A comparative study of text-independent speaker recognition systems using Gaussian mixture modeling and i-vector methods
Suma Paulose, Dominic Mathew, Abraham Thomas · 2017
Speaker recognition systems are increasingly being used in today's world as a means of biometrie in many real life applications. The accuracy of such a system decreases with noise, channel variations, duration of speech signals used etc. It is always believed that these systems work well when sufficient training data is available to get rid of these effects. In this paper, we make a comparison of text-independent speaker recognition system using two different feature extraction methods: Mel Frequency Cepstral Coefficients (MFCC) and Inner Hair Cell Coefficients (IHC). For classification we used Gaussian Mixture Modeling (GMM) and i-vector modeling using Probabilistic Linear Discriminant Analysis (PLDA). The recognition results for 100 different speakers were compared and it was observed that MFCC is better than IHC and i-vector modeling performs better for long utterances.