Feature Extraction Analysis in a Speaker Identification System

Human interaction and communication both rely heavily on language. To emphasize the properties of communication signals, speech recognizers use a parametric signal structure. There are numerous methods for removing elements from audio streams. In this study, four characteristic extraction techniques - the Mel frequency cepstral coefficient (MFCC), the bark frequency cepstral coefficient (BFCC), the Gamma frequency cepstral coefficient (GFCC), and the Linear Predictive Coding (LPC)—are used to assess the performance of the Gaussian mixture model (GMM) in speech recognition. The research analyses the computation time, the number of speakers to be researched, and the recognition speed to determine the optimal feature extraction method. The findings show that GMM-based MFCC performs better in recognition than BFCC, LPC, and GFCC.