⚠️ Deprecated. This plugin is deprecated and will be removed in a future release. Use livekit.agents.inference.TurnDetector instead — it ships with livekit-agents, requires no additional install, and replaces both the English and Multilingual text-based models with a unified audio end-of-turn detector.

This plugin introduces end-of-turn detection for LiveKit Agents using a custom open-weight model to determine when a user has finished speaking.

Traditional voice agents use VAD (voice activity detection) for end-of-turn detection. However, VAD models lack language understanding, often causing false positives where the agent interrupts the user before they finish speaking.

By leveraging a language model specifically trained for this task, this plugin offers a more accurate and robust method for detecting end-of-turns.

See https://docs.livekit.io/agents/build/turns/turn-detector/ for more information.

Usage

The recommended replacement is TurnDetector, available from livekit-agents directly:

from livekit.agents.inference import TurnDetector

session = AgentSession(
    ...
    turn_detection=TurnDetector(),
)

Usage with RealtimeModel

TurnDetector works with speech-to-speech models such as OpenAI's Realtime API:

session = AgentSession(
    ...
    llm=openai.realtime.RealtimeModel(),
    turn_detection=TurnDetector(),
)

Running your agent

This plugin requires model files. Before starting your agent for the first time, or when building Docker images for deployment, run the following command to download the model files:

python -m livekit.agents download-files

Downloaded model files

Model files are downloaded to and loaded from the location specified by the HF_HUB_CACHE environment variable. If not set, this defaults to $HF_HOME/hub (typically ~/.cache/huggingface/hub).

For offline deployment, download the model files first while connected to the internet, then copy the cache directory to your deployment environment.

Model system requirements

The end-of-turn model is optimized to run on CPUs with modest system requirements. It is designed to run on the same server hosting your agents.

The model requires <500MB of RAM and runs within a shared inference server, supporting multiple concurrent sessions.

License

The plugin source code is licensed under the Apache-2.0 license.

The end-of-turn model is licensed under the LiveKit Model License.