Automatic gender recognition using linear prediction coefficients and artificial neural network on speech signal

Automatic Gender Recognition (AGR) system is an intelligent machine inspired by the highly advanced skills of human cognitive and developed through adequate training to recognize the gender of a speaker as male or female. In this paper, speech from 93 speakers were extracted using Linear Prediction Coefficients (LPC). Pre-processing steps such as normalization, pre-emphasis, frame blocking and windowing were carried out prior to feature extraction. The LPC coefficients of different order was investigated to produce the optimum parameters for the developed AGR. Artificial Neural Network (ANN) was used as the recognition engine and Multi-Layer Perceptron (MLP) was adopted to train the ANN. The experimental results show the highest overall recognition rate that can be achieved by the proposed system was 93.3% in average and the results also indicate almost equal performance for the detection of male and female throughout the experiments.