livekit-plugins-turn-detector

What it is and what it does

This package adds end-of-turn detection to LiveKit Agents by using a custom language model trained specifically to recognize when a user has finished speaking, rather than relying on voice activity detection (VAD) alone. VAD models often cause false positives and interrupt users mid-sentence; this plugin's language understanding approach is more accurate and robust for real conversational agents.

The package depends on transformers, onnxruntime, numpy, jinja2, and livekit-agents. It downloads and caches model files locally (under 500MB) and runs inference on CPU alongside your agent. However, the package is deprecated: LiveKit has integrated an improved TurnDetector directly into livekit-agents itself, which ships with no additional install and replaces both English and Multilingual models with a unified audio-based detector.

Use it for:

• Building voice agents that need to detect when a user finishes speaking without interrupting mid-sentence
• Replacing VAD-only turn detection in existing LiveKit Agents to reduce false-positive interruptions
• Deploying multilingual voice agents where language understanding improves turn-taking accuracy
• Running speech-to-speech models (e.g., OpenAI Realtime API) with more intelligent end-of-turn signals

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Detects end-of-turn in voice conversations for LiveKit Agents using a language model trained for this task, replacing simpler voice activity detection with more accurate interruption prevention.

No—do not install this package. It is deprecated and will be removed in a future release. Use the built-in TurnDetector from livekit-agents instead, which requires no additional install, ships with the main package, and provides a unified audio-based detector that replaces this plugin's functionality. If you are already using this plugin, plan to migrate to the built-in alternative.

Install

livekit-plugins-turn-detector on PyPI

pip install livekit-plugins-turn-detector

uv add livekit-plugins-turn-detector

poetry add livekit-plugins-turn-detector

Installing livekit-plugins-turn-detector

Before you install

Low friction install with a pure-Python wheel. Actively maintained as of 1 day ago with 13002 repository stars. However, the package is deprecated and will be removed in a future release; the built-in TurnDetector in livekit-agents is the recommended replacement.

License in practice

Apache-2.0 permissive license on the plugin source code. The bundled end-of-turn model uses the LiveKit Model License, which you should review separately if model redistribution or modification is relevant to your use case.

Quickstart

pip install livekit-plugins-turn-detector

from livekit.agents import TurnDetector
from livekit.agents import AgentSession

session = AgentSession(
    turn_detection=TurnDetector(),
)

Requires Python >=3.10.0. Model files must be downloaded via `python -m livekit.agents download-files` before first use; they are cached in the location specified by HF_HUB_CACHE environment variable (defaults to ~/.cache/huggingface/hub).

Verify before relying

• Whether the deprecated status affects your deployment timeline or long-term maintenance plans
• Exact performance difference between this plugin and the built-in TurnDetector replacement
• Whether the <500MB RAM requirement and CPU-only design fit your infrastructure

Package facts

Evidence: livekit_plugins_turn_detector-1.6.10-py3-none-any.whl

Keywords: ai, audio, livekit, realtime, video, voice, webrtc

Tags

More Artificial Intelligence packages

Further reading