Hybrid feature extraction method of MFCC+GFCC helicopter noise based on wavelet decomposition

Abstract Aiming at the issue that recognition accuracy of traditional acoustic signal features is low for helicopter signals with wind noise in near field, a method extracting mixed MFCC+GFCC based on wavelet decomposition proposed. Firstly, three-layer and reconstruction are applied to signals; then, Mel-Frequency Cepstral Coefficients (MFCC) Gammatone-Frequency Cepstrum Coefficient (GFCC) respectively extracted approximation detail components; next, coefficients components which averaged combined those form hybrid feature parameters; finally, convolutional neural network used classify signal, realize correct signals. Experimental results show improved by almost 40% contrast other methods, such as MFCC GFCC, when SNR equal -5dB. Further, When -10dB, more than 49%, while methods cannot effectively recognize targets. The proposed extraction can significantly improve environment, provide reference near-field detection