Speech recognition in living rooms: Integrated speech enhancement and recognition system based on spatial, spectral and temporal modeling of sounds

摘要: Research on noise robust speech recognition has mainly focused dealing with relatively stationary that may differ from the conditions in most living environments. In this paper, we introduce a system can recognize presence of multiple rapidly time-varying sources as found typical family room. To deal such severe conditions, our exploits all available information about and noise; is spatial (directional), spectral temporal information. This realized model-based enhancement pre-processor, which consists two complementary elements, multi-channel speech-noise separation method information, followed by single channel algorithm uses long-term characteristics obtained clean examples. Moreover, to compensate for any mismatch remain between enhanced acoustic model, employs an adaptation technique combines conventional maximum likelihood linear regression dynamic adaptive compensation variance Gaussians model. Our proposed approaches human performance levels greatly improving audible quality substantially keyword accuracy.