Research of Window Function Influence on the Result of Arabic Speech Automatic Recognition

One of the causes of errors in automatic recognition is the influence of the frequency response of the communication channel on speech signal parameters, which are used for recognition. Normalization of parameters of a voice signal on mean value suppresses influence of the frequency response of communication link on the specified parameters. In case of normalization, the number of errors of automatic speech recognition is reduced. The analysis of the factors influencing the result of normalization of parameters of a voice signal on mean value is carried out. The results of the experiments, demonstrate the reduction of influence of the frequency response of communication link by the normalized parameters of a test signal when using window functions with the small level of side lobes of their frequency response. The average accuracy of the recognition system is above 95%.