Gender recognition using FB series expansion and SVM

Speech signal does not only carry linguistic information needed for communication among persons but also non-linguistic information relating to particular speaker. Non linguistic features are unique for each speaker due to physical length of vocal tract, vibration of vocal folds and speaking style of the speaker. Gender identification finds application in biometric security systems. Gender specific information can help researchers by reducing their search to half and computational complexity of the systems. After identifying gender specific information and removing undesired information of particular gender bandwidth of transmission can be utilized efficiently. It is demonstrated through experiments in literature that Male has fundamental frequency (F0) around 120 Hz and females has F0 around 200 Hz. There exists overlapping region ie between 120Hz to 200 Hz where both the gender can have F0.In this work, gender identification of speech signal using FB series expansion and SVM is proposed. FB expansion is used to extract features of speech signals and Support Vector Machine (SVM) is applied on feature vectors to recognize the gender of speakers. FB expansion is capable of extracting features (F0) from speech signals more accurately since it does not require filter to separate desired speech signal and hence is free from filter related problems. Feature vectors are given to SVM to classify features to determine gender of speakers. SVM used in the proposed method exhibits its suitability to classify non-separable class of data using hyper plane than any conventional kernels methods to classify data.