The implementation of continuous speech recognition system based on LabVIEW

Based on software of LABVIEW, the original sound signal was usually obtained by soundcard. The process to get clean sound signal was pre-emphasis, wavelet de-noising, adding window and port detection. Continuous speech recognition system was realized by VQ (vector quantization) and HMM (hidden Markov model) for training and recognition. The Mel frequency cepstrum coefficient and its difference were used as speech recognition characteristic parameter, and ameliorative Viterbi-Beam recognition algorithm is used in the system. Experimental data indicates that ameliorative Viterbi-Beam recognition algorithm decreased the calculation of the system and also enhanced its running speed. Furthermore, the successful rate of speech recognition is about 90%, and the whole system can reach the factual applied request.