The Research On Spectrogram Of A Particular Group Of Small-Vocabulary Recognition Algorithm

As one of the most common methods of speech recognition, speech-dependentrecognition has been applied in numerous applications of vehicle speech recognition,such as controlling basic motions of a vehicle, and it can improve the accuracy ofspeech recognition and reduce the response time.In this paper, we summarize common algorithms of speech recognition, andpresent a novel speaker-dependent speech recognition algorithm based on spectrogram.Compared with frame detection of speech signals in traditional speech recognitionalgorithms, speech recognition based on overall characteristics of spectrogram isproposed. It can highlight overall time-frequency characteristics, and introducemorphological image processing technique into the field of speech recognition.The proposed method is implemented under the environment of MATLAB7.1.First, transform recorded speech signals to Fourier spectrogram, and prepossess thespectrogram by morphological image processing techniques, including smoothing,normalization, binarization, etc; then, based on the characteristics of speech signals inthe spectrogram, make image line cross-correlation by an improved method, get themaximum coefficient of the line cross-correlation as the first eigen-parameter, and countthe number of formants, which are displayed in the spectrogram, as the secondeigen-parameter; finally, analyze the obtained data by support vector machine (SVM),and determine the recognition rate of speaker-dependent speech recognition. The paperattempts to recognize the Chinese control commands of vehicle speech recognitionsystem.