Based On The Design Of The Vc Advertising Speech Recognition System

TV Advertising is an important part of daily life, through which people know whatproducts they like or dislike. But as the important effects it takes on, it is more and moreurgent for us to monitor whether the right ads is on TV.In this paper, the real-time realization of Speech Recognition technology on commonplatform is investigated. On the base of the Speech Recognition theory, the primeprinciples and the methods of speech front-end detection, the audio characters extracting,acoustic modeling and its similarity estimation are studied.In speech front-end detection, we introduce that the short time average energy andaverage zero across rate. And we use twin thresholds method to deal with speech front-enddetection based on the result of simulation.Before extracting the audio characters, we learn that LPCC and MFCC are the usefulcharacters.During the match between two audios, we suffer a great of problems, like largeamountof data, and different lengths of two audios. To solve these problems, we apply thetechnology of K-means method, Vector Quantity and Dynamic Time Warping.Through simulations, the specificity of usual algorithms and the methods of parameterselection are compared, then a modeling method of column standard deviation arepresented.Based on the working above, a set of speech recognition system and thecorresponding demo software are built under VC. The system is examined with somenamed advertising. The results show that the system has a higher recognition rate.