VLLM & TTS Playground

A personal playground for testing and experimenting with Voice Language Models (VLLM) and Text-to-Speech (TTS) synthesis using the Maya1 model and SNAC audio codec.

Overview

This project uses:

• Maya1: An open-source voice AI model for generating emotional speech
• SNAC: A neural audio codec for decoding hierarchical audio tokens to waveforms
• Transformers: For model loading and inference

Features

• Generate speech with natural language voice descriptions
• Control voice characteristics (tone, pitch, pacing, emotion)
• Support for emotional tags like <angry>, <chuckle>, etc.
• 24kHz audio output

Setup

1. Create a virtual environment

python -m venv venv

Activate the virtual environment:

• Windows: venv\Scripts\activate
• Linux/Mac: source venv/bin/activate

2. Install dependencies

pip install -r requirements.txt

3. First-time model download

Note: On first run, the script will download the Maya1 model (~7GB) and SNAC decoder. This is a one-time download and may take several minutes depending on your internet connection.

4. Run the script

python main.py

Generated audio will be saved to outputs/output.wav.

Usage

Edit the voice description and text in main.py:

description = "Hissing, cold male voice, no clear age, sinister and menacing tone. High pitch, slow, deliberate pacing."
text = "There is no good and evil. <angry> There is only power, and those too weak to seek it. <chuckle> Foolish boy."

Run the script to generate speech with your custom parameters.

Requirements

• Python 3.8+
• CUDA-compatible GPU (recommended for faster inference)
• ~10GB disk space for models
• See requirements.txt for Python dependencies

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Maya Research for the Maya1 model
• SNAC audio codec