On a study of decreasing the processing time in the speech recognition system using the HMM (hidden Markov model) algorithm

The study of isolated-word speech recognition is now progressing and the recognizers have reached commercialization. Speech recognition has now been applied to communication systems in our environment. However, speech recognition performance on a large vocabulary is insufficient for commercial use, since there are many problems which are due to difficulties of speech database setup on a large vocabulary and the increase of processing time. Therefore research into these problems has been progressed. In this paper, we propose an improved method for the processing time of the recognition process. The variation of the speech signal may be predicted through short time analysis. Especially, in the case of voiced speech, the waveform is repeated with the same period. In this paper, we discuss the reduction of both processing time for speech database setup and speech recognition time if we use this characteristic of the speech signal.