nguyen-brat/nguyen-ngoc-ngan-zero-shot-tts

Vietnamese Text-to-Speech Demo

Zero-shot voice cloning for Vietnamese using VieNeu-TTS (Qwen2 0.5B + NeuCodec).

Features

• —Short-text synthesis — single sentence or paragraph
• —Long-form synthesis — multi-paragraph stories and articles with automatic sentence chunking
• —Voice cloning — clone any voice from a 3–10s reference WAV
• —Preset voices — 6 built-in Vietnamese speaker presets
• —Fine-tuned model — set the ADAPTER_REPO Space secret to a HuggingFace LoRA adapter repo

Adding a Fine-tuned Adapter

Set the ADAPTER_REPO Space secret to your HuggingFace Hub repo ID (e.g. your-username/vieneu-lora-adapter). The app will automatically load it and show a fine-tuned mode option.

If you have local seed clips for the fine-tuned voice, place seed_XX.wav + seed_XX.lab files in data/seed_clips/.