LiveKit Kyutai TTS Plugin

A high-quality text-to-speech plugin for LiveKit agents integrating Kyutai TTS with streaming implementation for real-time voice synthesis.

Features

• High-Quality Audio: Premium voice synthesis with natural-sounding speech
• Streaming Implementation: Real-time audio generation with low-latency streaming
• Excellent Performance: ~260ms time-to-first-byte (TTFB) on RTX 4090

Requirements

• LiveKit Agents v1.2 or higher
• Kyutai TTS server instance

Installation

1. Clone or download this plugin into your LiveKit-based agents project root directory
2. Set up the Kyutai TTS server using taresh18/delayed-streams-modeling
3. Ensure your server is running and accessible

Server Setup

Use the delayed streams modeling server for optimal performance:

Repository: taresh18/delayed-streams-modeling

Follow the setup instructions in the repository to get your Kyutai TTS server running with streaming capabilities.

Usage

Initialize your agent session with the KyutaiTTS plugin:

from your_plugin_path import kyutTTS

session = AgentSession(
    # ... other configuration
    tts=kyutTTS(
        base_url="<kyutai_server_url>",  # e.g., "http://localhost:8000"
        voice="expresso/ex04-ex02_happy_001_channel2_140s.wav",  # voice file path
    )
)

Performance Metrics

• Latency: ~260ms TTFB on RTX 4090 GPU
• Quality: High-quality voice synthesis with natural prosody
• Streaming: Real-time audio generation and playback
• Efficiency: Optimized for production use with streaming implementation