Extracting accent information from Urdu speech for forensic speaker recognition

Abstract

This paper presents a new method for extraction of accent information from Urdu speech signals. Accentis used in speaker recognition system especially in forensic cases and plays a vital role in discriminating people ofdifferent groups, communities and origins due to their different speaking styles. The proposed method is based onGaussian mixture model-universal background model (GMM-UBM), mel-frequency cepstral coefficients (MFCC), and adata augmentation (DA) process. The DA process appends features to base MFCC features and improves the accentextraction and forensic speaker recognition performances of GMM-UBM. Experiments are performed on an Urdu forensicspeaker corpus. The experimental results show that the proposed method improves the equal error rate and the accuracyof GMM-UBM by 2.5% and 3.7%, respectively.