Research On Isolated Word Speech Recognition Algorithm And System Simulation

Talking with the machines and trying to make machines understand nature language and execute relevant task, called speech recognition. Speech recognition technology is a high-technology that it can make machines recognize and understand the speech signal and then translate into relevant texts or orders. It mainly includes feature extraction technique, pattern matching criterion and model training techniques. It is concerned with many research fields of varied subjects and fully used in the high-tech fields.First, this paper analyses the research and development of speech recognition in detail, and has done a very good summary in such aspects as developing history, existing problems and developing direction of speech recognition, lading a good foundation for further study of speech recognition.Second, the speech signal preprocessing and feature extraction problems have been discussed. Discuss the methods of extracting the different characteristic parameters of speech recognition systematically, especially analyze LPCC (Linear Prediction Cepstrum Coefficient) and MFCC (Mel Frequency Cepstrum Coefficient) parameters which reflect cepstrum characteristic. Research shows MFCC is better than LPCC in increasing the recognition nature of the system. In order to reflect the dynamic performance of the characteristic parameters, this paper also proposes LPCC, MFCC one steps, two steps difference parameter.Third, we focused on the discussion on the application of classical methods in speech recognition, including the DTW (Dynamic Time Warping) and HMM (Hidden Markov Models). What’s more, some problems such as scaling, and parameter smoothing have also been discussed. Experiment results show using HMM in isolated word recognition, the recognition rate is higher than using DTW, but HMM algorithm needs more time. Aiming at the defects of the traditional HMM, on the basis of thorough research of the VQ (Vector Quantization) technology and the HMM, this paper proposes VQ-HMM method. Theory analysis and experiment results show this algorithm is easier, needs less recognition time, and its recognition rate is a little higher than the former.Finally, apply the MATLAB program, through the simulation experiment on isolated words speech recognition system voice signal acquisition, pretreatment, the feature extraction, training and identify links are verified, performance analysis and the results are reviewed.