Gammatone Feature-based Classification and Signal-to-Noise Ratio Estimation of Aircraft Noise in Combined Noise

The applicability of Gammatone Frequency Cepstral Coefficients (GFCCs) was studied by comparing its performance with Mel Frequency Cepstral Coefficients (MFCCs) in their combined noise environments. A multi-output artificial neural network model was developed to classify aircraft, road traffic, and combined noise, and to estimate the signal-to-noise (SNR) ratio when combined noise is detected. GFCC features were extracted using 64 Gammatone filters and 13 cepstral coefficients. Trained on 10 076 samples, the model achieved a classification accuracy of 99.94 % and an SNR estimation mean absolute error of 0.5432, outperforming MFCC-based models. These results demonstrate the effectiveness of GFCC in analyzing low-frequency dominant aircraft noise under combined noise conditions.