Font Size: a A A

Speech Recognition In Noise Immunity

Posted on:2007-01-24Degree:MasterType:Thesis
Country:ChinaCandidate:Y H ZhangFull Text:PDF
GTID:2208360185491201Subject:Computational Mathematics
Abstract/Summary:PDF Full Text Request
With the development of computer science, the research of speech recognition trends to the fields of large-vocabulary, continuous speech recognition and etc. But the performance of the speech recognition system degraded greatly under actual environment because of weakness in robustness, flexibility and self-adaptive, especially the noisy problem. As the preprocessing part of the cooperation with The Hong Kong Polytechnic University "Study of Control Technology for Building Services Based on the Recognition of the Multimedia Systems", it is development mostly that the feature extraction method, speech enhancement in speech signal process in this paper in order to improve the robustness under the actual environment.The paper elucidates the basis of hearing masking, and the traditional method used in speech recognition with masking feature. Based on the recognition algorithm of DTW and HMM, an new method to compute MFCC using the relationship between mel frequency and bark frequency is developed. CMN is also used to MFCC to improve the performance in noisy environment.Additionally, the paper introduces how to model the speech signal using GARCH which is used in financial application .The GARCH model can describe the feature of speech signal more exactly. This model is used in preprocessing of speech recognition to improve the quality of the speech.
Keywords/Search Tags:Speech recognition, Speech enhancement, Hearing masking, GARCH
PDF Full Text Request
Related items