Predictive coding with delayed decision.

This paper describes quantitative and qualitative improvements in 6400 b/s adaptive predictive voice coding when delayed decision is used to quantize the residual signal into one bit/ sample. Signal-to-quantizing noise curves as a function of the delay decision length provide quantitative indications of the improvements while listening tests of sentences processed with and without delayed decisions provide measures of qualitative performance. Although delayed decision does provide better waveform tracking and yields better than 2 dB improvements in signal-to-quantizing noise, auditory tests indicate that listeners cannot hear the improvement with decision delays of 5 or less. Consequently, the increased complexity of delayed decision with 6400 b/s adaptive predictive coding does not result in a useable improvement in voice transmission.