---
license: apache-2.0
pipeline_tag: text-to-speech
language:
  - en
  - zh
  - multilingual
tags:
  - vocence
  - bittensor
  - prompt-tts
  - omnivoice
  - sad-tts-v1
---

# Sad-tts-v1 (Vocence PromptTTS)

**Sad-tts-v1** is a **[Vocence SN78](https://github.com/Vocence-bt/vocence)** miner bundle (OmniVoice backbone). Hub: **`KGSS/Sad-tts-v1`** — natural-language **instruction** plus **text** in, **mono WAV** out via [`miner.py`](./miner.py).

Base weights align with **[k2-fsa/OmniVoice](https://huggingface.co/k2-fsa/OmniVoice)** · Apache-2.0 · [`omnivoice`](https://pypi.org/project/omnivoice/) runtime. Pin commits in `miner_deploy_sad_tts_v1.py` and [VOCENCE_HF.md](./VOCENCE_HF.md).

## Vocence contract

| Method | Role |
|--------|------|
| `Miner(path_hf_repo)` | Load checkpoint from a directory (or HF snapshot) containing `config.json`, `model.safetensors`, tokenizers, and `audio_tokenizer/`. |
| `warmup()` | One short synthesis to prime the stack. |
| `generate_wav(instruction, text)` | Returns `(float32 mono ndarray, sample_rate)`; typically **24 kHz**. |

Validators send **free-form** `instruction`. OmniVoice **voice-design** only accepts [whitelisted attribute tags](https://huggingface.co/k2-fsa/OmniVoice); **`miner.py` maps** keywords (gender, age, pitch, whisper, accent, Chinese dialects) to those tags. Unmatched instructions fall back to `runtime.default_instruct` in [`vocence_config.yaml`](./vocence_config.yaml).

## Repo layout

| File | Role |
|------|------|
| `miner.py` | Engine + NL → `instruct` mapping |
| `chute_config.yml` | Chutes image (PyTorch cu128 + `omnivoice`) |
| `vocence_config.yaml` | Limits, default voice tags, `num_step` / `guidance_scale` |
| Weight files | Shipped in this Hub repo (`model.safetensors`, `audio_tokenizer/`); see [VOCENCE_HF.md](./VOCENCE_HF.md) |

## Local quick check (GPU)

```bash
pip install omnivoice torch torchaudio  # match CUDA index from chute_config.yml
# Copy snapshot: huggingface-cli download k2-fsa/OmniVoice --local-dir ./OmniVoice_weights
python -c "
from pathlib import Path
from miner import Miner
m = Miner(Path('./OmniVoice_weights'))
m.warmup()
w, sr = m.generate_wav('Calm female voice, British accent.', 'Hello from OmniVoice on Vocence.')
print(w.shape, sr)
"
```

## License

**Apache-2.0** for this packaging layout and miner glue; **OmniVoice** weights and upstream code remain under their stated licenses on the [model card](https://huggingface.co/k2-fsa/OmniVoice).