.995604 / 0.932177 / 0.838884 best_acc=0.995698
974
+ now: 2022-04-05 23:51:42.518252
975
+ [97400] train_loss=0.0142976 valid_loss=0.0252632 valid_pos_acc=0.975637 valid_acc=0.99558 / 0.928696 / 0.831515 best_acc=0.995698
976
+ now: 2022-04-06 00:03:39.610613
977
+ [97600] train_loss=0.01445 valid_loss=0.0247523 valid_pos_acc=0.975652 valid_acc=0.995626 / 0.928967 / 0.837404 best_acc=0.995698
978
+ now: 2022-04-06 00:15:39.948279
979
+ [97800] train_loss=0.0138557 valid_loss=0.0250991 valid_pos_acc=0.975672 valid_acc=0.99553 / 0.928176 / 0.832755 best_acc=0.995698
980
+ now: 2022-04-06 00:27:35.432314
981
+ [98000] train_loss=0.015877 valid_loss=0.0247035 valid_pos_acc=0.975576 valid_acc=0.995548 / 0.928882 / 0.835904 best_acc=0.995698
982
+ now: 2022-04-06 00:39:42.196877
983
+ [98200] train_loss=0.0141082 valid_loss=0.0244627 valid_pos_acc=0.975843 valid_acc=0.995676 / 0.928219 / 0.833121 best_acc=0.995698
984
+ now: 2022-04-06 00:51:45.295429
985
+ [98400] train_loss=0.0146779 valid_loss=0.0248106 valid_pos_acc=0.975962 valid_acc=0.995632 / 0.927196 / 0.83131 best_acc=0.995698
986
+ now: 2022-04-06 01:03:48.072286
987
+ [98600] train_loss=0.0150793 valid_loss=0.024931 valid_pos_acc=0.975741 valid_acc=0.995587 / 0.927742 / 0.832345 best_acc=0.995698
988
+ now: 2022-04-06 01:15:49.874914
989
+ [98800] train_loss=0.0138506 valid_loss=0.0256198 valid_pos_acc=0.976136 valid_acc=0.995611 / 0.929698 / 0.830346 best_acc=0.995698
990
+ now: 2022-04-06 01:27:46.508890
991
+ [99000] train_loss=0.0118288 valid_loss=0.026124 valid_pos_acc=0.976116 valid_acc=0.995596 / 0.930481 / 0.833991 best_acc=0.995698
992
+ now: 2022-04-06 01:39:52.754135
993
+ [99200] train_loss=0.0115159 valid_loss=0.0254321 valid_pos_acc=0.975875 valid_acc=0.995689 / 0.93085 / 0.835924 best_acc=0.995698
994
+ now: 2022-04-06 01:51:59.498058
995
+ [99400] train_loss=0.0118415 valid_loss=0.0261764 valid_pos_acc=0.976184 valid_acc=0.995604 / 0.930402 / 0.832654 best_acc=0.995698
996
+ now: 2022-04-06 02:04:00.516174
997
+ [99600] train_loss=0.0111765 valid_loss=0.0253784 valid_pos_acc=0.976212 valid_acc=0.995789 / 0.930556 / 0.834549 best_acc=0.995789
998
+ now: 2022-04-06 02:16:10.608263
999
+ [99800] train_loss=0.0122449 valid_loss=0.0257746 valid_pos_acc=0.976166 valid_acc=0.995726 / 0.931839 / 0.838174 best_acc=0.995789
1000
+ now: 2022-04-06 02:28:12.819393
1001
+ [100000] train_loss=0.0122801 valid_loss=0.0260771 valid_pos_acc=0.975372 valid_acc=0.99555 / 0.929529 / 0.838038 best_acc=0.995789
1002
+ now: 2022-04-06 02:40:11.627022
1003
+ testing ...
1004
+ reloading best accuracy model ...
1005
+ valid_best_acc=0.995789 test_loss=0.0266495 test_pos_acc=0.975942 test_acc=0.995605 / 0.918997 / 0.806213
GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/version ADDED
@@ -0,0 +1 @@
 
 
1
+ v2.0
LICENSE ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ MIT License
2
+
3
+ Copyright (c) 2026 JodieRuth
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy
6
+ of this software and associated documentation files (the "Software"), to deal
7
+ in the Software without restriction, including without limitation the rights
8
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
9
+ copies of the Software, and to permit persons to whom the Software is
10
+ furnished to do so, subject to the following conditions:
11
+
12
+ The above copyright notice and this permission notice shall be included in all
13
+ copies or substantial portions of the Software.
14
+
15
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21
+ SOFTWARE.
README.md ADDED
@@ -0,0 +1,220 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - zh
4
+ license: other
5
+ license_name: mixed
6
+ license_link: https://huggingface.co/shiqianyan/TangYuGe-GPT-SoVITS-v2ProPlus/blob/main/THIRD_PARTY.md
7
+ library_name: gpt-sovits
8
+ pipeline_tag: text-to-speech
9
+ base_model: lj1995/GPT-SoVITS
10
+ tags:
11
+ - text-to-speech
12
+ - tts
13
+ - voice-cloning
14
+ - gpt-sovits
15
+ - gpt-sovits-v2proplus
16
+ - chinese
17
+ - anime
18
+ - galgame
19
+ - character-voice
20
+ ---
21
+
22
+ # TangYuGe GPT-SoVITS v2ProPlus(唐语歌语音合成模型)
23
+
24
+ **English summary.** A Chinese text-to-speech model of the character *Tang Yuge (唐语歌)* from the visual novel *Sparkling Memories (夏日花火)*, fine-tuned from the official [GPT-SoVITS](https://github.com/RVC-Boss/GPT-SoVITS) **v2ProPlus** base models. This repository is self-contained: it ships the fine-tuned GPT and SoVITS weights, all pretrained models required by GPT-SoVITS at inference time, 19 scene reference clips with their prompt texts, ready-to-use profiles and scripts. Clone GPT-SoVITS at the verified commit, install its Python dependencies, run `scripts/install_pretrained.py`, and you can synthesize speech. Chinese only.
25
+
26
+ 本仓库是基于 GPT-SoVITS **v2ProPlus** 底模微调的唐语歌(《夏日花火》角色)中文语音合成模型,并附带推理所需的全部预训练模型。下载本仓库、准备好 GPT-SoVITS 运行环境后即可直接使用。
27
+
28
+ 旧版本(GPT-SoVITS v2Pro 微调)见 [shiqianyan/TangYuGe-GPT-SoVITS-v2pro](https://huggingface.co/shiqianyan/TangYuGe-GPT-SoVITS-v2pro)。本版本在相同训练数据上改用 v2ProPlus 底模,背景底噪和电音明显更少,推荐优先使用本版本。
29
+
30
+ ## 模型信息
31
+
32
+ | 项目 | 内容 |
33
+ | --- | --- |
34
+ | 模型类型 | GPT-SoVITS 少样本语音克隆 / 文本转语音(TTS) |
35
+ | 底模 | GPT-SoVITS v2ProPlus(`s2Gv2ProPlus.pth`、`s2Dv2ProPlus.pth`)与 GPT 底模 `s1v3.ckpt` |
36
+ | 微调方式 | 全参数微调(非 LoRA),SoVITS 与 GPT 两个模型分别训练 |
37
+ | 语言 | 中文(普通话) |
38
+ | 输出 | 32 kHz 单声道 WAV |
39
+ | 训练数据 | 角色唐语歌的游戏语音 561 条,约 43 分钟 |
40
+ | 推荐采样参数 | `top_k=5`、`top_p=0.8`、`temperature=0.7`,切句方式 `cut5` |
41
+ | 验证环境 | GPT-SoVITS 提交 `ea2d2a8`、Python 3.10、PyTorch 2.4.0 + CUDA 12.4、NVIDIA RTX 4060 Laptop 8GB |
42
+
43
+ ## 适用范围
44
+
45
+ - 适合:为唐语歌这一角色生成中文台词语音,例如同人创作、游戏 Mod、角色对话机器人、个人学习研究。
46
+ - 不适合:非中文文本;需要与原声完全一致的商业配音;任何冒充真实人物、误导他人或违法的用途。生成内容请标注为 AI 合成。
47
+
48
+ ## 快速开始
49
+
50
+ ### 1. 准备 GPT-SoVITS
51
+
52
+ ```bash
53
+ git clone https://github.com/RVC-Boss/GPT-SoVITS.git
54
+ cd GPT-SoVITS
55
+ git checkout ea2d2a8
56
+ ```
57
+
58
+ 按 GPT-SoVITS 官方说明安装 Python 依赖(Python 3.10,推荐 NVIDIA GPU)。本仓库已附带所需预训练模型,不需要再单独下载 GPT-SoVITS 的预训练模型包。
59
+
60
+ ### 2. 下载本仓库
61
+
62
+ 建议与 `GPT-SoVITS` 放在同一个父目录下:
63
+
64
+ ```bash
65
+ cd ..
