Non-linear predictive vector quantization of speech

TL;DR

This paper introduces a non-linear predictive vector quantizer for speech coding using neural networks, evaluates its design quality, and discusses potential improvements despite current limitations.

Contribution

It presents a novel non-linear PVQ approach for speech coding based on MLPs and proposes a method to assess quantizer design quality.

Findings

Non-linear PVQ does not outperform non-linear scalar predictor

Method to evaluate quantizer correlation exploitation

Potential for future improvements in PVQ design

Abstract

In this paper we propose a Non-Linear Predictive Vector quantizer (PVQ) for speech coding, based on Multi-Layer Perceptrons. We also propose a method to evaluate if a quantizer is well designed, and if it exploits the correlation between consecutive outputs. Although the results of the Non-linear PVQ do not improve the results of the non-linear scalar predictor, we check that there is some room for the PVQ improvement.