Global maximum likelihood voice decoding
Richard A. Dean · International Conference on Acoustics, Speech, and Signal Processing · 2002
Research on a technique for improving speech coder performance in the presence of errors is presented. It is shown that significant performance advantages can be achieved by combining conventional maximum-likelihood decoding with the context within the speech signal to make global decisions. The hidden Markov model (HMM) is integrated into a combined channel and speech decoder. A theoretical development of a global maximum-likelihood decoder is presented. Results of testing in channel errors are presented, demonstrating enhanced error performance.>