66
+ hf download shiqianyan/TangYuGe-GPT-SoVITS-v2ProPlus --local-dir TangYuGe-GPT-SoVITS-v2ProPlus
67
+ ```
68
+
69
+ ```text
70
+ <workspace>/
71
+ ├── GPT-SoVITS/
72
+ └── TangYuGe-GPT-SoVITS-v2ProPlus/
73
+ ```
74
+
75
+ ### 3. 安装预训练模型
76
+
77
+ ```bash
78
+ python TangYuGe-GPT-SoVITS-v2ProPlus/scripts/install_pretrained.py --gptsovits-root GPT-SoVITS
79
+ ```
80
+
81
+ 脚本会校验文件哈希,再把 `GPT-SoVITS-overlay/` 中的模型复制到 GPT-SoVITS 对应位置(已存在且相同的文件会跳过)。
82
+
83
+ ### 4. 生成语音
84
+
85
+ 以下命令均使用 GPT-SoVITS 的 Python 环境执行:
86
+
87
+ ```bash
88
+ # 默认使用主 profile(严肃、长句)
89
+ python TangYuGe-GPT-SoVITS-v2ProPlus/scripts/run_tangyuge_infer.py --text "这里填写要合成的中文文本。" --output output.wav
90
+
91
+ # 切换场景:换一个 profile
92
+ python TangYuGe-GPT-SoVITS-v2ProPlus/scripts/run_tangyuge_infer.py --voice-profile TangYuGe-GPT-SoVITS-v2ProPlus/profiles/morning.yaml --text "早上好呀,今天也要加油哦。" --output morning.wav
93
+ ```
94
+
95
+ 若两个目录不在同一父目录下,追加 `--gptsovits-root <GPT-SoVITS 路径>`。
96
+
97
+ ### 5. 启动 HTTP 服务(可选)
98
+
99
+ ```bash
100
+ python TangYuGe-GPT-SoVITS-v2ProPlus/scripts/start_tangyuge_bridge.py
101
+ ```
102
+
103
+ 启动后在 `127.0.0.1:9881` 提供 `/tts` 接口(后端 GPT-SoVITS `api_v2` 运行在 `9880`)。服务会根据文本自动选择场景,也可显式指定:
104
+
105
+ ```bash
106
+ curl -X POST http://127.0.0.1:9881/tts -H "Content-Type: application/json; charset=utf-8" \
107
+ -d '{"text": "早上好呀,今天也要加油哦。", "profile": "morning"}' --output output.wav
108
+ ```
109
+
110
+ 请求字段:`text`(必填)、`profile` / `scene` / `reference`(选择场景)、`top_k`、`top_p`、`temperature` 等。辅助接口:`GET /health`、`GET /profiles`、`GET /references`。
111
+
112
+ ### 在 GPT-SoVITS WebUI 中使用
113
+
114
+ 安装预训练模型后,在 WebUI 推理页选择:
115
+
116
+ - GPT 模型:`weights/tangyuge_v2pp-e14.ckpt`
117
+ - SoVITS 模型:`weights/tangyuge_v2pp_e4_s564.pth`
118
+ - 参考音频:`references/` 中任一文件,**参考文本必须使用 `manifests/references.yaml` 中与之绑定的 `prompt_text`**
119
+ - 采样参数:`top_k=5`、`top_p=0.8`、`temperature=0.7`
120
+
121
+ ## 场景参考音频(reference)
122
+
123
+ GPT-SoVITS 的语气和节奏很大程度上由参考音频决定。本仓库提供 19 个场景,每个场景是一条原游戏语音及其台词文本,打包为 `profiles/*.yaml`:
124
+
125
+ | Profile | 适用文本 |
126
+ | --- | --- |
127
+ | `main_hybrid_mid`(默认) | 严肃说明、长台词、情绪稳定的表达 |
128
+ | `morning` | 早安、问候、短句寒暄 |
129
+ | `neutral_polite` | 普通说明、礼貌回答 |
130
+ | `soft_daily` | 温柔、日常、低压交流 |
131
+ | `question_clear` / `question_firm` | 关切式疑问 / 质询式疑问 |
132
+ | `lively_daily` / `casual_invite` | 轻快短句 / 熟人邀请 |
133
+ | `narration_calm` | 旁白、叙述 |
134
+ | `apology` | 道歉、礼貌退让 |
135
+ | `bright_fireworks` / `high_brightness_exclaim` / `exclaim` | 明亮期待 / 开心短句 / 惊讶感叹 |
136
+ | `intimate_reassure` / `romantic_soft` | 亲密安抚 / 含蓄恋爱向 |
137
+ | `command_gentle` | 提醒、劝阻 |
138
+ | `food_daily` / `cafeteria_ordering` | 饮食日常 / 点单 |
139
+ | `low_energy_discussion` | 克制解释、舒缓讨论 |
140
+
141
+ 更换参考音频时必须同时更换对应的参考文本,否则发音与节奏会明显变差。
142
+
143
+ ## 仓库结构
144
+
145
+ ```text
146
+ ├── README.md
147
+ ├── LICENSE # 本仓库作者整理内容的许可(MIT)
148
+ ├── THIRD_PARTY.md # 各组件的来源与许可
149
+ ├── weights/ # 微调权重(fp16)
150
+ │ ├── tangyuge_v2pp-e14.ckpt # GPT(文本→语义)
151
+ │ └── tangyuge_v2pp_e4_s564.pth # SoVITS(语义→声学)
152
+ ├── GPT-SoVITS-overlay/ # 推理与再训练所需的预训练模型,与 GPT-SoVITS 目录同构
153
+ │ └── GPT_SoVITS/
154
+ │ ├── pretrained_models/ # 底模、BERT、HuBERT、声纹模型、语种识别模型
155
+ │ └── text/G2PWModel/ # 中文多音字模型
156
+ ├── configs/tts_infer_pack.yaml # GPT-SoVITS 推理配置
157
+ ├── profiles/ # 19 个场景 profile
158
+ ├── references/ # 19 条场景参考音频
159
+ ├── samples/ # 合成示例
160
+ ├── scripts/ # 安装、推理、HTTP 服务脚本
161
+ └── manifests/
162
+ ├── references.yaml # 参考音频与参考文本的绑定清单
163
+ ├── training_report.md # 训练与评测报告
164
+ └── checksums.sha256 # 文件校验清单
165
+ ```
166
+
167
+ ## 训练细节
168
+
169
+ - 数据:从游戏中提取的唐语歌语音 702 条(约 48.8 分钟),筛选 1–12 秒、有效汉字不少于 6 个的 561 条(约 43.1 分钟)。用语音识别逐条核对过台词文本,未发现标注错误。
170
+ - SoVITS:batch size 4,学习率 1e-4,fp16 混合精度,共训练 8 个 epoch,选用第 4 个 epoch(564 步)。
171
+ - GPT:batch size 4,学习率 0.002,16-bit 混合精度,共训练 15 个 epoch,选用第 14 个 epoch。
172
+ - 选型:在 25 句压力测试集、4 条参考音频、固定随机种子下,对多个 epoch 组合与采样参数做客观评测,再经两轮人工盲听确定最终组合。完整过程见 `manifests/training_report.md`。
173
+
174
+ ## 评测
175
+
176
+ 与旧版 v2Pro 微调模型在相同文本、相同参考音频、相同采样参数下对比(配对 bootstrap,95% 置信区间不含 0 视为显著):
177
+
178
+ | 指标 | 结果 |
179
+ | --- | --- |
180
+ | 停顿处底噪电平 | 降低约 5–6 dB(显著),接近原声水平 |
181
+ | DNSMOS 背景噪声分(BAK) | +0.21(显著) |
182
+ | DNSMOS 整体质量分(OVRL) | +0.12(显著) |
183
+ | 音高细碎抖动 | 降低约 9%(显著) |
184
+ | 发音准确率(ASR 拼音错误率) | 与旧版无显著差异,约 1–2% |
185
+ | 声纹相似度 | 与旧版无显著差异 |
186
+
187
+ ## 局限性
188
+
189
+ - 仅支持中文;中英混合文本效果未验证。
190
+ - 极短的句子(如“好呀,就这样吧”)偶尔会只读出前半句,这是 GPT-SoVITS 自回归生成的共性问题;保守的采样参数可以减少但不能消除。
191
+ - 效果依赖参考音频与参考文本的配对;参考文本写错会导致发音、节奏异常。
192
+ - 音色与语气受限于约 43 分钟的训练数据,强烈情绪(大喊、哭泣等)表现有限。
193
+ - 在 CPU 上推理时,部分 Intel 混合架构处理器(已在 i7-13650HX 上复现)的 PyTorch oneDNN 多线程卷积会算出错误结果;GPU 推理不受影响。CPU 推理请设置 `torch.backends.mkldnn.enabled = False` 或使用单线程。
194
+
195
+ ## 许可证与来源
196
+
197
+ 本仓库包含多个来源的组件,各自遵循其许可证,详见 [THIRD_PARTY.md](THIRD_PARTY.md):
198
+
199
+ - 微调权重、配置、profile、脚本与文档:MIT(见 `LICENSE`)。
200
+ - 训练语音与参考音频:来源于《Sparkling Memories》(夏日花火)角色唐语歌的游戏语音。据发布者说明,该部分内容按 Apache-2.0 条款使用与发布。
201
+ - GPT-SoVITS 底模:MIT;chinese-roberta-wwm-ext-large:Apache-2.0;chinese-hubert-base:MIT;ERes2NetV2 声纹模型(3D-Speaker):Apache-2.0;G2PW(g2pW):Apache-2.0;fastText 语种识别模型 `lid.176.bin`:CC BY-SA 3.0。
202
+
203
+ 再分发时请保留 `LICENSE` 与 `THIRD_PARTY.md`。
204
+
205
+ ## 致谢
206
+
207
+ - [GPT-SoVITS](https://github.com/RVC-Boss/GPT-SoVITS)
208
+ - [HFL Chinese-BERT-wwm](https://github.com/ymcui/Chinese-BERT-wwm)、[TencentGameMate chinese_speech_pretrain](https://github.com/TencentGameMate/chinese_speech_pretrain)
209
