Fact · Speech recognition
The docstring of OpenAI's open-source Whisper code says a segment is treated as silent when its no-speech probability is above the threshold and its average log probability is below the log probability threshold.
Checked against its source on October 17, 2026
In the source's own words
“If the no_speech probability is higher than this value AND the average log probability over sampled tokens is below”
whisper/transcribe.py (openai/whisper, main branch), OpenAI, GitHub (openai/whisper), n.d., transcribe docstring.
Open the original- What it measures
- The check runs on each 30-second window, after the model has decoded it. It is a per-segment decision made by the decoder, not a timer on the audio.
- What it does not show
- The rule decides whether to skip a window that the model already decoded, so it does not tell a live app when a speaker has finished.