Fact · Speech recognition
0.6no-speech probability threshold
In OpenAI's open-source Whisper command line, the --no_speech_threshold option defaults to 0.6.
Checked against its source on October 17, 2026
In the source's own words
“"--no_speech_threshold", type=optional_float, default=0.6, help="if the probability of the”
whisper/transcribe.py: decoding defaults and command-line options (openai/whisper), OpenAI, GitHub (openai/whisper), 2026, Command-line options, argparse definition of --no_speech_threshold.
Open the original- What it measures
- Whisper decodes audio in 30-second windows. The no-speech check runs after the model has decoded a window, to decide whether to skip that window as silence.
- What it does not show
- The check works on a window the model has already decoded. It does not tell a live app that the speaker has finished a sentence. Other tools may use different defaults.