Suppression Rule for Speech Recognition Friendly Noise Suppressors

Audio signal enhancement often involves the application of a time-varying filter, or suppression rule, to the frequency-domain transform of a corrupted signal. Known approaches use rules derived under Gaussian models and interpret them as spectral estimators in a Bayesian statistical framework. While this mathematical approach provides rules that satisfy  certain optimization criteria these rules are not optimal when the enhanced signal is for a speech recognition engine. In this paper we present the approach and the results for creation of a speech recognition friendly suppression rule. The described approach increases the average speech recognition rate in Aurora 2 tests from 52.47% to 77.69% while maintaining performance for low noise utterances.