+ - [3D-Speaker](https://github.com/modelscope/3D-Speaker)、[g2pW](https://github.com/GitYCC/g2pW)、[fastText](https://fasttext.cc/)
210
+
211
+ ## 引用
212
+
213
+ ```bibtex
214
+ @misc{tangyuge_gptsovits_v2proplus,
215
+ title = {TangYuGe GPT-SoVITS v2ProPlus},
216
+ author = {shiqianyan},
217
+ year = {2026},
218
+ howpublished = {\url{https://huggingface.co/shiqianyan/TangYuGe-GPT-SoVITS-v2ProPlus}}
219
+ }
220
+ ```
THIRD_PARTY.md ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Third-Party Components and Data Provenance
2
+
3
+ 本仓库为了“下载即可使用”,随包分发了 GPT-SoVITS 推理与再训练所需的预训练模型。各组件保持其原始许可证;再分发本仓库或其中任一组件时,请同时保留本文件与 `LICENSE`。
4
+
5
+ ## 1. 本仓库作者的内容(MIT)
6
+
7
+ 微调权重(`weights/`)、配置(`configs/`)、场景 profile(`profiles/`)、清单(`manifests/`)、脚本(`scripts/`)、示例音频(`samples/`)与文档,按 `LICENSE`(MIT)发布。
8
+
9
+ ## 2. 训练语音与参考音频
10
+
11
+ - 来源:《Sparkling Memories》(夏日花火)角色“唐语歌”的游戏语音。
12
+ - 用途:微调训练;`references/` 中的 19 条原始语音用作推理时的参考音频。
13
+ - 许可:据发布者说明,该部分内容按 Apache-2.0 条款使用与发布。
14
+
15
+ ## 3. 随包分发的预训练模型(`GPT-SoVITS-overlay/`)
16
+
17
+ | 文件 / 目录(相对 GPT-SoVITS 根目录) | 组件 | 来源 | 许可证 |
18
+ | --- | --- | --- | --- |
19
+ | `GPT_SoVITS/pretrained_models/v2Pro/s2Gv2ProPlus.pth` | SoVITS v2ProPlus 生成器底模 | [lj1995/GPT-SoVITS](https://huggingface.co/lj1995/GPT-SoVITS) | MIT |
20
+ | `GPT_SoVITS/pretrained_models/v2Pro/s2Dv2ProPlus.pth` | SoVITS v2ProPlus 判别器底模(仅训练) | [lj1995/GPT-SoVITS](https://huggingface.co/lj1995/GPT-SoVITS) | MIT |
21
+ | `GPT_SoVITS/pretrained_models/s1v3.ckpt` | GPT 底模 | [lj1995/GPT-SoVITS](https://huggingface.co/lj1995/GPT-SoVITS) | MIT |
22
+ | `GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large/` | 中文 BERT(文本特征) | [hfl/chinese-roberta-wwm-ext-large](https://huggingface.co/hfl/chinese-roberta-wwm-ext-large) | Apache-2.0 |
23
+ | `GPT_SoVITS/pretrained_models/chinese-hubert-base/` | 中文 HuBERT(语音特征) | [TencentGameMate/chinese-hubert-base](https://huggingface.co/TencentGameMate/chinese-hubert-base) | MIT |
24
+ | `GPT_SoVITS/pretrained_models/sv/pretrained_eres2netv2w24s4ep4.ckpt` | ERes2NetV2 声纹模型(v2Pro / v2ProPlus 必需) | [3D-Speaker](https://github.com/modelscope/3D-Speaker),经 [lj1995/GPT-SoVITS](https://huggingface.co/lj1995/GPT-SoVITS) 分发 | Apache-2.0 |
25
+ | `GPT_SoVITS/text/G2PWModel/` | G2PW 中文多音字模型 | [g2pW](https://github.com/GitYCC/g2pW),经 GPT-SoVITS 分发 | Apache-2.0 |
26
+ | `GPT_SoVITS/pretrained_models/fast_langdetect/lid.176.bin` | fastText 语种识别模型 | [fastText language identification](https://fasttext.cc/docs/en/language-identification.html)(Facebook AI Research) | CC BY-SA 3.0 |
27
+
28
+ 说明:
29
+
30
+ - GPT-SoVITS 底模由 [RVC-Boss/GPT-SoVITS](https://github.com/RVC-Boss/GPT-SoVITS) 项目发布于 [lj1995/GPT-SoVITS](https://huggingface.co/lj1995/GPT-SoVITS)(MIT)。本仓库中的文件与官方文件逐字节相同,SHA-256 见 `manifests/checksums.sha256`。
31
+ - 推理只需要 BERT、HuBERT、声纹模型、G2PW 与语种识别模型;GPT / SoVITS 底模仅在基于本模型继续训练时使用。
32
+ - `lid.176.bin` 采用 CC BY-SA 3.0:再分发须署名,修改后的版本须以相同许可发布。
33
+
34
+ ## 4. 运行所需但不随包分发
35
+
36
+ - GPT-SoVITS 程序代码:[RVC-Boss/GPT-SoVITS](https://github.com/RVC-Boss/GPT-SoVITS)(MIT),本仓库在提交 `ea2d2a8` 上验证。
37
+ - Python 依赖:按 GPT-SoVITS 官方说明安装,遵循各依赖自身的许可证。
configs/tts_infer_pack.yaml ADDED
@@ -0,0 +1,57 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # GPT-SoVITS 推理配置。相对路径以 GPT-SoVITS 根目录为工作目录;权重由 profiles/*.yaml 覆盖为本仓库 weights/。
2
+ custom:
3
+ bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
4
+ cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
5
+ device: cuda
6
+ is_half: true
7
+ t2s_weights_path: ../weights/tangyuge_v2pp-e14.ckpt
8
+ version: v2ProPlus
9
+ vits_weights_path: ../weights/tangyuge_v2pp_e4_s564.pth
10
+ v1:
11
+ bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
12
+ cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
13
+ device: cpu
14
+ is_half: false
15
+ t2s_weights_path: GPT_SoVITS/pretrained_models/s1bert25hz-2kh-longer-epoch=68e-step=50232.ckpt
16
+ version: v1
17
+ vits_weights_path: GPT_SoVITS/pretrained_models/s2G488k.pth
18
+ v2:
19
+ bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
20
+ cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
21
+ device: cpu
22
+ is_half: false
23
+ t2s_weights_path: GPT_SoVITS/pretrained_models/gsv-v2final-pretrained/s1bert25hz-5kh-longer-epoch=12-step=369668.ckpt
24
+ version: v2
25
+ vits_weights_path: GPT_SoVITS/pretrained_models/gsv-v2final-pretrained/s2G2333k.pth
26
+ v2Pro:
27
+ bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
28
+ cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
29
+ device: cpu
30
+ is_half: false
31
+ t2s_weights_path: GPT_SoVITS/pretrained_models/s1v3.ckpt
32
+ version: v2Pro
33
+ vits_weights_path: GPT_SoVITS/pretrained_models/v2Pro/s2Gv2Pro.pth
34
+ v2ProPlus:
35
+ bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
36
+ cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
37
+ device: cpu
38
+ is_half: false
39
+ t2s_weights_path: GPT_SoVITS/pretrained_models/s1v3.ckpt
40
+ version: v2ProPlus
41
+ vits_weights_path: GPT_SoVITS/pretrained_models/v2Pro/s2Gv2ProPlus.pth
42
+ v3:
43
+ bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
44
+ cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
45
+ device: cpu
46
+ is_half: false
47
+ t2s_weights_path: GPT_SoVITS/pretrained_models/s1v3.ckpt
48
+ version: v3
49
+ vits_weights_path: GPT_SoVITS/pretrained_models/s2Gv3.pth
50
+ v4:
51
+ bert_base_path: GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large
52
+ cnhuhbert_base_path: GPT_SoVITS/pretrained_models/chinese-hubert-base
53
+ device: cpu
54
+ is_half: false
55
+ t2s_weights_path: GPT_SoVITS/pretrained_models/s1v3.ckpt
56
+ version: v4
