A HMM-based integrated method for speaker-independent speech recognition

In this paper a new integrated method for speaker-independent speech recognition is proposed. This method incorporates three models based on the HMM, i.e., the semi-continuous hidden Markov model, the simple trajectory model and the Markov chain model to recognize an input utterance. The training procedure and recognition strategy as well as related computation problems are described in detail in this paper. Experiments on a Mandarin speech database CIDS showed that the proposed method can significantly decrease the error rate of the speaker independent (SI) system and it is a promising method to improve SI recognition systems employing existing techniques. The idea given in this paper provides a new way for speech recognition