DSP Speech-Recognition System Based on Mel-Frequency Cepstrum Coefficients
Bo Xu · Communications technology · 2007
By regard Mel-Frequency Cestrum Coefficients as character parameters for speech-recognition, and using Dynamic Time Warping recognition algorithm, the speech-recognition with the kernel of Digital Signal Processor TMS320VC549 is implemented. 24 grades Mel-Frequency Cestrum Coefficients are used as character parameters. Software system directly programs the DSP, which use MFCC stay-point algorithm to simulate character parameters, use DTW algorithm to search match paths. Hardware system uses the single-chip AT89C51 to realize communication with TMS320VC549, and is able to produce even higher recognition accuracy in presence of channel noise and spectral distortion.