Applied Research On Specific Word Chinese Speech Recognition System

At present, the speech recognition technology has already been widely applied in diverse areas, such as: household electronics,intellectual toys, the database voice inquiry of the business system, industrial voice control, autodialing of telephone and telecom system, etc., and will affirmatively become the interface of the next generation operating system. This paper makes an attempt on the investigation in DTW and HMM with their applications to Chinese speech recognition.At first, various aspects of the fundamentals of the speech recognition system are discussed in detail. Then preprocessing and feature extraction are followed. After that are the windowing, noise filtering, endpoint detection. Two important speech analysis methods for speech recognition: MFCC and LPCC are intensively investigated and compared. In the result, MFCC is selected as feature selector for speech signals.At the second phase, several speech recognition algorithms, including the DTW and HMM, are thoroughly discussed .We have experimentally proven the improved DTW outbalancing the fundamental DTW by recognition efficiency. Also, we proposed a novel seeking method for confirming reference patterns. The method is that firstly computing the DTW cumulative distances between each other of several characteristic vector sequences for each word, then summing up these distances to everyone vector sequence from all other vector sequences for the same word, selecting the minimum from these summed distances and the correspondent characteristic vector sequence regarded as the reference pattern for that word .This novel seeking method has experimentally shown excellent recognition accuracy. Then we address HMM's fundamentals and HMM's parameter selection problems. The last but the most important, both of DTW and HMM algorithms are implemented on a PC, and the comparisons between these two algorithm for efficiency and accuracy are carried out and the results are listed in this paper.