The Research Of Fusion LPCC And MFCC Feature Parameters In Speech Recognition Technology

With the rapid progress and development of computer technology and communication Technology, there is an urgent need to achieve the purpose of the interaction between human and computers, it is to make the machine can understand human's words or commands, which prompted the speech recognition technology has been rapid development and progress. After 60 years of research and development, the research in speech recognition technology has achieved some good results, but there are some difficult technical problems in fluently communicating with the computer by human language, so the experts and researchers should do further research and exploration in speech recognition technology.Some fundamentals, key technology and system framework of speech recognition technology are introduced in this paper. The major components of the speech recognition system was studied and analyzed as a whole: preprocessing, feature extraction, model training section, the library section of the model parameters and pattern matching section. This paper focuses on the detailed derivation and concrete implementation steps of the LPCC and MFCC feature extraction algorithm. LPCC feature extraction is that firstly to do linear coding LPC analysis and calculation, and then LPC coefficients are obtained, the last step to compute its cepstrum coefficient to obtain LPCC coefficients. MFCC feature extraction is that firstly to do FFT transform, and then to pass through MEL filter bank, and last step to do logarithmic calculation and DCT transformation to obtain MFCC coefficients. In addition, a new feature parameters LPMFCC was introduced, and its extraction process was combined with the above of two feature extraction process: LPC coefficients applicable to Mel cepstrum calculation. On the basis of comparative analysis LPCC and MFCC's advantages and disadvantages. This paper presents a fusion LPCC and MFCC feature extraction algorithm parameters—A Based on Fisher Criterion fusion feature extraction algorithm, and designed two algorithms implementations.This paper focuses on the research and improvement of the speech signal feature extraction algorithm. Based on researching and analyzing in feature extraction algorithm of LPCC and MFCC, this paper presents a feature extraction algorithm of fusion LPCC and MFCC. Evaluating the merits of the algorithm is a standard voice recognition system recognition rate, timeliness and noise immunity. In this paper, I have done a large number of comparative experiments with MATLAB simulation software to analysis these characteristic parameters on effect of the speech recognition system performance.The results show that the fusion characteristic parameters based on Fisher criterion are easier to characterize the speech signal characteristics in extracting algorithm parameters than the original characteristic parameters LPCC and MFCC feature parameters. It improves the recognition rate of speech recognition system, and background noise immunity performance has also been strengthened. Also the terms of impact on the effectiveness of the system is not very big, so the fusion feature extraction algorithm proposed in this paper is more suitable for speech recognition system and for being put into practical use.