Nonlinear prediction with neural nets in ADPCM

TL;DR

This paper investigates the application of neural network-based nonlinear predictive models in speech waveform coding, demonstrating improved signal quality metrics over traditional methods.

Contribution

It introduces a novel neural network-based nonlinear prediction scheme for speech coding, showing practical implementation and performance gains.

Findings

Improved SEGSNR by 1-2 dB with the proposed model

Effective in adaptive quantization from 2 to 5 bits

Highlights potential for nonlinear models in real speech coding applications

Abstract

In the last years there has been a growing interest for nonlinear speech models. Several works have been published revealing the better performance of nonlinear techniques, but little attention has been dedicated to the implementation of the nonlinear model into real applications. This work is focused on the study of the behaviour of a nonlinear predictive model based on neural nets, in a speech waveform coder. Our novel scheme obtains an improvement in SEGSNR between 1 and 2 dB for an adaptive quantization ranging from 2 to 5 bits.