Fact · Speech recognition
80Mel-frequency filters
The log-Mel spectrogram function in the Whisper code uses 80 Mel-frequency filters by default, and it supports only 80 or 128 filters.
Checked against its source on October 17, 2026
In the source's own words
“The number of Mel-frequency filters, only 80 and 128 are supported”
whisper/audio.py: the audio constants and log-Mel code in the openai/whisper repository, OpenAI, GitHub, openai/whisper repository (main branch), 2022, log_mel_spectrogram, parameter n_mels.
Open the original- What it measures
- The function takes a 16 kHz waveform, applies a short-time Fourier transform with a 400-sample window and a 160-sample hop, multiplies the power by a Mel filter bank, and takes the log. The default is set in the function signature as n_mels: int = 80.
- What it does not show
- Describes the open-source Whisper code, not every recognizer. Other systems choose different numbers of Mel bands, such as the 128 option that Whisper also supports.