Research And Development Of Small-scale Speech Recognition System

Speech is an important tool to communicate between human being and machines.Speech recognition means the machines understand the speech. Now speechrecognition technology has been broadly applied. This paper small-scale speechrecognition system aims to the small-scale vocabulary and isolated-words speechrecognition. This paper is discussed from two sides. On the one side, the paper focuses on the pattern recognition, the principle andmodels of speech recognition. We build the model to recognize the speech in thesmall-scale vocabulary. Firstly, the principle of the speech recognition, including thedigital speech model and speech processing, is introduced. Speech signal framing andendpoint detection are emphasized. Speech feature extraction is one of the importantparts of speech recognition. Liner Predictive Cepstrum Coefficients (LPCC) andMel-Frequency Cepstrum Coefficients (MFCC), two speech feature parameters arediscussed. We choose the MFCC as the speech feature parameter. The principle ofDynamic Time Warping (DTW) arithmetic and Hidden Markov Model are discussedand used to speech recognition system. Building the DTW model to speakerdependent speech recognition, we perform the speech recognition experiment for'0~9' 10 numbers and 23 Chinese words. The result of the experiment is good. Aswell as building the Continues Gauss Distribution HMM model to speakerindependent speech recognition system, we got the basis result after the experimentfor 10 numbers words. On the other side the paper focuses on the software development. We set up thesystem of command recognition embedded in the endoscope imaging system usingthe soft development kit, Speech API . Based on the system of command recognition,the result of experiments on the recognition for 23 Chinese command words is good.Further, we resolve the rejection recognition problem and bring forward the methodsto improve the recognition result.