Speech recognition models can derive subband and modulation power from 22 gammatone filters
Evidence
Although gammatone filters provide a good approximation of the human auditory filters, they suffer from a few shortcomings. They cannot emulate the level-dependent characteristics of the auditory filters. In addition, the impulse response of a gammatone filter has a relatively slow onset, which brings on problems when modelling phenomena involving temporally short sounds, such as the precedence effect. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 1614 (
#complicates, via linked claim)
Categorization
- XREF: connects to: Gammatone filters fail on level-dependence and fast-onset sounds — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 1614; The envelope of each auditory filter output is processed by a modulation filterbank; Auditory filter bandwidth scales nonlinearly with center frequency (+1 more)
Related
- Gammatone filters match the auditory nerve’s measured impulse response (similarity 0.93)
- Gammatone filters fail on level-dependence and fast-onset sounds (similarity 0.90)
- Gammatone filter banks model human auditory filtering using measured nerve responses (similarity 0.89)
- Speech perception models decompose subband signals into acoustic and modulation envelopes (similarity 0.88)
- A model of consonant perception extracts modulation power from gammatone-filtered speech envelopes. (similarity 0.88)
- DRNL filter outputs smear temporal fine structure into double impulses (similarity 0.84)
- Modulation power is captured by filters tuned to 2, 4, 8, and 16 Hz (similarity 0.83)
- Auditory filter bandwidth scales nonlinearly with center frequency (similarity 0.80)
- Filter banks split broadband signals into sub-bands much like cochlear hearing does (similarity 0.80)
- Filtering in the frequency domain is simply multiplicative on spectra (similarity 0.79)
- The envelope of each auditory filter output is processed by a modulation filterbank (similarity 0.79)
- The auditory system acts as a band-pass filter at each signal frequency. (similarity 0.78)
- Frequency-modulation detection relies on the steep lower excitation slope at 27 dB per Bark (similarity 0.78)
- Auditory filters respond asymmetrically, favoring frequencies below center (similarity 0.78)
Broader Connections
- Gammatone filters match the auditory nerve’s measured impulse response
#corroborates(0.90) - The envelope of each auditory filter output is processed by a modulation filterbank
#expands(0.82) - Gammatone filters fail on level-dependence and fast-onset sounds
#complicates(0.78) — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 1614 - Auditory filter bandwidth scales nonlinearly with center frequency
#expands(0.72)