Speaker identification: A way to reduce call-sign confusion events

Abstract

This paper examines the development of a speaker identification system (SIS) for future aeronautical communication systems. SIS promises to improve flight safety by reducing the incidence of call-sign confusion events. However, the practical development of such a system faces many challenges, especially related to the signal corruption by the channel noise. Due to the dynamic motion of aircraft, the aeronautical channel experiences high Doppler shifts and fading due to multipath propagation. This means that the SIS is required to be robust against such perturbations. In the proposed system, aeronautical channel noise was generated and mixed with speech signals to get the testing data. Four spectral features including Linear Predictive Cepstral Coefficients (LPCC), Perceptual Linear Prediction (PLP), Mel Frequency Cepstral Coefficient (MFCC) and Gammatone Frequency Cepstral Coefficient (GFCC) were extracted and then Support Vector Machines (SVM) were used for classification. The performance of the system was evaluated using noiseless and noisy signals from the ATCOSIM speech corpus. The experimental results show that the better recognition rate is obtained for GFCC under noisy conditions as compared to PLP, LPCC and MFCC.