The Technology Research Of Speech Recognition System For Isolated Words

Speech recognition is a very popular research area in recent years.With the development of technology it has been widely used in practical applications,and among them isolation word plays important role.The isolated word speech recognition system is mainly used in automatic control,such as driving,robot manipulation,instruments and equipment manipulation,as well as the collection of dial-up,smart toys,household appliances manipulation.,especially when human's hands have been occupyed or are unable to use and them must implement the vioce contrl,so the isolated word speech recognition system play the unreplaceable role.This paper focus on the reserch and implement of isolated command word recognition with DTW(Dynamic Time Warp )based on MATLAB GUI.According to the process of isolated word speech recognition system,begin with the front-end of speech signal,the paper described each step in details respectively, including sample,pre-emphasis,enframe and window,until the the selection of characteristic parameters as well as the recognition algorithm.Ultimately it designed a small vocabulary isolated word recognition system,which display the main result with GUI(Graphical User Interfaces),and the realization of speech recording is realtime which has excellent interfaces.Endpoint detection is one of the important parts in speech recognition.Efficient endpoint detection can reduce the processing time,eliminate the disturbance of non-speech segment noise,and improve the system's recognition performance.This paper improved the endpoint detection algorithm based on the traditional double threshold algorithm with increaming the length of tolerable non-voice,which detected the voice interval of speech signal that have more than one words effectively,and laid a good foundation for following steps.There are mainly two kinds of characteristic parameter of speech signal:LPCC (linear prediction cepstrum coefficient) and MFCC(Mel frequency scale cepstrum parameter).MFCC simulates human hearing characteristic fully,and has the excellent recognition performance and the anti-noise ability,so chooses it as the recognition parameter in this paper.In the practical application,the MFCC and its first-order difference coefficient were used.DTW(Dynamic Time Warping) and HMM(Hidden Markov Model) is a mainstream technology of recognition algorithm,because this paper is based on small vocabulary isolated word recognition system,so DTW will be able to obtain good results.At the same time,for DTW,this article improve algorithm in terms of recognition rate and implementation efficiency including proposed relax start and end point and improved local function,which saving computation time and improved implementation efficiency.Finally,under MATLAB environment,with the aid of GUIDE this good development kit,the paper displayed the entire related process of recognition system with the form of GUI,including real-time voice recording and playback,endpoint detection and playbacking as well as the last recognition result output based on DTW. The whole related process of system show for us by a way of graphical interface clear,which can achieves the the dual effect of audio-visual!...