Study Of Feature Extraction In Speech Recognition

As the number of automobiles is increasing, the transportation environment is becoming moreand more complicated. It has become one of the major causes of traffic accidents that new driveshandle the cars incorrectly for their nervous or the lack of training. The crisis control method andautomobile active safety system are in absence nowadays. The breaking system based on speechcommand recognition technology will become one of the most effective methods to solve theproblem.In our study, we analyzed the characteristic of the speech command and found out thedifferences between the speech and speech command signal. We studied the peak frequency of thespeech command which was stable and used it to distinguish the speech command from speech signalof one identical speaker. Moreover, Mel-Frequency Cepstrum Coefficients was introduced to separatethe same words of different speakers. A speech command recognition method based on templatematching technology was proposed, the experiments showed that our method was effective. We alsoanalyzed the high-SNR spectra band segment method using spectral entropy. The experiment resultswere promising and will help broad the application of speech command recognition system in caractive safety.