LiveKit Plugins Turn Detector

LiveKit Plugins Turn Detector

LiveKit Proxy End of Speech Detection Plugin Based on Customized Model

0
  • Provide speech end detection based on language models to improve accuracy and robustness.
  • Compared with traditional VAD models, it reduces false positives and avoids agents interrupting before the end of the user's speech.
  • • Support integration with VoicePipeAgent, making it convenient for developers to use.
  • Model files are required, which can be downloaded through the command line.
  • Optimize to run on CPU with moderate system requirements.
  • Supports multiple concurrent sessions and is suitable for running on proxy servers.
  • Future versions will reduce CPU and memory requirements.

Product Details

LiveKit Plugin Turn Detector is a plugin for LiveKit Agents that introduces end-to-end end of speech detection by using a customized open weight model to determine when a user completes their speech. Compared to traditional Acoustic Activity Detection (VAD) models, this plugin utilizes a language model specifically trained for this task, providing a more accurate and robust method for detecting the end of speech. The current version only supports English and is not recommended for use in other languages.