Table 2. Correlation coefficients for different feature sets.
L2 on features | L2 on reduced features | Gaussian kernels on reduced features | |
Fourier-based Spectrum | - | - | 0.69 |
Auditory Spectrum | 0.473 | - | 0.739 |
Global Spectro-temporal Modulations | 0.509 | - | 0.701 |
Separable Spectro-temporal Modulations | 0.561 | 0.561 | 0.830 |
Full cortical Model | 0.611 | 0.607 | 0.944 |
Neurophysiological STRFs | - | - | 0.73 |
Each row represents the correlation coefficient between the model and human similarity matrix using a direct Euclidian distance the specific feature itself (left); on a reduced dimension of the features (middle column) or using the Gaussian kernel distance (right column). Auditory Spectrum: time-average of the cochlear filterbank; Global Spectro-temporal Modulations: model STRFs averaged in time and frequency; Separable Spectro-temporal modulations: model STRFs averaged separately in rate and scale, and then in time; Full cortical model: STRFs averaged in time; Neurophysiological STRFs: as the full cortical model, but with STRFs collected in primary auditory cortex of ferrets.