Enhanced Mining of Audio Signals from Optimal Intrinsic Mode Functions Through Statistical Analysis

Automated speaker recognition system is extremely important in areas such as Forensic and Defence. Performance of an automated speaker recognition system is dependent on feature extraction and classification. As speech is a non stationary signal and most of the information is present in the low frequency region, sub band coding or multi resolution analysis is favored. Empirical Mode Decomposition (EMD) is used for decomposing the signal and pitch, period, number of peaks, mean, standard deviation, skewness, kurtosis, energy, zero crossing rate, second order moment and third order moment are used for aggregating these coefficients. Adaptive Resonance Theory (ART) is chosen as the classifier. Sensitivity is the metric used for studying the performance of the proposed technique. From the results, it is also proved that the proposed ART based speaker recognition system using Empirical Mode Decomposition parameters gives 100% sensitivity for the acquired research database.