voicedesign_t0

PromptTTS model based on Qwen3-TTS-12Hz-1.7B-Base, with small updates from fine-tuning on additional training data.

Contents

• miner.py — Vocence PromptTTS engine (class Miner: __init__, warmup, generate_wav)
• chute_config.yml — Chutes build config (image, node selector, chute settings)
• vocence_config.yaml — runtime options (sample_rate, adapter, limits)
• model.safetensors — fine-tuned model weights
• speech_tokenizer/ — RVQ speech tokenizer
• tokenizer_config.json, vocab.json, merges.txt — text tokenizer assets

Training notes

Small incremental update over the base model, fine-tuned on a modest instruction-annotated dataset. Intended for Vocence subnet (Bittensor SN78) PromptTTS evaluation.