Research On Speech Recognition In Complex Noise Environment

The deepening of human-computer interaction requires automatic speech recognition(ASR)to be robust to the full range of real-world noise and other acoustic distorting conditions.The low accuracy of speech recognition in complex noise environment has aroused wide concern among scholars.Current methods can be roughly divided into three categories: feature based method,noise environment classification method,and speech enhancement method.In this essay,GFCC is used as speech feature,a method based on CNN and LSTM is designed,extract the training speech spectrum,and use the attention network to perform adaptive feature refinement.Then the attention map is multiplied with the input feature map to achieve speech recognition in a noisy environment.The main content of the thesis includes:(1)Basic tasks of speech signal processing.The processing tasks are classified into three categories,and the speech recognition problems related to this essay are introduced,and the acoustic models used in the three sub-problems including text recognition,voice print recognition and emotion recognition are expounded.(2)The processing method of speech recognition problem,the purpose of pre-filtering,pre-weighting,framing,endpoint detection and other operations of speech signals,and the characteristics of common speech signals and the performance of each feature in the noise environment,so as to introduce the influence of noise on speech recognition problem.(3)Different features have different recognition abilities in noise environment.Compared with traditional mainstream features,GFCC features have better anti-noise performance and are superior to MFCC features in different data.(4)Speech recognition in noise environment based on neural network.Introduced the limitations of the traditional noise reduction model and the noise classification model,as well as the advantages of the attention model.Experimental results show that the proposed algorithm performs well in different scene noises,in the case of known noise types,the results are similar,in unknown noise,results increases about 3%,effectively improves the accuracy of speech recognition in low SNR environment,and basically realizes speech recognition in complex noise environment.