Analysis of constant-Q filterbank based representations for speech emotion recognition

This work analyzes the constant-Q filterbank-based time-frequency representations for speech emotion recognition (SER). Constant-Q filterbank provides non-linear spectro-temporal representation with higher frequency resolution at low frequencies. Our investigation reveals how increased low-frequency benefits SER. The time-domain comparative analysis between short-term mel-frequency spectral coefficients (MFSCs) and features, namely transform (CQT) continuous wavelet (CWT), that provide time-invariance low-frequencies. robustness against irrelevant temporal variations in pitch, especially low-arousal emotions. corresponding frequency-domain over different classes shows better of pitch harmonics constant-Q-based than MFSC. These advantages are further consolidated by SER performance extensive evaluation features four publicly available databases six advanced deep neural network architectures as back-end classifiers. inferences this study hint toward suitability potentiality