Automatic enlargement of speech corpus for speaker recognition

This research deals with the problem of recognition when only a few samples are available for training of the system. To avoid the low recognition rate caused by such type of speech corpus, automatic techniques for the enlargement of speech corpus are proposed in this paper. These techniques are: lengthening of sample by automatic segmentation, automatic noise addition at different sound-to-noise ratios (SNRs), and lengthening of reversed sample. Different combinations of samples, generated by the proposed techniques, are used to obtain the high recognition rate. These techniques have shown promising result.