Mimicking voice recognition using MFCC-GMM framework
M V Unnikrishnan, Rajeev Rajan · 2017 International Conference on Trends in Electronics and Informatics (ICEI) · 2017
Evaluating the quality of mimicked speech has started more attention nowadays since it may affect speaker verification system as in spoof attack. In this paper, mel frequency cepstral coefficients (MFCC) are effectively utilized for evaluating the quality of text independent mimicked speech. Gaussian mixture model (GMM) based classifier is used in the classification phase to make decision based on the log-likelihood score. The experiment evaluates the competence of 5 artists in mimicking 5 target speakers and rank them according to the scores of a classifier. This is tested against the mean score obtained through the perception test conducted on 20 listeners. If the artist with highest mean score is identified as rank-1 by the proposed system, a hit occurs. The results show the promise of MFCC in evaluating voice mimicking performance.