ISCA Archive RSR 1997
ISCA Archive RSR 1997

Automatic speech recognition for time-varying channels

Keith M. Pouting

Any mismatch between training and test conditions can cause difficulty for current automatic speech recognition systems. This paper presents a simple tracking algorithm to compensate for changing transmission channel characteristics during recognition. The algorithm is designed to work in the context of a continuous single-pass speech recognition system and to be robust to speech recognition errors. Experiments show that it can successfully compensate for mismatched microphones and microphone positions.