混合语音识别模型的设计与仿真研究

摘要

研究语音识别率问题,语音信号是一种非平稳信号,含有大量噪声信息,目前大多教识别算法线性理论,难以正确识别语音信号非线性变化过程,识别正确率低.通过将隐马尔可夫模型(HMM)和SVM相结合组成一个混合抗噪语音识别模型(HMM-SVM).同时用HMM模型对语音信号时序进行建模,并得到待识别语音信号的输出概率,然后将输出概率作为SVM的输入进行学习,得到语音分类信息,最后通过利用HMM-SVM识别结果做出正确识别决策.仿真结果表明,HMM-SVM提高语音识别正确率,尤其在低信噪比环境下,明显改善了语音识别系统的性能.%Study noisy speech recognition rate. The speech signals are non-stationary signals which contain a lot of noises, and most of the current recognition algorithms are based on linear theory, therefor, the correct recognition rate is low. The HMM model and SVM( HMM-SVM) were combined to build a noisy speech recognition model. The HMM-SVM model first the HMM model was used in modelling the speech signal time series and to calculate the out-put probability of the the speech signals to be recognized. Then the output probability was used as input of SVM for learning to acquire the information voices classification. Finally. The HMM-SVM identification results were used to make the correct decision. The simulation results show that, compared with the single HMM or SVM model, the HMM-SVM model can improve the accuracy of speech recognition. Especially in low SNR environment, HMM-SVM can significantly improve the performances of speech recognition systems.