Порівняльний аналіз ефективності ідентифікації дикторів за спектральними характеристиками

Abstract

This work investigates the accuracy of automatic speaker identification using three spectral features: Mel-frequency cepstral coefficients (MFCC), spectral centroid, and spectral roll-off. An experiment was conducted on eight speakers based on a Java-implemented system. Mathematical descriptions of the features and similarity metrics used in the software implementation are provided. The results can be used in the design of voice biometric systems where a balance between computational complexity and reliability is required.