kikiri-tts
Copyright 2026 semidark

This project contains code from the following open-source projects:

────────────────────────────────────────────────────────────────────────────────

kokoro/ — Kokoro TTS inference package (git submodule)
  Source:  https://github.com/hexgrad/kokoro
  Fork:    https://github.com/semidark/kokoro (branch: main)
  License: Apache License 2.0
  Author:  hexgrad
  Changes: Fork includes German language support (`'d'`) in pipeline/CLI.

────────────────────────────────────────────────────────────────────────────────

StyleTTS2/ — Training code (git submodule)
  Source:  https://github.com/yl4579/StyleTTS2
  Fork:    https://github.com/semidark/StyleTTS2 (branch: main)
  License: MIT License
  Author:  Yinghao Aaron Li (yl4579)
  Changes: See the commit history on the main branch for a full
           list of patches. Summary:
           - Migrated weight_norm/spectral_norm to parametrizations API
           - Fixed multiple training bugs (F0 shape, checkpoint save order,
             missing .train() mode, hardcoded breakpoint, torch.load defaults)
           - Added Kokoro-82M symbol table (kokoro_symbols.py)
           - Filtered sequences exceeding PLBERT max length (meldataset.py)

────────────────────────────────────────────────────────────────────────────────

The following directories contain original work by this project's contributors:

  scripts/     — Dataset preparation, voicepack extraction, inference testing
  configs/     — Training configuration for German fine-tuning
  docs/        — Training guide and documentation
  training/    — Training data lists and OOD texts (audio files excluded)