← All stories
● Covered by 1 source · 1 reportMedium impact1 neutral

RealtimeVAD detects end-of-speech up to 110ms faster than Silero VAD

🔄 Updated 8d ago
New to BrevFeed? We gather this story from every outlet covering it into one summary — ranked by real-world impact, not just the latest headline — so you never miss what matters. What is BrevFeed? →

Key points

  • Detects end-of-speech 100-110 ms earlier than Silero VAD.
  • Combines Silero VAD with real-time audio level monitoring.
  • Fires 'start', 'pause', and 'end' events.
  • Runs on one CPU thread, requires numpy and onnxruntime.

Faster End-of-Speech Detection

RealtimeVAD is a new voice activity detection (VAD) tool designed to identify when a speaker has stopped talking more quickly than existing solutions. It achieves this by integrating the Silero VAD for speech detection with an additional, faster check that monitors audio levels every 10 milliseconds. This dual approach allows RealtimeVAD to detect the cessation of speech up to 110 milliseconds earlier than Silero VAD alone.

Impact on Voice Assistants

The primary benefit of RealtimeVAD is reducing latency in voice assistant workflows. Processes such as speech recognition, large language model (LLM) calls, and voice output typically wait for a VAD to confirm the user has finished speaking. By providing an 'end' event sooner, RealtimeVAD can accelerate these subsequent steps, potentially leading to more responsive conversational AI systems.

Technical Implementation and Performance

RealtimeVAD processes 16 kHz mono audio and outputs 'start', 'pause', and 'end' events. It reports the end of speech a median of 100 ms before Silero's probability drops, and 206 ms before Silero's VADIterator reports the end, based on tests with 3,279 real turns. The tool runs on a single CPU thread, faster than real-time, and requires only numpy and onnxruntime, with the Silero model bundled without needing PyTorch.

Considerations and Current Status

While offering faster detection, RealtimeVAD also fires more frequently in short gaps between words, approximately ten times more often than Silero. Additionally, its lead in detection can diminish with quiet speakers on noisy lines. The tool is currently in an alpha (0.0.5) status, indicating that its API may still undergo changes.

✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →

The daily brief

One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.

One email a day. Unsubscribe in one click, any time.

Today's brief

Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.

~4 min · 3 stories · Nov 21

▶ Play today's brief Listen on Spotify

New every morning, and the back catalogue is archived by date.

Reporting from

RealtimeVAD is a new voice activity detection (VAD) tool that identifies the end of speech significantly faster than the Silero VAD, by combining Silero's speech detection with a real-time audio level comparison. This earlier detection can reduce latency in voice assistant interactions, as subsequent processes like speech recognition and LLM calls can begin sooner.