Text Dependent Speaker Identification from Disguised Voice Using Feature Extraction and Classification

Speaker identification is used in voice-controlled software, phone fraud detection systems, information retrieval systems, etc. In this proposed research work, created a text-dependent speaker identification system that can classify the features of speech signals and identify the speakers. Mel Frequency Cepstral Coefficients (MFCC), Delta MFCC (AMFCC), and Double Delta MFCC (AAMFCC) feature extraction techniques are used for extracting the features of normal and disguised voice. In MFCC feature extraction, frames the speech signal, calculates FFT, applies Mel filter bank, and calculates DCT after logarithm. For the statistical analysis it is computed the mean and correlation coefficients by using MFCC, AMFCC and AAMFCC feature extraction techniques. After feature extraction, feature based decision tree (DT) and support vector machine (SVM) classifiers are used for computing the classification efficiency. After classification it is computed that 88.89% and 93.33 %, classification efficiency from DT and SVM classifier respectively. In the existing results classification efficiency was computed 83.73 % and 84.67% by using DT and SVM classification method. After comparing it is found that the proposed model for speaker identification based on feature extraction and classification method are better than the existing method.