Mel-frequencies Stochastic Model for Gender Classification based on Pitch and Formant

Abstract

Speech recognition applications are becoming more and more useful nowadays. Before this technology is applied, the first step is test the system to measure the reliability of system.  The reliability of system can be measured using accuracy to recognize the speaker such as speaker identity or gender.  This paper introduces the stochastic model based on mel-frequencies to identify the gender of speaker in a noisy environment.  The Euclidean minimum distance and back propagation neural networks were used to create a model to recognize the gender from his/her speech signal based on formant and pitch of Mel-frequencies. The system uses threshold technique as identification tool. By using this threshold value, the proposed method can identifies the gender of speaker up to 94.11% and the average of processing duration is 15.47 msec. The implementation result shows a good performance of the proposed technique in gender classification based on speech signal in a noisy environment.