MFCC-Based Forensic Speaker Recognition From Disguised Voice Using Machine Learning Algorithms

Automatic Speaker Recognition (ASR) identifies individuals from voice signals, but voice disguise remains a major challenge. This work detects disguised speech using MFCC features along with statistical measures such as mean, variance, and correlation. A custom dataset of artificially disguised voices is used in a text-dependent setup, and classifiers including k-NN, SVM, BPNN, and RBFNN are evaluated. Results show that combining statistical features with MFCC improves classification accuracy compared to using MFCC alone.