Detecting emotion change instant in speech signal using spectral patterns in pitch coherent single frequency filtering spectrogram

Emotional intelligence means an ability to understand how others are feeling so that one can adjust according to others' emotional needs. Machines with emotional intelligence must understand human emotions as well also be informative about their emotional changes. Several researchers have tried to detect emotions from speech but little work is done in the area where machines could predict emotional changes. In the proposed work we have tried to analyze the spectrum of the glottal source to predict emotional changes. The glottal source spectrum is represented as a single-frequency filtering spectrogram. Single frequency spectrogram has high resolution in the time and frequency domain simultaneously. A single-frequency filtering spectrogram is fed to EfficientNetB7 to predict time-instant emotional changes. The performance of the proposed method is evaluated on two datasets IEMOCAP and RAVDESS. The proposed method shows high accuracy in comparison to other state-of-the-art techniques used for emotion prediction.