endo5501/audio.cpp

1---2license: mit3language:4  - ja5tags:6  - text-to-speech7  - audio8---9 10# Irodori-TTS model assets for audio.cpp (NovelViewer)11 12Runtime model assets for the [endo5501/audio.cpp](https://github.com/endo5501/audio.cpp) fork13(Irodori-TTS engine used by NovelViewer). This repository repackages the minimal file set14required by the audio.cpp `irodori_tts` safetensors loader, laid out as sibling directories:15 16```17Irodori-TTS-600M-v3-VoiceDesign/18  model.safetensors19  model_config.json20llm-jp-3-150m/21  tokenizer.json22Semantic-DACVAE-Japanese-32dim/23  weights.safetensors   (converted from upstream weights.pth)24```25 26## Sources and licenses27 28| Asset | Upstream | License |29|---|---|---|30| Irodori-TTS-600M-v3-VoiceDesign | [Aratako/Irodori-TTS-600M-v3-VoiceDesign](https://huggingface.co/Aratako/Irodori-TTS-600M-v3-VoiceDesign) | MIT (+ ethical restrictions, see below) |31| llm-jp-3-150m tokenizer | [llm-jp/llm-jp-3-150m](https://huggingface.co/llm-jp/llm-jp-3-150m) | Apache-2.0 |32| Semantic-DACVAE-Japanese-32dim | [Aratako/Semantic-DACVAE-Japanese-32dim](https://huggingface.co/Aratako/Semantic-DACVAE-Japanese-32dim) | MIT |33 34`Semantic-DACVAE-Japanese-32dim/weights.safetensors` is a format conversion35(PyTorch `weights.pth` → safetensors) of the upstream checkpoint; weights are unmodified.36 37## Ethical restrictions (inherited from Irodori-TTS)38 39In addition to the MIT license terms, the upstream Irodori-TTS model states ethical40restrictions on use (e.g., prohibiting impersonation without consent and unlawful use).41See the upstream model card for the authoritative text:42https://huggingface.co/Aratako/Irodori-TTS-600M-v3-VoiceDesign43