Estimating unreliable packets in subband coded speech

Abstract

A robust digital mobile radio communication system using 12-kb/s embedded subband speech coding and rate-compatible punctured convolutional (RCPC) codes with generalized Viterbi decoding for combined error detection and correction is examined. Unreliable subband-coded speech side-information is identified and replaced using various estimation strategies. By using a realistic simulation model with 4-DPSK (differential phase-shift keying) modulation over a mobile channel, improvements of 1 to 4.4 dB in recovered speech segmental SNR (signal-to-noise ratio) are obtained compared to a system with only channel-based error correction. Informal subjective tests support these gains.