# kikiri_pipeline — additional dependencies for the FlowTTS-style HF pipeline. # # These are in ADDITION to the base kikiri-tts deps in pyproject.toml: # torch, transformers, huggingface_hub, loguru, misaki[en,de], soundfile, numpy # # Install with: # pip install -r requirements-pipeline.txt # # or with uv: # uv pip install -r requirements-pipeline.txt # HuggingFace dataset loading (humair025/koku) datasets>=2.18.0 # YAML config loading pyyaml>=6.0 # Audio resampling fallback (if torchaudio not available) + F0 via pyin librosa>=0.10.0 # TensorBoard logging tensorboard>=2.14.0 # Progress bars (used by dataset prep) tqdm>=4.66.0 # Optional: W&B logging (uncomment to enable) # wandb>=0.16.0 # Optional: speaker clustering (only needed if you re-run scripts/prepare_dataset.py # on local MP3s — not needed for the HF dataset pipeline) # resemblyzer>=0.1.4 # scikit-learn>=1.3.0 # Optional: mlx-whisper transcription (only needed if you transcribe local MP3s) # mlx-whisper>=0.1.0