Entropy-constrained spike modulus quantization in a bio-inspired universal audio coder

We propose an optimal quantizer for the moduli (spike amplitudes) of thematching pursuit on gammatone/gammachirp kernels that have been shown previously to be very efficient for audio coding. The quantizer optimization is performed by Genetic Algorithm (GA). We show that the optimal quantization values found by GA can be approximated by a “piecewise uniform” quantizer. The latter approximation is faster and has less overhead (i.e., there is no need for codebook transmission). Based on perceptual evaluation of audio quality, transparent quality is obtained for both the optimal and piecewise uniform quantization approaches. We also studied the performance of in-loop quantization vs. out-of-loop quantization when computational cost can be afforded.