| With the development of speech recognition technology, an increasing number of voice products have come into people's life. In recent years, the popularity of vehicle increasingly improves and customers are demanding more and better vehicular functions. Applying speech recognition techniques and then controlling ancillary equipment in cars are able to improve the cars' comfortableness and safety. The vehicular speech recognition system has the characteristic of instantaneity, robustness and high recognition rate.This article studies the isolate word speech recognition of a small amount of Chinese words spoken by a group of particular people. The article presents the basic theories and processing flow of speech recognition in first chapter. Then the article introduces each procedure of speech recognition and the programming and simulation experiment using MATLAB in subsequent chapters. It contains the contents of using digital high-pass filter to filter out most of noise, putting forwards a parameter applied for endpoint detection that combine Zoning spectral entropy, C0 complexity and spectral energy of every frame as the featured parameter of endpoint detection, extracting 24 dimension Mel Frequency Cepstral Coefficients(MFCC) as the featured parameter of speech recognitio, dividing the voice command into verbs and nouns based on the length of voice signal and distinguishing these two parts by applying different recognition algorithm, obtaining nine algorithm by combing the algorithm of these two parts together and applying Grey Fixed Weight Clustering method to evaluate each combinational algorithm based on the evaluation index of vehicular speech recognition and then choosing the best recognition algorithm.Based on the result of the experiment, recognition algorithm of verbs applies BP neural net while recognition algorithm of noun applies Hidden Markov Model (HMM). This combinational algorithm is more adaptable to in-vehicle speech recognition system in this study. |
PDF Full Text Request