Envelope modeling for speech and audio processing using distribution quantization

Envelope models are common in speech and audio processing: for example, linear prediction is used for modeling the spectral envelope of speech, whereas audio coders use scale factor bands for perceptual masking models. In this work we introduce an envelope model called distribution quantizer (DQ), with the objective of combining the accuracy of linear prediction and the flexibility of scale factor bands. We evaluate the performance of envelope models with respect to their ability to reduce entropy as well as their correlation to the original signal magnitude. The experiments show that in terms of entropy, distribution quantization and linear prediction are comparable, whereas for correlation, distribution quantization is better. Furthermore the coefficients of distribution quantization are independent and thus more flexible and easier to quantize than linear predictive coefficients.