ICTNLP/indextts2_decoder

1---2language:3- en4- zh5license: other6license_name: bilibili-model-use-license-agreement7license_link: https://github.com/ictnlp/FreezeEmpath/blob/main/licenses/IndexTTS2.txt8tags:9- speech10- audio-to-audio11- freezeempath12---13 14# IndexTTS2-derived waveform decoder for FreezeEmpath15 16This package converts FreezeEmpath speech tokens into 22,050 Hz mono waveforms, using components from [IndexTTS2](https://github.com/index-tts/index-tts), Amphion/MaskGCT, and NVIDIA BigVGAN.17 18[FreezeEmpath model](https://huggingface.co/ICTNLP/FreezeEmpath) · [Code and instructions](https://github.com/ictnlp/FreezeEmpath) · [Paper](https://arxiv.org/abs/2604.18159)19 20## Contents and use21 22- `config.yaml`: semantic codec and flow-matching configuration.23- `semantic_codec.safetensors`: semantic codec weights.24- `s2mel.pth`: flow-matching model weights.25- `bigvgan_generator.pt`: vocoder weights.26- `style.pt`, `prompt_condition.pt`, `ref_mel.pt`: fixed voice conditions used by the released example.27 28Use the FreezeEmpath GitHub code and its bundled BigVGAN architecture configuration:29 30```bash31python inference/codes2wav.py \32  --cfg_path checkpoints/indextts2_codes2wav/config.yaml \33  --model_dir checkpoints/indextts2_codes2wav \34  --input_file outputs/answers.jsonl \35  --output_dir outputs/wav --seed 4236```37 38The input must contain `prediction_units` speech tokens generated by `inference/infer.py --s2s`; this package is not a standalone text-to-speech interface. The example uses a fixed voice, not user-selected speaker prompts.39 40## License41 42This IndexTTS2-derived package is subject to the [bilibili Model Use License Agreement](https://github.com/ictnlp/FreezeEmpath/blob/main/licenses/IndexTTS2.txt). Individual BigVGAN and Amphion components retain their upstream MIT notices. See [third-party notices](https://github.com/ictnlp/FreezeEmpath/blob/main/THIRD_PARTY_NOTICES.md).43 44Any modifications made to the original model in this Derivative Work are not endorsed, warranted, or guaranteed by the original right-holder of the original model, and the original right-holder disclaims all liability related to this Derivative Work.45