BMLD-based prediction models operate per spectro-temporal unit, not per channel

Evidence

The key feature to note about the hearing modeling spectrogram is that the sound is analyzed via a bank of filters that model the auditory filters in terms of their shape and bandwidth. As a result, the frequency axis is based on the ERB (equivalent rectangular bandwidth) scale itself, since the output from each filter is plotted with an equal distance on the frequency axis of the spectrogram. — David M. Howard, Jamie Angus, Acoustics and Psychoacoustics, 1501 (#expands, via linked claim)

Categorization

• XREF: connects to: Spectro-temporal SNR units span 23 ms and one ERB bandwidth; Hearing-modeling spectrograms use an ERB frequency axis from modeled auditory filters. — David M. Howard, Jamie Angus, Acoustics and Psychoacoustics, 1501

Broader Connections

• Spectro-temporal SNR units span 23 ms and one ERB bandwidth #expands (0.85)
• Hearing-modeling spectrograms use an ERB frequency axis from modeled auditory filters. #expands (0.75) — David M. Howard, Jamie Angus, Acoustics and Psychoacoustics, 1501