Vocazee

A voice cloning and text-to-speech application that can generate speech in any voice.

Technical Details

• Frontend: React
• Backend: FastAPI
• Text-to-speech: Tortoise TTS

Installation

1. Clone the repository:

git clone https://github.com/taeefnajib/vocazee.git
cd vocazee

2. Build and start the containers:

docker compose up --build

Note: The first build will take some time as it downloads necessary AI models (>1GB). This is a one-time setup.

3. Access the application:

• Frontend: http://localhost:3000
• Backend API: http://localhost:8000
• API Documentation: http://localhost:8000/docs

Usage

Training a Custom Voice

1. From the Web Interface:

   • Go to http://localhost:3000
   • Switch to the "Train Custom Voice" tab
   • Enter a name for your voice
   • Record a clear audio file of your voice reading the provided text
   • Click "Train Voice"
   • Wait for the training to complete (usually takes 1-2 minutes)

2. From Command Line (Advance): Go to server directory. Make sure you create and activate a virtual environment and install dependencies by running pip install -r requirements.txt. Now follow the steps:

   # 1. First, process your audio file
   python generate_voice.py --input_file path/to/your/audio.wav --output_dir voices/your_voice_name
   
   # 2. Generate voice embeddings
   python save_embeddings.py --voice_dir voices/your_voice_name
   
   # 3. Cache voice latents for faster generation
   python cache_voice_latents.py --voice_dir voices/your_voice_name

Tips for best results:

• Use high-quality audio with minimal background noise
• Record in a quiet environment
• Speak clearly and at a natural pace
• Aim for at least 120 seconds of audio

Generating Speech

1. From the Web Interface:

   • Go to http://localhost:3000
   • Select a trained voice from the dropdown
   • Enter or paste the text you want to convert to speech
   • Toggle "High Quality" if desired (slower but better quality)
   • Click "Generate Speech"
   • Once complete, use the audio player to listen or download the generated audio

2. Using the API directly:

   curl -X POST http://localhost:8000/generate-speech \
     -H "Content-Type: application/json" \
     -d '{
       "text": "Your text here",
       "voice_name": "your_voice_name",
       "high_quality": false
     }'

Voice Directory Structure

Each voice in the voices directory should have the following structure:

voices/
└── your_voice_name/
    ├── original.wav         # Original audio file
    ├── chunks               # Processed audio chunks
    ├── voice_latents.pth    # Cached voice latents
    └── embeddings.pt        # Voice embeddings

API Endpoints

• POST /create-voice: Train a new voice
• GET /voices: List all available voices
• POST /generate-speech: Generate speech from text
• GET /audio/{generation_id}/{part}: Get generated audio file

Troubleshooting

1. If the server is slow on first request:

   • This is normal as models are being loaded
   • Subsequent requests will be faster

2. If voice training fails:

   • Ensure audio is clear and has minimal background noise
   • Try recording a longer sample
   • Check if the audio format is supported (WAV recommended)

3. If speech generation is stuck:

   • Check server logs using docker logs vocazee-server-1
   • Ensure the voice model exists and is properly trained
   • Try with a shorter text first

License

MIT License