57
+ vits_weights_path: GPT_SoVITS/pretrained_models/gsv-v4-pretrained/s2Gv4.pth
manifests/checksums.sha256 ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ae05a32dee4b7a4616f3c08f48afc3b4a868226418bfcb4b3c9ebc232cf22fa7 .gitattributes
2
+ 69cdb94d6cc99c38f3f0be04445460a541e3f929b02463cd887d943a9b8e9f44 configs/tts_infer_pack.yaml
3
+ c3e5060a1277e0f078cc6be9da4528a605dba6ece93018981fe2c820e5c7b103 GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-hubert-base/config.json
4
+ dcd684124d06722947939d41ea6ae58dbf10968c60a11a29f23ddc602c64a29b GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-hubert-base/preprocessor_config.json
5
+ 24164f129c66499d1346e2aa55f183250c223161ec2770c0da3d3b08cf432d3c GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-hubert-base/pytorch_model.bin
6
+ 3d57de2fd7e80d0e5c8ff194f0bbb6baa10df7e43fc262a0cc71298a78b0a3e5 GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large/config.json
7
+ e53a693acc59ace251d143d068096ae0d7b79e4b1b503fa84c9dcf576448c1d8 GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large/pytorch_model.bin
8
+ 173796956820ea27bd14f76bf28162607ff4254807e2948253eb5b46f5bb643b GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/chinese-roberta-wwm-ext-large/tokenizer.json
9
+ 7e69ec5451bc261cc7844e49e4792a85d7f09c06789ec800fc4a44aec362764e GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/fast_langdetect/lid.176.bin
10
+ 87133414860ea14ff6620c483a3db5ed07b44be42e2c3fcdad65523a729a745a GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/s1v3.ckpt
11
+ 4f5a0bf73c61eb41b174e1bb54e7ee3c83233892be8e0af1f187024e8e581a35 GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/sv/pretrained_eres2netv2w24s4ep4.ckpt
12
+ 635cd84bf6f7f9b8d41c88c7106f81d782c794c61f931845214ea037b0c5bef2 GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/v2Pro/s2Dv2ProPlus.pth
13
+ d42a22bbbf65fb2bbdd45ad6a66841156977db45c7aabe0a6992ff378d9c7d3b GPT-SoVITS-overlay/GPT_SoVITS/pretrained_models/v2Pro/s2Gv2ProPlus.pth
14
+ 0fe90e26be7653023772b3ad2d08952961bbf295fb39ba685c7172d3d9769f52 GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/bopomofo_to_pinyin_wo_tune_dict.json
15
+ 4832fb6a553a9c5ee0c18b5cb698ce0acc312f324ee77ee61c340266d228184d GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/char_bopomofo_dict.json
16
+ b6154e494355d14a8d2c8a38d07720f77b34479e26aab21a5f63495ae904da76 GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/config.py
17
+ 2eb3c71fd95117b2e1abef8d2d0cd78aae894bbe7f0fac105ddc9c32ce63cbd0 GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/g2pW.onnx
18
+ fe7ed53098648441a94f414ed5fd3395789d1dceb98dc1e584489c2d43036b69 GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/MONOPHONIC_CHARS.txt
19
+ cc7e5a3eb377e3289af5e3a6be7505ac11a0d21422f80cfbf161059798f8671b GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/POLYPHONIC_CHARS.txt
20
+ d56fd692e4fd0090b3e11741f10238ea69bf4b380c0d65894646941e1231aed4 GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/record.log
21
+ cf063c2ce8ac1403d4dca358f0568dfd32684e7a4412e24bc6e7cd1a7cd0cefb GPT-SoVITS-overlay/GPT_SoVITS/text/G2PWModel/version
22
+ aecef7eea19798acfbac05fdea7017e3a76c8ce46c422ece4d9e9223f295bdd7 LICENSE
23
+ f5e42a565108b1bdb14a7d2be18057a8e53a6022bbe10562383e8cd7d806acd6 manifests/references.yaml
24
+ 94d33682c273af657050d3fddf462638a8f9080286830b9d00947a138e5254b0 manifests/training_report.md
25
+ d5bbf02da4b0686890d2e2ffc7fb9edad576f1383d9ce51b8dee9e23406b4bf1 profiles/apology.yaml
26
+ 5088cc31f30e9b85649d16d2e6e714cbc1ec1c4b9296c32f3a4b6df194298568 profiles/bright_fireworks.yaml
27
+ 14d6cbf0244571b62d327126bb1ea7fb4a0d1311de3afe9ff82866dec1abd3b5 profiles/cafeteria_ordering.yaml
28
+ 45367b01ea8bf74918e0e6ec430cf619b148a53790f3427ed928cb7ac58863c3 profiles/casual_invite.yaml
29
+ fe35700b132fd8b4cdff03f26730be7792169b026432c6d7a0229bab1ae751d6 profiles/command_gentle.yaml
30
+ 051f608a451edca0325daef8be31c9313bda8f55b66fa52b973dce6255a13d98 profiles/exclaim.yaml
31
+ 97b520cb5dc9db0f9f04c53508b0a4e837cab14d8ba9767d25eabacf475c0a74 profiles/food_daily.yaml
32
+ 14e26f059d8ddc20f67584673b4f2cc801c04119e231c39a33d77c092fde8982 profiles/high_brightness_exclaim.yaml
33
+ b4a3875c50700d0a70e1c0800b8bcefead1ac1add266c2d89c3635f562dc95af profiles/intimate_reassure.yaml
34
+ 89dbb99c1ef879d80ab2403871590f09cf2960fc4ce3ed35504d39e11130ab64 profiles/lively_daily.yaml
35
+ ac07309018614a2a3d52b357ef783f1193214dee7506cf0530fd4dd673a314c2 profiles/low_energy_discussion.yaml
36
+ d00c2d926458189fd4dba9e636b62a49865c0bfa641e44d42c70f604df13f49a profiles/main_hybrid_mid.yaml
37
+ 0d0d1e2a1526e13b6b1ad4ee4c82a7866c6de37dce35646ca7359638b590ea45 profiles/morning.yaml
38
+ 5c92e9e62c2b1711b0685fa03aaf9588c3a0e1b23c9498691f9d96297a48bf15 profiles/narration_calm.yaml
39
+ e0632f4c410bb7c28e49faf0db7a61cf70383b2fbfd4da29819a02d7c483928d profiles/neutral_polite.yaml
40
+ 38040a2cc47453a784b0d75c29b9af3d11b115b86ac71fcbebd340a692b80e05 profiles/question_clear.yaml
41
+ ad34f34d6e4cbb9bc7cbf1905de0594327c1b087809aba9135066228949dd947 profiles/question_firm.yaml
42
+ f9e287689dfc4ff495aac2b6ae975c89bf94d61491bd08936d5a76629711279e profiles/romantic_soft.yaml
43
+ 9cfad5379390b997c1254fc52b41228a98181e50f0a7dc0d5b3dc29bc697bb4e profiles/soft_daily.yaml
44
+ 315c0f709139f3c9e2ca552b2e60c41c0f613990d0b72ece35c49b74b3a1f97f README.md
45
+ b7b74e3b4353bff0b7d69128ea879607acba64cff4b1185927c2b2db83c120f0 references/apology_script32_L0040.wav
46
