Research And Realization Of The Intelligent Wheelchair Speech Recognition Control Technology Based On Feature Extraction

Speech recognition technology can be thought of a comprehensivetechnology which has relationship with acoustics, phonetics, linguistics,computer science, signal processing, artificial intelligence, etc., it isbecoming the key technology in the information technology human-machineinterface, and has important theoretical meaning and vast potential for futuredevelopment. Currently the intelligent wheelchair is a research hotpot as animportant research field of the series of the elder-aid and handicapped-aidrobots, which combines speech recognition technology with intelligentwheelchair control technology. Such as: disabled people can use simplecommand to control the intelligent wheelchair etc. So researching practicalintelligent wheelchair speech recognition controlling system not only has agood meaning of theory but also has a great value in practicality.This paper introduces the basic principle of the intelligent wheelchairvoice command-word recognition and control technology, and studies HiddenMarkov Model (HMM) speech recognition algorithm based on the mixedfeature parameter of Mel Frequency Cepstral Coefficients (MFCC) and fractaldimension, and achieves the real-time speech control to the intelligentwheelchair.The paper constructs a speaker-independent, isolated word, smallvocabulary speech recognition system based on Hidden Markov Model (HMM)in the Matlab platform. The system completes the main process of speechrecognition, including pretreatment, endpoint detection, feature extraction andpattern matching. It establishes the training sample database and test sampledatabase for the intelligent wheelchair command words, and completesChinese voice command words:"Forward","Backward","Turn left","Turnright","Stop"training and recognition using Hidden Markov Model. The testresult indicates that the system's recognition rate achieves above 85%.The effect of the different feature parameters to the system performanceis analyzed through simulation experiments. In the phase of extraction ofspeech feature parameters, because of the fractal characteristics in speechtime domain waveform, fractal dimension may be regarded as a kind of feature parameters in command-word recognition. So we propose a newspeech feature extraction method with the mixed feature parameter, whichcombines the traditional Mel Frequency Cepstral Coefficients (MFCC) orLinear Predictive Cepstral Coefficients (LPCC) with fractal dimension as thefeature parameter. Thus, such mixed feature parameter can show the speechfeature better and improve the system's recognition rate. The simulationresults show that mixed feature parameter is better than single traditionalMFCC or LPCC feature parameter in recognition rate.The experiments of using command words to control intelligentwheelchair show that the speech recognition system reaches the expectationeffect in laboratory conditions. The speech recognition system discussed inthis paper can be used not only for intelligent wheelchair controlling byspeech but also for laying the foundation to develop more complex intelligentwheelchair control technology based on multi-sensor and multi-informationfusion in the future.