---
license: apache-2.0
pipeline_tag: text-to-speech
library_name: qwen-tts
tags:
  - audio
  - tts
  - qwen
  - voice-design
  - english
  - vocence
---

# vocence-ninja

A Vocence-compliant PromptTTS miner repo. Provides text-to-speech via the
`generate_voice_design(text, instruct)` API on top of
[`Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign`](https://huggingface.co/Qwen/Qwen3-TTS-12Hz-1.7B-VoiceDesign).

Designed to be deployed on [Chutes](https://chutes.ai) under the [Vocence
subnet](https://github.com/Vocence-78/vocence) using the canonical wrapper.

## Files

| File | Purpose |
|------|---------|
| `miner.py` | PromptTTS engine. Class `Miner(path_hf_repo)`, `warmup()`, `generate_wav(instruction, text)` |
| `vocence_config.yaml` | Runtime + generation config. **Top-level `model_name` must match your on-chain commitment.** |
| `chute_config.yml` | Image / NodeSelector / Chute settings for `chutes build` |
| `model.safetensors` | Model weights |
| `config.json` | HuggingFace `Qwen3TTSConfig` |
| `tokenizer_config.json`, `vocab.json`, `merges.txt`, `preprocessor_config.json` | Text tokenizer |
| `speech_tokenizer/` | 12 Hz acoustic codec |