Research On The Speech Recognition Technology In Human Computer Interaction

After a brief discussion on the future of speech recognition technology in the field of human-computer interaction,this paper deeply studies the knowledge of speech recognition technology.First of all,after a lot of research on the technology of speech recognition,due to the endpoint detection is one of the most important part,this paper proposed an endpoint detection method which is based on MFCC cosine value with only one single threshold.Secondly,in order to improve the quality of the feature parameters,this paper studies the speech enhancement algorithm,and proposes a speech enhancement method based on LMS algorithm and spectral subtraction.Finally,this paper proposes a speech recognition method which combines these two algorithms,and not only verified this method under the experimental condition,but also developed a software based on this.The main contents and innovations of this paper are as follows:1)Endpoint detection is the most important part of speech recognition,so a large number of related algorithms are studied.And based on the the former people's research findings which is the VAD algorithm based on Euclidean distance MFCC with double thresholds,this paper proposed an algorithm based on MFCC cosine value with only one single threshold,as called MFCC_COS.By calculating the cosine value,MFCC_COS algorithm avoid the numerical sensitive problem of the euclidean distance.By using signal threshold,it reduced a larger error probability of the double thresholds.The algorithm is simple and performs well in the noise environment,and with the increase of the noise intensity,the detection accuracy will not be reduced too fast.Compared with the traditional algorithm,it has better robustness.2)The noise in the real environment can not be avoided,so this paper will do the speech enhancement before extracting the feature parameters.After studying a large number of speech enhancement methods,we proposed a new speech enhancement method which is based on spectral subtraction and LMS algorithm,as called LMSSS.This algorithm eliminates the problem of music noise after spectral subtraction,and also avoids the filter delay problem of LMS algorithm.The experimental results show that this method is more effective than the use of spectral subtraction alone or LMS algorithm alone.And in the experimental range,the stronger the noise are,the more obvious advantages it has.3)The paper finally proposed one speech recognition method combines LMSSS algorithm and MFCC_COS algorithm.And the experiment data shows that the recognition rate of the proposed method has been further improved,and in the noise environment,it showed a very good robustness.In addition,in the end of this paper,this speech recognition method had been used for developing a software.