+ dc043035484e6253380e8262bd8630370d57057817e24b0121bcb79abee6b393 references/bright_fireworks_script22_L0948.wav
47
+ 57e41eb33ff65c4ee48792a852ad8b3dec0d433539859f73357bd0f822345801 references/cafeteria_ordering_script1_L0930.wav
48
+ 0f452a82cf2c831ea136d43dee3d12d875e37dad30106ec60cbead18d927a438 references/casual_invite_script21_L0249.wav
49
+ a84e77da40c75bc82ee3869f32e53a26c4024c38b1f270b6cb636e9bfb43fdcb references/command_gentle_script43_L0650.wav
50
+ 118c4532438106aaac5f897259e5289f7d4fcc7893467e24dad5193c5204b489 references/exclaim_script32_L0066.wav
51
+ fd08fd812a70f6c5021d3fa3eb8f893d45c72a94efb0dbdfcd32db01993ff96c references/food_daily_script43_L0998.wav
52
+ c762b74072321d7b2da574747a10d54884f0efe75d8101e70e92b7e93f692649 references/high_brightness_exclaim_script31_L0541.wav
53
+ 25a2048645310916476c8ba38df130d2eb79a0bdd68ed5279dd68ad467d3e1d5 references/intimate_reassure_script41_L0688.wav
54
+ 1ecf29b1d98c22bddbc245518ecb14e5f53ef56074d8d70540316ad58e4e265b references/lively_daily_script1_L0960.wav
55
+ 0cf7bb4f0ec0f05ba24804f9d6b61d8804185d92b3e26e1d1cf73e7888947d5a references/low_energy_discussion_script22_L0493.wav
56
+ 9a56987fa564a0b95632b8047daab6408685b788c213e0621734cec752dcdc51 references/morning_script43_L0993.wav
57
+ 2f0d79edd1e13acbf390b34f672b53dffd501535e299ed915d62c8998797be37 references/narration_calm_script1_L0869.wav
58
+ 7ff2353ae358311ab6032100502c25678d40cf2602683de04342921bbebee425 references/neutral_polite_script1_L0824.wav
59
+ af91b24dfce44ca19953bb3f43141b13461dc035de34252bcde3287fc1f745e1 references/question_clear_script34_L0420.wav
60
+ 62ba101d8be2c8906a9b43cae50d58e21c30dfd15de4d301e3c64d8cde55cdea references/question_firm_script33_L0206.wav
61
+ f820a73d5c905cd1a7a1fae26eb1aedfabe208a82602dc8d90c19e73482bdec5 references/romantic_soft_script34_L0448.wav
62
+ e7be698478d556d566cc676cdc52366abc77d95ee648e48fdb2701a498335f62 references/serious_script41_L0682.wav
63
+ 4f26e47e986ab5df9d8976e15a0f200fcbd50f3859485b41b870c357cc696954 references/soft_daily_script22_L0687.wav
64
+ 19d5a18a52fa0990451145708c52e4173ce822eadf943d7091fe8e1b5731bca1 samples/sample_long_statement.wav
65
+ 53a5b55e524d21b72bfae34d32a1f68251ac2eb14b24351819931ccd24411701 samples/sample_morning.wav
66
+ b2b9c2857a51b24b6d5868295ee320edc69c898233487aa553948027589d3684 scripts/install_pretrained.py
67
+ 13941b67f9a3fc73a1c937c26c3bfbd0308f135051063689747f82ef21408ab6 scripts/repo_paths.py
68
+ 7a93cc6d784796314ba39c65f4601ff623d28ac214bee3e14fb22010f017be37 scripts/run_tangyuge_infer.py
69
+ 717acdae8ded701d8b86adec65aaecdfc59e932c8ffdbcf63d7176903ffab4ea scripts/start_tangyuge_api.py
70
+ 036744c1e99fcdb0ce26ca1a1aa8af8341d7f31a7173d13e47b1433ef0041428 scripts/start_tangyuge_bridge.py
71
+ 74eb916e55d1aae115d3554c70ebdfff6c391dcdb7f708d6b0a2c05a99c4aa1c THIRD_PARTY.md
72
+ feea339f552ef580f96e43c49fa5d6f08989db1c4874e4c0d7cadf135159cdb5 weights/tangyuge_v2pp-e14.ckpt
73
+ dcb6042184355144dc15d118e88802fd4bdb494bd105da7c72ac42454d322899 weights/tangyuge_v2pp_e4_s564.pth
manifests/references.yaml ADDED
@@ -0,0 +1,338 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ voice_pack: tangyuge_v2pp_final
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ selected_final: main_hybrid_mid
5
+ weights:
6
+ gpt: weights/tangyuge_v2pp-e14.ckpt
7
+ sovits: weights/tangyuge_v2pp_e4_s564.pth
8
+ default_sampling:
9
+ top_k: 5
10
+ top_p: 0.8
11
+ temperature: 0.7
12
+ references:
13
+ - id: serious_script41_L0682
14
+ profile: profiles/main_hybrid_mid.yaml
15
+ file: references/serious_script41_L0682.wav
16
+ prompt_text: 气也撒了,如果你还不收敛,给在场辛苦的工作人员增添麻烦,丢人现眼的,也只有你自己了。
17
+ source_script: 41.csv
18
+ source_line: 682
19
+ source_voice_id: 4486
20
+ duration_sec: 8.391927
21
+ scene_tags:
22
+ - final
23
+ - serious
24
+ - long_statement
25
+ - firm
26
+ recommendation: 主线最终版。适合长台词、严肃说明、压住情绪的表达。
27
+ - id: morning_script43_L0993
28
+ profile: profiles/morning.yaml
29
+ file: references/morning_script43_L0993.wav
30
+ prompt_text: 曦铭?这么早就来学校了呀。
31
+ source_script: 43.csv
32
+ source_line: 993
33
+ source_voice_id: 43033
34
+ duration_sec: 3.034082
35
+ scene_tags:
36
+ - morning
37
+ - greeting
38
+ - short_sentence
39
+ recommendation: 适合早安、问候和短句寒暄,优先改善“早”等短句咬字。
40
+ - id: neutral_polite_script1_L0824
41
+ profile: profiles/neutral_polite.yaml
42
+ file: references/neutral_polite_script1_L0824.wav
43
+ prompt_text: 何同学好,当然可以呀,只要我知道的,我会言无不尽。
44
+ source_script: 1.csv
45
+ source_line: 824
46
+ source_voice_id: 418
47
+ duration_sec: 5.404762
48
+ scene_tags:
49
+ - neutral
50
+ - polite
51
+ - explanation
52
+ recommendation: 适合普通说明、礼貌回答和平稳叙述。
53
+ - id: soft_daily_script22_L0687
54
+ profile: profiles/soft_daily.yaml
55
+ file: references/soft_daily_script22_L0687.wav
56
+ prompt_text: 你先随意转转,柜台还有咖啡出售,我先去帮奶奶做最后的准备。
57
+ source_script: 22.csv
58
+ source_line: 687
59
+ source_voice_id: 2574
60
+ duration_sec: 4.977596
61
+ scene_tags:
62
+ - soft
63
+ - daily
64
+ - gentle
65
+ recommendation: 适合日常、温柔、低压交流。
66
+ - id: question_clear_script34_L0420
67
+ profile: profiles/question_clear.yaml
68
+ file: references/question_clear_script34_L0420.wav
69
+ prompt_text: 我还没来得及详细问,他是不是生病了?
70
+ source_script: 34.csv
71
+ source_line: 420
72
+ source_voice_id: 3821
73
+ duration_sec: 3.11415
74
+ scene_tags:
75
+ - question
76
+ - clear
77
+ - concerned
78
+ recommendation: 适合疑问句、关切式发问和自然上扬尾音。
79
+ - id: question_firm_script33_L0206
80
+ profile: profiles/question_firm.yaml
81
+ file: references/question_firm_script33_L0206.wav
82
+ prompt_text: 所以,你是要让龙主席看到你的这份论述稿?
