A model of consonant perception extracts modulation power from gammatone-filtered speech envelopes.
Evidence
The subband power P, in dB, was calculated using 22 fourth-order gammatone filters with equivalent rectangular bandwidths. The center frequencies were spaced on a third-octave grid, covering a range from 63 Hz to 8 kHz. The Hilbert envelope of each filter output was extracted and low-pass filtered using a first-order Butterworth filter with a cut-off frequency of 150 Hz. The envelopes were down-sampled to a sampling rate of 1050 Hz. The modulation power Pmod , in dB, was obtained using the subband envelope extraction described above, followed by a modulation filterbank consisting of 3 second-order band-pass filters (center frequencies of 4, 8, and 16 Hz) in parallel with one third-order low-pass filter (cut-off frequency of 2 Hz). — Dijk, Pim van, Physiology, Psychoacoustics and…, 5602
Categorization
Category: Definitional Content
- DEFINE: Defines the full signal-processing pipeline (gammatone filters to modulation filterbank) that produces subband power and modulation power representations.
Related
- Speech recognition models can derive subband and modulation power from 22 gammatone filters (similarity 0.92)
- Speech perception models decompose subband signals into acoustic and modulation envelopes (similarity 0.91)
- Gammatone filter banks model human auditory filtering using measured nerve responses (similarity 0.84)
- Modulation power is captured by filters tuned to 2, 4, 8, and 16 Hz (similarity 0.84)