A Feature Level Fusion Scheme for Robust Speaker Identification

For speaker identification purposes, features are first extracted and then compared with those of the training set to find the closest match. So, finding effective and robust features for classifying speakers is beneficial to improve the overall identification performance, especially in the presence of noise. In this paper, a new method of feature extraction based on feature fusion is proposed, where Gammatone Frequency Cepstral Coefficients (GFCC) and wavelet components are extracted and fused for training and testing the Support Vector Machines (SVM) classifier. The performance of the proposed scheme is validated and compared with conventional GFCC using clean and noise corrupted signals from Voxforge database. From the experimental results, it is evident that our algorithm has a higher identification accuracy compared to baseline GFCC.