Speech Emotion Recognition using non-linear Teager energy based features in noisy environments

In this study, Teager-energy based Mel-frequency cepstral coefficients (TEMFCCs) are proposed for Automatic Speech Emotion Recognition (ASER) in noisy environments. TEMFCCs are obtained by taking the absolute value of the Teager-energy operator (TEO) of the short-time Fourier transform of the signal (STFT), warping it to a Mel-frequency scale, and taking the discrete cosine transform (DCT) of the log-Mel Teager-energy spectrum. Experiments on classification of discrete emotion categories show that TEMFCCs are more robust than MFCCs in noisy conditions, while TEMFCCs and MFCCs perform similarly for clean conditions.