Research Methods Of Speech Recognition Of Specific Two Words Chinese Vocabulary Based On Spectrogram

The human dream is to let the computer understand human language since the birth of computer thinking.With the rapid development of electronic products,more and more people are eager to get rid of shackles of the keyboard instead of speech input for humanity's input.Especially the input of Chinese characters,has always been a major problem in computer applications.Therefore the use of Chinese speech man-machine interaction is becoming a very important research subject.As the use of a higher frequency commonly used in modern Chinese vocabulary words are 56008,of which the pentasyllable and five syllable above words are 162,the four syllable words are 5855,the three syllable words are 6459,the disyllabic words are 40351,and the monosyllabic words are 3181.So two syllable words are accounted for 72% of the proportion of all words,plays an important role in the common words.This paper selects 10 Chinese vocabulary of two words speech recognition algorithm,has a strong representation.Traditional voice analysis are based on a short smooth assumptions,time-frequency localization information of speech signal is obtained by using Fourier transform with fixed window,with a short speech frame as the basic unit for processing the segmentation method destroys the integrity of syllable carries information,to a certain extent,affects the effect of speech recognition.The speech recognition technology applied to the field of image processing,the speech spectrogram of two words Chinese vocabulary for feature analysis,extracted and used four methods for feature extraction of the spectrogram,with width row projection,column projection and two in width with row projection of the spectrogram,the 6 level wavelet packet decomposition of wideband and narrowband spectrogram is carried out by using two discrete DB4 wavelet bases respectively,calculates the energy level details of each layer,vertical detail and diagonal detail energy values.The extracted feature set of these four methods as feature vector recognition,using SVM as the classifier for recognition of two words Chinese vocabulary.The algorithm uses the overall characteristics of spectrogram word by word for speech recognition,to highlight the overall time-frequency characteristics of speech signals,according to the Chinese characteristics,each voice command is taken as an image for lexical study to ensure that the integrity of the statement,is helpful to improve the speech recognition accuracy and robustness of system.The image samples are noise free by means of image processing technology,although noise free speech files relative to noise free speech samples effect is very poor,but this article also tries and explores the system,systematically,and at the same time provides the important basis and clue for the further research of the speech enhancement method.