Speaker identification using FBCC in Malayalam language
Drisya Vasudev, K.K. Anish Babu · 2014
Speaker identification attempts to determine the best possible match from a group of certain speakers, for any given input speech signal. The text-independent speaker identification system does the task to identify the person who speaks regardless of what is said. The first step in speaker identification is the extraction of features. In this proposed method, the Bessel features are used as an alternative to the popular techniques like MFCC and LPCC. The quasi-stationary nature of speech signal is more efficiently represented by damped sinusoidal basis function that is more natural for the voiced speech signal. Since Bessel functions have damped sinusoidal as basis function, it is more natural choice for the representation of speech signals. Here, Bessel features derived from the speech signal is used for creating the Gaussian mixture models for text independent speaker identification. A set of ten speakers is used for modelling using Gaussian mixtures. The proposed system is made to test over the Malayalam database obtaining an efficiency of 98% which is promising.