Talkscriber integration

[robosina]

Add Talkscriber Plugin for LiveKit Agents

Overview

This PR introduces a new plugin that integrates Talkscriber's voice AI services into LiveKit Agents, providing both Speech-to-Text (STT) and Text-to-Speech (TTS) capabilities.

Features

Speech-to-Text (STT)

• ✅ Real-time streaming transcription via WebSocket
• ✅ Interim and final transcript support
• ✅ Multilingual support with automatic language detection
• ✅ Smart turn detection using ML model for better endpoint detection
• ✅ Translation capabilities
• ✅ Configurable turn detection timeout

Text-to-Speech (TTS)

• ✅ High-quality voice synthesis
• ✅ Multiple voice options
• ✅ Audio buffering control for latency optimization
• ✅ Connection pooling for efficient resource management

Configuration & Deployment

• ✅ Environment variable support for flexible deployment:
  • STT_SERVER_HOST, STT_SERVER_PORT, STT_SERVER_USE_SSL
  • TTS_SERVER_HOST, TTS_SERVER_PORT, TTS_SERVER_USE_SSL
• ✅ API key authentication via TALKSCRIBER_API_KEY

Installation

pip install livekit-plugins-talkscriber

Or with LiveKit Agents:

pip install "livekit-agents[talkscriber]"

Usage Example

from livekit.agents import VoiceAssistant
from livekit.plugins import talkscriber

# Create STT instance
stt = talkscriber.STT(
    language="en",
    interim_results=True,
    enable_turn_detection=True,
)

# Create TTS instance
tts = talkscriber.TTS(
    voice="alloy",
    speed=1.0,
    pitch=0.0,
)

# Use in your agent
assistant = VoiceAssistant(
    stt=stt,
    tts=tts,
    # ... other configuration
)

Technical Details

Dependencies

• livekit-agents[codecs]>=1.2.14
• numpy>=1.26
• aiohttp
• websockets
• loguru

Architecture

• WebSocket-based real-time communication for low latency
• REST API fallback for chunked synthesis
• Proper error handling and reconnection logic
• Efficient audio buffer management
• Type-safe implementation with modern Python type hints

Testing

• ✅ Linter errors fixed (type annotations, bare except clauses, etc.)
• ✅ Environment variable configuration tested
• ✅ Follows LiveKit plugin architecture and conventions

Files Changed

• New plugin: livekit-plugins/livekit-plugins-talkscriber/
  • livekit/plugins/talkscriber/stt.py (629 lines)
  • livekit/plugins/talkscriber/tts.py (657 lines)
  • livekit/plugins/talkscriber/__init__.py
  • livekit/plugins/talkscriber/log.py
  • livekit/plugins/talkscriber/version.py
  • pyproject.toml
  • README.md
  • CHANGELOG.md
• Updated: pyproject.toml (added plugin to workspace)
• Updated: livekit-agents/pyproject.toml (added optional dependency)

Breaking Changes

None - This is a new plugin addition.

Related Links

• Talkscriber Client: https://github.com/Talkscriber/ts-client
• LiveKit Docs: https://docs.livekit.io

Checklist

• Code follows LiveKit plugin conventions
• Documentation (README) included
• Type annotations added
• Linter errors resolved
• Environment variable support for flexible deployment
• Error handling and logging implemented
• Compatible with Python 3.9+

---

This plugin expands LiveKit's voice AI capabilities by adding Talkscriber as a supported provider for both STT and TTS, giving developers more options for building real-time voice applications.

[CLAassistant]

All committers have signed the CLA.