Fig 3. Signal processing steps prior to Temporal (rate) PCA.

The original acoustic signal (a) is passed through a low-dimensional spectral filterbank, yielding a small set of core spectral band outputs, (b). The parameters of the low-dimensional spectral filterbank were determined in the prior Spectral PCA procedure. The envelope, (c), is extracted from each spectral band output using the Hilbert transform. Each envelope is further passed through a high-dimensional modulation filterbank, yielding a set of high-dimensional modulation rate envelopes, (d). This rate-filtering is performed for each spectral band envelope, but for simplicity, only the modulation rate envelopes from a single spectral band are shown in this figure. Finally, the power profiles of the modulation rate envelopes (bold blue line) are entered into a temporal (rate) PCA to identify patterns of covariation across modulation rates.