Speech Recognition Of Two-word Chinese Vocabulary By Applying Fourier Transform To The Spectrogram

This paperproposes a Fourier transform of the broad-band andnarrow-bandspectrogram.Observing the eigenvalues Obtained by segmentationprojection frequency to domain image and fusion of the broad-band and narrow-band spectrogram eigenvalues,forming a new recognition algorithm of two-word Chinese vocabulary.The algorithm does not use the previous speech recognition algorithm to recognize speech signal frame by frame,but uses the whole characteristic of spectrogram to achieve speech recognition integrity,which can highlight the overall time-frequency characteristics of speech signal.This method takes spectrogram as a visual image.Image processing technology extract speech recognition parametersto achieve speech recognition.Becausespectrogram speech characteristics reflectedin the texture structure,and image texture structure is more readily describedbytheimageof frequency domain.In this paper,Fourier transform is applied twice to the broad-bandand narrow-band spectrogram to convert spatial domain tofrequencydomain.Recognition of two-word Chinese vocabulary is achieved.The algorithm is simulated by MATLAB R2013 a in this paper.First,the recordedspeech samples are preprocessed with Cool Edit Pro2.0 software,andquantizingandnormalizingthem.Then,usingMATLABR2013asoftwaretoprogramandconstructthebroad-bandand narrow-band spectrogram byFouriertime-frequencyanalysis.Fouriertransformtwicetogetthe image frequenc y domain,and carry on the binarydoublingwidth lineprojection andcolumnprojection.The sup port vector machine(SVM)isconsidered as classifierforrecognizingspeech of two-word Chine se vocabulary.Theresults using this methodshow aremarkablerecognition rate of 96.8% for sp ecificand 98.8% for non-specific.The proposedmethodprovides a new research ideafor theover al two-word Chinesevocabularyrecognition.Because wavelet transform is a time window and frequency window can change the time-frequency analysis method.Therefore,this paperattempts toconstruct thewaveletspectrogramfor recognizing speechof two-word Chinese vocabulary.As the recording of a large number of samples of the work is more cumbersome,so wewant toachieve thenon-specificspeechrecognitionthrougha single template.But in the actual process encountered a variety of problems,the experimental results are not ideal,still need to do further research and discussionfollow-up.