metadata

title: Amphion Vevo Voice Conversion
emoji: 🎤
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 4.8.0
app_file: app.py
pinned: false
python_version: '3.10'

Amphion's Vevo - Voice Conversion & TTS

This is a Gradio web interface for the Vevo voice conversion model from the Amphion toolkit. It supports:

Voice conversion (transferring both style and timbre)
Timbre-only conversion
Text-to-Speech with voice cloning

Usage

Select mode:
- Voice: Convert voice with both style and timbre transfer
- Timbre: Convert only the timbre of the voice
- TTS: Generate speech from text with voice cloning
Upload audio files based on mode:
- Source Audio: Your input audio (for voice and timbre modes)
- Reference Style: Style reference (for voice and TTS modes)
- Reference Timbre: Voice reference (required for all modes)
For TTS mode:
- Enter the text you want to convert to speech
- Optionally provide reference text
- Select source and reference languages
Adjust Flow Matching Steps (1-64, default: 32)
- Higher values give better quality but take longer
- Lower values are faster but may reduce quality
Click "Generate" to create the converted audio

Sample Files

Sample audio files are available in the Amphion/models/vc/vevo/wav/ directory:

arabic_male.wav
source.wav

Technical Requirements

Python 3.10+
CUDA-capable GPU recommended for faster inference
Minimum 12GB storage space for models

Models

The application automatically downloads required models from Hugging Face:

Content Tokenizer (vq32)
Content-Style Tokenizer (vq8192)
Autoregressive Transformer
Flow Matching Transformer
Vocoder