Loading
Vous avez un workflow ComfyUI ? Obtenez une documentation lisible directement dans votre navigateur.Rendre un workflow lisible.Essayer maintenant ↗
higgs-audio-2-community
Source du modèle
Description de la source
INT8-HQ ONNX export of k2-fsa/OmniVoice,
optimized to run on CPU with onnxruntime (no GPU, no PyTorch needed at inference time).
Sources
1 sourceVérifié 19 sept.
Artefacts du modèle
3 artefactsaudio_tokenizer_decoder_int8/model.onnx
onnx · 226 KB · SHA-256 58833acb4e01…7d30 · Hugging Face
Téléchargeraudio_tokenizer_encoder_int8/model.onnx
onnx · 501 KB · SHA-256 7876269f66f3…803c · Hugging Face
TéléchargerExtraits de sources
2 extraits| Folder | Role | Size |
|---|---|---|
omnivoice_lm_int8_hq/ | Diffusion LM (INT8-HQ, audio head kept FP32) | ~611 MB |
audio_tokenizer_decoder_int8/ | tokens → 24 kHz waveform | ~82 MB |
audio_tokenizer_encoder_int8/ | reference audio → codec tokens (voice cloning) | ~378 MB |
tokenizer.json, config.json | text tokenizer + config | small |
Each model is
model.onnx+model.onnx_data(ONNX external data) — download both files.
| Model | Input | Output |
|---|---|---|
| LM | input_ids[batch,8,seq] int64; audio_mask[batch,seq] bool; attention_mask[batch,1,seq,seq] bool; position_ids[batch,seq] int64 | logits[batch,8,seq,1025] f32 |
| Encoder | audio[batch,1,num_samples] f32 @24kHz | audio_codes[batch,8,num_frames] int64 |
| Decoder | audio_codes[batch,8,num_frames] int64 | audio[batch,1,num_samples] f32 |
The sampling loop (diffusion + CFG) runs in Python; the LM / encoder / decoder run via ONNX.
Reference inference code: AFun9/Omnivoice-onnx
(infer_onnx.py). Example:
python infer_onnx.py --variant int8hq --num-step 16 --only demo_voice_clone
RTF ~3–7 depending on the CPU (i9-14900KF ~2.8–6.7; slower CPUs are slower). Suitable for
batch / offline use, not real-time. Increase --num-step for higher quality (slower).
This export bundles two components under different licenses — you must comply with both:
omnivoice_lm_int8_hq/): derived from Qwen3-0.6B / OmniVoice — Apache-2.0.audio_tokenizer_*_int8/): derived from Higgs Audio 2 (Boson AI) —
Boson Higgs Audio 2 Community License (based on the Meta Llama 3 Community License); see
LICENSE. It includes an acceptable-use policy and Llama-style terms.Because it contains the Higgs Audio 2 component, the repo license is declared as other.
This is an ONNX export + quantization only — weights are unchanged; all rights belong to the
original authors (k2-fsa/OmniVoice, Boson AI, Qwen).
omnivoice_lm_int8_hq/model.onnx
onnx · 1,28 MB · SHA-256 4a79b508bf2c…944e · Hugging Face
--- language: - vi - en - zh - multilingual pipeline_tag: text-to-speech tags: - tts - onnx - int8 - voice-cloning - omnivoice - cpu library_name: onnxruntime base_model: k2-fsa/OmniVoice license: other license_name: higgs-audio-2-community license_link: LICENSE --- # OmniVoice ONNX (INT8-HQ) — CPU inference **INT8-HQ ONNX** export of [k2-fsa/OmniVoice](https://huggingface.co/k2-fsa/OmniVoice), optimized to run **on CPU** with `onnxruntime` (no GPU, no PyTorch needed at inference time). - Multilingual TTS (646 languages incl. **Vietnamese**) + zero-shot **voice cloning** + voice design. - **24 kHz** output, 8 codebooks, frame rate 75. - INT8-HQ: cosine > 0.9999 vs FP32, perceptually almost indistinguishable. ## Components | Folder | Role | Size | |---|---|---| | `omnivoice_lm_int8_hq/` | Diffusion LM (INT8-HQ, audio head kept FP32) | ~611 MB | | `audio_tokenizer_decoder_int8/` | tokens → 24 kHz waveform | ~82 MB | | `audio_tokenizer_encoder_int8/` | reference audio → codec tokens (voice cloning) | ~378 MB | | `tokenizer.json`, `config.json` | text tokenizer + config | small | > Each model is `model.onnx` + `model.onnx_data` (ONNX external data) — download **both** files. ## ONNX I/O | Model | Input | Output | |---|---|---| | **LM** | `input_ids[batch,8,seq]` int64; `audio_mask[batch,seq]` bool; `attention_mask[batch,1,seq,seq]` bool; `position_ids[batch,seq]` int64 | `logits[batch,8,seq,1025]` f32 | | **Encoder** | `audio[batch,1,num_samples]` f32 @24kHz | `audio_codes[batch,8,num_frames]` int64 | | **Decoder** | `audio_codes[batch,8,num_frames]` int64 | `audio[batch,1,num_samples]` f32 | ## Inference The sampling loop (diffusion + CFG) runs in Python; the LM / encoder / decoder run via ONNX. Reference inference code: [AFun9/Omnivoice-onnx](https://github.com/AFun9/Omnivoice-onnx) (`infer_onnx.py`). Example: ```bash python infer_onnx.py --variant int8hq --num-step 16 --only demo_voice_clone ``` ## Performance (CPU) RTF ~3–7 depending on the CPU (i9-14900KF ~2.8–6.7; slower CPUs are slower). Suitable for batch / offline use, not real-time. Increase `--num-step` for higher quality (slower). ## License This export bundles **two components under different licenses** — you must comply with **both**: - **Diffusion LM** (`omnivoice_lm_int8_hq/`): derived from Qwen3-0.6B / OmniVoice — **Apache-2.0**. - **Audio tokenizer** (`audio_tokeniz...
Source context: 41 downloads · 2 likes · Pipeline text-to-speech · Library onnxruntime · Repo ct03/omnivoice-onnx-int8hq