Hidden Markov model environmental compensation for automatic speech recognition on hand-held mobile devices

This paper is concerned with applying hidden Markov model compensation techniques for improving the performance of automatic speech recognition (ASR) based services on hand-held mobile devices. The implementation and evaluation of an ASR based task for a mobile, hand–held device is presented, along with a set of compensation techniques that are used to compensate speaker independent hidden Markov models with respect to environmental and transducer variability. A technique for combined environment/transducer compensation is shown to significantly reduce the effects of environmental mismatch. The overall performance degradation with respect to clean conditions was reduced from 41.7 percent to 10.4 percent for speech spoken through a far–field microphone in an office environment, and from 79.2 percent to 39.8 percent for the same transducer in a noisy cafeteria environment.