Processing Steps

1. Multi-Resolution STFT (MR-STFT)
   (FFTs of different lengths combined)
   • High time resolution at high frequencies
   • High frequency resolution at low frequencies
2. Compute the Excitation Pattern by nonuniformly resampling the MR-STFT frequency axis
   • Spectral sampling interval $\approx$ critical bandwidth of hearing (Bark or ERB spacing)
   • Interpolation kernel is shaped like an auditory filter (frequency and presentation-level dependent)
   • Nonuniform frequency sampling can be accomplished by frequency warping (e.g., to Bark or ERB frequency scale) followed by uniform sampling¹