83
+ source_script: 33.csv
84
+ source_line: 206
85
+ source_voice_id: 3546
86
+ duration_sec: 4.451474
87
+ scene_tags:
88
+ - question
89
+ - firm
90
+ - serious
91
+ recommendation: 适合带判断感、质询感的疑问句。
92
+ - id: lively_daily_script1_L0960
93
+ profile: profiles/lively_daily.yaml
94
+ file: references/lively_daily_script1_L0960.wav
95
+ prompt_text: 好呀,刚刚讲到哪里来着?
96
+ source_script: 1.csv
97
+ source_line: 960
98
+ source_voice_id: 505
99
+ duration_sec: 2.863968
100
+ scene_tags:
101
+ - lively
102
+ - daily
103
+ - light
104
+ recommendation: 适合轻快、日常、有一点活泼感的短句。
105
+ - id: casual_invite_script21_L0249
106
+ profile: profiles/casual_invite.yaml
107
+ file: references/casual_invite_script21_L0249.wav
108
+ prompt_text: 嘿,同桌,一起去食堂吃午饭吗?
109
+ source_script: 21.csv
110
+ source_line: 249
111
+ source_voice_id: 2117
112
+ duration_sec: 2.907347
113
+ scene_tags:
114
+ - casual
115
+ - invite
116
+ - friendly
117
+ recommendation: 适合熟人邀请、轻松交流和亲近感短句。
118
+ - id: narration_calm_script1_L0869
119
+ profile: profiles/narration_calm.yaml
120
+ file: references/narration_calm_script1_L0869.wav
121
+ prompt_text: 老四呀,他依旧在村子里过着恬淡平静的耕耘生活。
122
+ source_script: 1.csv
123
+ source_line: 869
124
+ source_voice_id: 448
125
+ duration_sec: 6.00551
126
+ scene_tags:
127
+ - narration
128
+ - calm
129
+ - descriptive
130
+ recommendation: 适合叙述、旁白、说明性长句和较平缓的讲述。
131
+ - id: apology_script32_L0040
132
+ profile: profiles/apology.yaml
133
+ file: references/apology_script32_L0040.wav
134
+ prompt_text: 抱歉,是我冒昧失礼了。
135
+ source_script: 32.csv
136
+ source_line: 40
137
+ source_voice_id: 3287
138
+ duration_sec: 2.648549
139
+ scene_tags:
140
+ - apology
141
+ - restrained
142
+ - polite
143
+ recommendation: 适合道歉、克制、礼貌退让和低冲突短句。
144
+ - id: bright_fireworks_script22_L0948
145
+ profile: profiles/bright_fireworks.yaml
146
+ file: references/bright_fireworks_script22_L0948.wav
147
+ prompt_text: 那么,关于烟火大会的实现计划,就重新启动吧。
148
+ source_script: 22.csv
149
+ source_line: 948
150
+ source_voice_id: 2737
151
+ duration_sec: 4.198639
152
+ scene_tags:
153
+ - bright
154
+ - fireworks
155
+ - uplifting
156
+ recommendation: 适合偏明亮、带期待感和目标感的句子。
157
+ - id: exclaim_script32_L0066
158
+ profile: profiles/exclaim.yaml
159
+ file: references/exclaim_script32_L0066.wav
160
+ prompt_text: 就是这个!读起来十分可爱呢!
161
+ source_script: 32.csv
162
+ source_line: 66
163
+ source_voice_id: 3299
164
+ duration_sec: 3.535578
165
+ scene_tags:
166
+ - exclaim
167
+ - excited
168
+ - emphatic
169
+ recommendation: 适合惊讶、感叹、带明显情绪起伏的短句。
170
+ - id: intimate_reassure_script41_L0688
171
+ profile: profiles/intimate_reassure.yaml
172
+ file: references/intimate_reassure_script41_L0688.wav
173
+ prompt_text: 亲爱的,已经没事了,不去理她就好。
174
+ source_script: 41.csv
175
+ source_line: 688
176
+ source_voice_id: 4490
177
+ duration_sec: 3.195329
178
+ scene_tags:
179
+ - intimate
180
+ - reassure
181
+ - affectionate
182
+ recommendation: 适合亲密安抚、哄劝、低压陪伴感句子。
183
+ - id: command_gentle_script43_L0650
184
+ profile: profiles/command_gentle.yaml
185
+ file: references/command_gentle_script43_L0650.wav
186
+ prompt_text: 你小点声,别让老师听见啦。
187
+ source_script: 43.csv
188
+ source_line: 650
189
+ source_voice_id: tyg-5
190
+ duration_sec: 3.499206
191
+ scene_tags:
192
+ - command
193
+ - gentle
194
+ - caution
195
+ recommendation: 适合提醒、劝阻、轻命令和压低声量的句子。
196
+ - id: food_daily_script43_L0998
197
+ profile: profiles/food_daily.yaml
198
+ file: references/food_daily_script43_L0998.wav
199
+ prompt_text: 喏,给你带的早餐,快趁热吃吧。
200
+ source_script: 43.csv
201
+ source_line: 998
202
+ source_voice_id: 43037
203
+ duration_sec: 3.252517
204
+ scene_tags:
205
+ - food
206
+ - daily
207
+ - caring
208
+ recommendation: 适合早餐、食堂、便当和生活照料感句子。
209
+ - id: romantic_soft_script34_L0448
210
+ profile: profiles/romantic_soft.yaml
211
+ file: references/romantic_soft_script34_L0448.wav
212
+ prompt_text: 可以吗亲爱的?会不会占用你太多时间了?
213
+ source_script: 34.csv
214
+ source_line: 448
215
+ source_voice_id: 3830
216
+ duration_sec: 3.321927
217
+ scene_tags:
218
+ - romantic
219
+ - soft
220
+ - intimate
221
+ recommendation: 适合偏恋爱向、含蓄亲密、带一点试探感的句子。
222
+ - id: cafeteria_ordering_script1_L0930
223
+ profile: profiles/cafeteria_ordering.yaml
224
+ file: references/cafeteria_ordering_script1_L0930.wav
225
+ prompt_text: 您好,我想要一份草莓小蛋糕,再来一杯奶茶。
226
+ source_script: 1.csv
227
+ source_line: 930
228
+ source_voice_id: 485
229
+ duration_sec: 3.636757
230
+ scene_tags:
231
+ - cafeteria
232
+ - ordering
233
+ - polite_daily
234
+ recommendation: 适合点单、食堂、饮品甜点和带礼貌感的生活场景。
235
+ - id: high_brightness_exclaim_script31_L0541
236
+ profile: profiles/high_brightness_exclaim.yaml
237
+ file: references/high_brightness_exclaim_script31_L0541.wav
238
+ prompt_text: 那太好了,能帮到你我就很开心。
239
+ source_script: 31.csv
240
+ source_line: 541
241
+ source_voice_id: 3261
242
+ duration_sec: 2.997166
243
+ scene_tags:
244
+ - bright
245
+ - happy
246
+ - exclaim
247
+ recommendation: 适合更明亮、更开心、更轻扬的积极短句。
248
+ - id: low_energy_discussion_script22_L0493
249
+ profile: profiles/low_energy_discussion.yaml
250
+ file: references/low_energy_discussion_script22_L0493.wav
251
+ prompt_text: 所以即使在知道之前答应,也绝对没关系的。
252
+ source_script: 22.csv
253
+ source_line: 493
254
+ source_voice_id: 2453
255
+ duration_sec: 3.309093
256
+ scene_tags:
257
+ - low_energy
258
+ - discussion
259
+ - calm
260
+ recommendation: 适合低能量讨论、克制解释、舒缓安抚式说明。
261
+ version: v2ProPlus
262
+ previous_v2pro_scene_sampling:
263
+ apology:
264
+ top_k: 12
265
+ top_p: 0.95
266
+ temperature: 0.9
267
+ bright_fireworks:
268
+ top_k: 12
269
+ top_p: 0.95
270
+ temperature: 0.9
271
+ cafeteria_ordering:
272
+ top_k: 12
273
+ top_p: 0.95
274
+ temperature: 0.9
275
+ casual_invite:
