livekit-plugins-tenvad 1.0.6

Agent Framework Plugin for TEN VAD

LiveKit Plugins – TEN VAD

livekit-plugins-tenvad provides seamless integration of the TEN-framework/ten-vad voice activity detection (VAD) plugin into the LiveKit ecosystem.

This plugin enables real-time speech activity detection with low-latency inference, optimized for streaming, conversational AI, and livekit-agents integration.

✨ Features

• 🔌 LiveKit plugin integration — plug-and-play support for LiveKit workflows
• 🤖 Compatible with livekit-agents — extend agents with real-time VAD capabilities
• 🎤 Accurate voice activity detection powered by TEN VAD
• ⚡ Low-latency inference (~0.17ms avg per frame) suitable for real-time use
• 📊 Benchmark validated against Silero VAD (faster and more continuous speech detection)
• 🛠️ Configurable & extensible within the LiveKit plugin system

🔧 Installation

# from PyPI
uv pip install livekit-plugins-tenvad

# from source
uv pip install git+https://github.com/dangvansam/livekit-plugins-tenvad.git

🔌 Usage

from livekit.plugins import tenvad

vad = tenvad.VAD.load(
    activation_threshold=0.5,
    min_silence_duration=0.3,
    min_speech_duration=0.15,
    max_buffered_speech=30,
    prefix_padding_duration=0.1,
    padding_duration=0.1
)

📊 Run Benchmark

git clone https://github.com/dangvansam/livekit-plugins-tenvad.git

cd livekit-plugins-tenvad

# install dependencies for testing
uv pip install .[test]

python test/benchmark.py test/sample.wav outputs silero,ten

Benchmark Results

Metric	Silero VAD	TEN VAD
Speech segments	95	41
Total speech	19.01s (13.0%)	114.98s (78.8%)
Avg inference time	0.22ms	0.17ms
Min inference time	0.18ms	0.14ms
Max inference time	9.76ms	0.78ms

Highlights:

• TEN VAD is ~1.27× faster per frame
• Detects longer continuous speech compared to Silero
• Provides lower latency with fewer false segment splits

Visualizations

Long audio

Short audio

Citations

@misc{TEN VAD,
  author       = {TEN Team},
  title        = {TEN VAD: A Low-Latency, Lightweight and High-Performance Streaming Voice Activity Detector (VAD)},
  year         = {2025},
  publisher    = {GitHub},
  journal      = {GitHub repository},
  howpublished = {\url{https://github.com/TEN-framework/ten-vad.git}},
  email        = {[e-mail]}
}