Speech Recognition Based on Support Vector Machine and Error Correcting Output Codes

摘要

A method is proposed based on application of Error Correcting Output Codes Support Vector Machine (ECOC-SVM) in order to get better results of speech recognition. Some uncorrelated SVMs are constructed based on ECOC matrix codes to improve the integrated performance of fault tolerance of classification model. This paper gives four commonly-used encodings of ECOC. By comparing the results with that of speech recognition based on HMM, the experiments indicate that the ECOC method is more suitable for speech recognition, among which the predicting accuracy of one-versus-one is the highest of all.