276
+ top_k: 12
277
+ top_p: 0.95
278
+ temperature: 0.9
279
+ command_gentle:
280
+ top_k: 12
281
+ top_p: 0.95
282
+ temperature: 0.9
283
+ exclaim:
284
+ top_k: 12
285
+ top_p: 0.95
286
+ temperature: 0.9
287
+ food_daily:
288
+ top_k: 12
289
+ top_p: 0.95
290
+ temperature: 0.9
291
+ high_brightness_exclaim:
292
+ top_k: 12
293
+ top_p: 0.95
294
+ temperature: 0.9
295
+ intimate_reassure:
296
+ top_k: 12
297
+ top_p: 0.95
298
+ temperature: 0.9
299
+ lively_daily:
300
+ top_k: 12
301
+ top_p: 0.95
302
+ temperature: 0.9
303
+ low_energy_discussion:
304
+ top_k: 12
305
+ top_p: 0.95
306
+ temperature: 0.9
307
+ main_hybrid_mid:
308
+ top_k: 12
309
+ top_p: 0.95
310
+ temperature: 0.9
311
+ morning:
312
+ top_k: 14
313
+ top_p: 0.97
314
+ temperature: 0.92
315
+ narration_calm:
316
+ top_k: 12
317
+ top_p: 0.95
318
+ temperature: 0.9
319
+ neutral_polite:
320
+ top_k: 12
321
+ top_p: 0.95
322
+ temperature: 0.9
323
+ question_clear:
324
+ top_k: 12
325
+ top_p: 0.95
326
+ temperature: 0.9
327
+ question_firm:
328
+ top_k: 12
329
+ top_p: 0.95
330
+ temperature: 0.9
331
+ romantic_soft:
332
+ top_k: 12
333
+ top_p: 0.95
334
+ temperature: 0.9
335
+ soft_daily:
336
+ top_k: 12
337
+ top_p: 0.95
338
+ temperature: 0.9
manifests/training_report.md ADDED
@@ -0,0 +1,73 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # 训练与评测报告:TangYuGe GPT-SoVITS v2ProPlus
2
+
3
+ 本文记录本模型的数据、训练配置、选型方法与评测结果,供复现与对比参考。
4
+
5
+ ## 1. 数据
6
+
7
+ | 项目 | 内容 |
8
+ | --- | --- |
9
+ | 来源 | 《Sparkling Memories》(夏日花火)中角色唐语歌的全部语音台词,702 条,约 48.8 分钟,44.1 kHz 单声道 |
10
+ | 筛选规则 | 时长 1–12 秒,去除标点与语气词后有效汉字不少于 6 个;台词外层「」去掉 |
11
+ | 训练集 | 561 条,约 43.1 分钟 |
12
+ | 标注核对 | 用 Paraformer 语音识别逐条转写,与台本按无调拼音比对:平均差异 2.0%,无一条超过 30%;差异几乎都是人名同音字与语气词写法,台本可信 |
13
+ | 音质 | 原声停顿处底噪约 -56 dB,属干净录音 |
14
+
15
+ ## 2. 训练配置
16
+
17
+ 底模:GPT-SoVITS v2ProPlus(SoVITS `s2Gv2ProPlus.pth` / `s2Dv2ProPlus.pth`,GPT `s1v3.ckpt`)。两个模型均为全参数微调。
18
+
19
+ | 阶段 | batch size | 学习率 | 精度 | 训练轮数 | 选用 |
20
+ | --- | --- | --- | --- | --- | --- |
21
+ | SoVITS | 4 | 1e-4(文本编码器 ×0.4) | fp16 混合精度 | 8 epoch | 第 4 epoch(564 步) |
22
+ | GPT | 4 | 0.002 | 16-bit 混合精度 | 15 epoch | 第 14 epoch |
23
+
24
+ 另训练了 SoVITS batch size 2 的对照组(6 epoch),用于确认 batch size 的影响。
25
+
26
+ 硬件:NVIDIA RTX 4060 Laptop 8GB;SoVITS 每 epoch 约 2 分钟,GPT 每 epoch 不到 1 分钟。
27
+
28
+ ## 3. 评测方法
29
+
30
+ - 测试文本:25 句,覆盖短句、长句、疑问、感叹、省略号停顿、连续顿号等容易出错的情形。
31
+ - 参考音频:4 条(严肃长句、早安问候、关切疑问、温柔日常),每条配对原台词文本。
32
+ - 固定随机种子,同一文本、同一参考、同一采样参数下比较不同模型。
33
+ - 指标:
34
+
35
+ | 指标 | 含义 | 方向 |
36
+ | --- | --- | --- |
37
+ | 拼音错误率(PER) | Paraformer 识别合成语音,与原文按无调拼音比对,衡量吞字、错字 | 越低越好 |
38
+ | DNSMOS SIG / BAK / OVRL | 微软 DNSMOS P.835 估计的语音失真 / 背景噪声 / 整体质量 | 越高越好 |
39
+ | 停顿底噪电平 | 非静音帧 RMS 的 10 分位,衡量停顿处的嘶声与底噪 | 越低越好 |
40
+ | 音高抖动 | 浊音段基频曲线的二阶差分,衡量“字音发飘” | 越低越好 |
41
+ | 声纹相似度 | ERes2NetV2 声纹向量与训练语音平均声纹的余弦相似度 | 越高越好 |
42
+
43
+ - 规模:共合成约 4600 条语音,涵盖 36 组 GPT × SoVITS epoch 组合、GPT 与 SoVITS 各自的逐 epoch 精扫、5 组采样参数。
44
+ - 统计:结论用配对 bootstrap(5000 次重采样)与符号检验核查,95% 置信区间不含 0 才视为显著。
45
+
46
+ ## 4. 结果
47
+
48
+ ### 与旧版 v2Pro 微调模型对比
49
+
50
+ 同一测试集、4 条参考音频,两者均使用 `top_k=12 / top_p=0.95 / temperature=0.9`(旧版的推荐参数),本模型为 GPT 第 14 epoch + SoVITS 第 4 epoch。
51
+
52
+ | 指标 | 旧版 v2Pro | 本模型 | 差异 |
53
+ | --- | --- | --- | --- |
54
+ | 停顿底噪电平 | -48.5 dB | -54.3 dB | 低约 6 dB(显著) |
55
+ | DNSMOS BAK | 3.42 | 3.62 | +0.2(显著) |
56
+ | DNSMOS OVRL | 2.59 | 2.72 | +0.13(显著) |
57
+ | 音高抖动 | 0.234 | 0.211 | 约 -9%(显著) |
58
+ | 拼音错误率 | 1.4% | 1.3% | 无显著差异 |
59
+ | 声纹相似度 | 0.888 | 0.901 | 无显著差异 |
60
+
61
+ 作为参考,原声录音的 DNSMOS BAK 为 3.46、OVRL 为 2.67,停顿底噪约 -56 dB,音高抖动 0.226。
62
+
63
+ ### 选型
64
+
65
+ - SoVITS:第 4 epoch 在所有 GPT epoch 搭配下底噪都最低(比相邻 epoch 低 1–3 dB);SoVITS epoch 对发音准确率无显著影响。batch size 2 对照组 DNSMOS 略高但底噪高约 2.5 dB,不是更优。
66
+ - GPT:第 10–14 epoch 发音最稳定,第 15 epoch 底噪上升;第 13 与第 14 epoch 客观指标无法区分,经人工试听选第 14 epoch。
67
+ - 采样参数:保守参数(`top_k=5`、`top_p=0.8`、`temperature=0.7`)在短句上有减少吞字的趋势,两轮人工试听均优于原参数,定为默认。
68
+ - 推理精度:fp16 与 fp32 无可测差异,默认 fp16。
69
+
70
+ ## 5. 已知问题
71
+
72
+ - 短句偶发提前结束(如“好呀,就这样吧”只读出“好呀”),新旧模型都存在,是 GPT-SoVITS 自回归生成的共性问题。
73
+ - 部分 Intel 混合架构 CPU 上 PyTorch oneDNN 多线程卷积会算错(已在 i7-13650HX 复现);本模型训练与推理均在 GPU 完成,不受影响。CPU 推理请关闭 oneDNN 或使用单线程。
profiles/apology.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_apology
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: apology_script32_L0040
13
+ ref_audio_path: ../references/apology_script32_L0040.wav
14
+ prompt_text: 抱歉,是我冒昧失礼了。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - apology
27
+ - restrained
28
+ - polite
29
+ usage: 适合道歉、克制、礼貌退让和低冲突短句。
profiles/bright_fireworks.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_bright_fireworks
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: bright_fireworks_script22_L0948
13
+ ref_audio_path: ../references/bright_fireworks_script22_L0948.wav
14
+ prompt_text: 那么,关于烟火大会的实现计划,就重新启动吧。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - bright
27
+ - fireworks
28
+ - uplifting
29
+ usage: 适合偏明亮、带期待感和目标感的句子。
profiles/cafeteria_ordering.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_cafeteria_ordering
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: cafeteria_ordering_script1_L0930
13
+ ref_audio_path: ../references/cafeteria_ordering_script1_L0930.wav
14
+ prompt_text: 您好,我想要一份草莓小蛋糕,再来一杯奶茶。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - cafeteria
27
+ - ordering
28
+ - polite_daily
29
+ usage: 适合点单、食堂、饮品甜点和带礼貌感的生活场景。
profiles/casual_invite.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_casual_invite
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: casual_invite_script21_L0249
13
+ ref_audio_path: ../references/casual_invite_script21_L0249.wav
14
+ prompt_text: 嘿,同桌,一起去食堂吃午饭吗?
