Implementation of a Voice Dialing System Based On HTK

摘要

Putting speech recognition technology into practice, a speaker-independent digital voice dialing system is realized in an embedded system. In the speech recognition process, the Mel Frequency Cepstrum Coefficients (MFCC) is used as the feature parameter, the Continuous Hidden Markov Model (CHMM) is chosen as the training and recognition model and realized by HTK. This speaker-independent digit recognition is tested on embedded system. The result indicates the accuracy of this system reaches up to 98% to speaker-independent case and it takes 3.55 seconds. The speech recognition is implemented in software, which is easier transplant to different platform and has potential application prospects.