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - casual
27
+ - invite
28
+ - friendly
29
+ usage: 适合熟人邀请、轻松交流和亲近感短句。
profiles/command_gentle.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_command_gentle
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: command_gentle_script43_L0650
13
+ ref_audio_path: ../references/command_gentle_script43_L0650.wav
14
+ prompt_text: 你小点声,别让老师听见啦。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - command
27
+ - gentle
28
+ - caution
29
+ usage: 适合提醒、劝阻、轻命令和压低声量的句子。
profiles/exclaim.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_exclaim
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: exclaim_script32_L0066
13
+ ref_audio_path: ../references/exclaim_script32_L0066.wav
14
+ prompt_text: 就是这个!读起来十分可爱呢!
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - exclaim
27
+ - excited
28
+ - emphatic
29
+ usage: 适合惊讶、感叹、带明显情绪起伏的短句。
profiles/food_daily.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_food_daily
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: food_daily_script43_L0998
13
+ ref_audio_path: ../references/food_daily_script43_L0998.wav
14
+ prompt_text: 喏,给你带的早餐,快趁热吃吧。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - food
27
+ - daily
28
+ - caring
29
+ usage: 适合早餐、食堂、便当和生活照料感句子。
profiles/high_brightness_exclaim.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_high_brightness_exclaim
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: high_brightness_exclaim_script31_L0541
13
+ ref_audio_path: ../references/high_brightness_exclaim_script31_L0541.wav
14
+ prompt_text: 那太好了,能帮到你我就很开心。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - bright
27
+ - happy
28
+ - exclaim
29
+ usage: 适合更明亮、更开心、更轻扬的积极短句。
profiles/intimate_reassure.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_intimate_reassure
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: intimate_reassure_script41_L0688
13
+ ref_audio_path: ../references/intimate_reassure_script41_L0688.wav
14
+ prompt_text: 亲爱的,已经没事了,不去理她就好。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - intimate
27
+ - reassure
28
+ - affectionate
29
+ usage: 适合亲密安抚、哄劝、低压陪伴感句子。
profiles/lively_daily.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_lively_daily
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: lively_daily_script1_L0960
13
+ ref_audio_path: ../references/lively_daily_script1_L0960.wav
14
+ prompt_text: 好呀,刚刚讲到哪里来着?
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - lively
27
+ - daily
28
+ - light
29
+ usage: 适合轻快、日常、有一点活泼感的短句。
profiles/low_energy_discussion.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_low_energy_discussion
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: low_energy_discussion_script22_L0493
13
+ ref_audio_path: ../references/low_energy_discussion_script22_L0493.wav
14
+ prompt_text: 所以即使在知道之前答应,也绝对没关系的。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - low_energy
27
+ - discussion
28
+ - calm
29
+ usage: 适合低能量讨论、克制解释、舒缓安抚式说明。
profiles/main_hybrid_mid.yaml ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_main
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: final_selected
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: serious_script41_L0682
13
+ ref_audio_path: ../references/serious_script41_L0682.wav
14
+ prompt_text: 气也撒了,如果你还不收敛,给在场辛苦的工作人员增添麻烦,丢人现眼的,也只有你自己了。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - final
27
+ - serious
28
+ - long_statement
29
+ - firm
30
+ usage: 当前主线最终版,适合严肃说明、长台词和情绪偏稳的表达。
profiles/morning.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_morning
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: morning_script43_L0993
13
+ ref_audio_path: ../references/morning_script43_L0993.wav
14
+ prompt_text: 曦铭?这么早就来学校了呀。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - morning
27
+ - greeting
28
+ - short_sentence
29
+ usage: 适合早安、问候、短句寒暄;优先改善“早”等短句咬字。
profiles/narration_calm.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_narration_calm
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: narration_calm_script1_L0869
13
+ ref_audio_path: ../references/narration_calm_script1_L0869.wav
14
+ prompt_text: 老四呀,他依旧在村子里过着恬淡平静的耕耘生活。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - narration
27
+ - calm
28
+ - descriptive
29
+ usage: 适合叙述、旁白、说明性长句和较平缓的讲述。
profiles/neutral_polite.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_neutral_polite
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: neutral_polite_script1_L0824
13
+ ref_audio_path: ../references/neutral_polite_script1_L0824.wav
14
+ prompt_text: 何同学好,当然可以呀,只要我知道的,我会言无不尽。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - neutral
27
+ - polite
28
+ - explanation
29
+ usage: 适合普通说明、平稳回答和礼貌叙述。
profiles/question_clear.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_question_clear
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: question_clear_script34_L0420
13
+ ref_audio_path: ../references/question_clear_script34_L0420.wav
14
+ prompt_text: 我还没来得及详细问,他是不是生病了?
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - question
27
+ - clear
28
+ - concerned
29
+ usage: 适合疑问句、关切式发问和上扬尾音。
profiles/question_firm.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_question_firm
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: question_firm_script33_L0206
13
+ ref_audio_path: ../references/question_firm_script33_L0206.wav
14
+ prompt_text: 所以,你是要让龙主席看到你的这份论述稿?
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - question
27
+ - firm
28
+ - serious
29
+ usage: 适合带判断感、质询感的疑问句。
profiles/romantic_soft.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_romantic_soft
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: romantic_soft_script34_L0448
13
+ ref_audio_path: ../references/romantic_soft_script34_L0448.wav
14
+ prompt_text: 可以吗亲爱的?会不会占用你太多时间了?
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - romantic
27
+ - soft
28
+ - intimate
29
+ usage: 适合偏恋爱向、含蓄亲密、带一点试探感的句子。
profiles/soft_daily.yaml ADDED
@@ -0,0 +1,29 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ name: tangyuge_v2pp_final_soft_daily
2
+ character: 唐语歌
3
+ source_game: Sparkling Memories
4
+ model_family: GPT-SoVITS
5
+ version: v2ProPlus
6
+ status: scene_reference
7
+ tts_config: ../configs/tts_infer_pack.yaml
8
+ weights:
9
+ gpt: ../weights/tangyuge_v2pp-e14.ckpt
10
+ sovits: ../weights/tangyuge_v2pp_e4_s564.pth
11
+ prompt:
12
+ reference_id: soft_daily_script22_L0687
13
+ ref_audio_path: ../references/soft_daily_script22_L0687.wav
14
+ prompt_text: 你先随意转转,柜台还有咖啡出售,我先去帮奶奶做最后的准备。
15
+ prompt_lang: zh
16
+ defaults:
17
+ text_lang: zh
18
+ text_split_method: cut5
19
+ media_type: wav
20
+ batch_size: 1
21
+ top_k: 5
22
+ top_p: 0.8
23
+ temperature: 0.7
24
+ scene:
25
+ tags:
26
+ - soft
27
+ - daily
28
+ - gentle
29
+ usage: 适合日常、温柔、低压交流。
references/apology_script32_L0040.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b7b74e3b4353bff0b7d69128ea879607acba64cff4b1185927c2b2db83c120f0
3
+ size 234012
references/bright_fireworks_script22_L0948.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dc043035484e6253380e8262bd8630370d57057817e24b0121bcb79abee6b393
3
+ size 370432
references/cafeteria_ordering_script1_L0930.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57e41eb33ff65c4ee48792a852ad8b3dec0d433539859f73357bd0f822345801
3
+ size 320806
references/casual_invite_script21_L0249.wav ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0f452a82cf2c831ea136d43dee3d12d875e37dad30106ec60cbead18d927a438
3
+ size 256472