Table 8. Comparison of the SD experiments with existing methods.

Database	Reference	Feature	Accuracy(%)
RAVDESS	Bhavan et al. (2019)	Spectral centroids, MFCC and MFCC derivatives	75.69
RAVDESS	Proposed approach	AlexNet + FS + RF	86.79
RAVDESS	Proposed approach	AlexNet + FS + SVM	88.77
SAVEE	Özseven (2019)	OpenSmile features	72.39
SAVEE	Proposed approach	AlexNet + FS + RF	86.79
SAVEE	Proposed approach	AlexNet + FS + SVM	88.77
Emo-DB	Guo et al. (2018)	Amplitude spectrogram and phase information	91.78
Emo-DB	Chen et al. (2018)	3-D ACRNN	82.82
Emo-DB	Meng et al. (2019)	Dilated CNN + BiLSTM	90.78
Emo-DB	Özseven (2019)	OpenSMILE features	84.62
Emo-DB	Bhavan et al. (2019)	Spectral centroids, MFCC and MFCC derivatives	92.45
Emo-DB	Proposed approach	AlexNet + FS + MLP	95.80
Emo-DB	Proposed approach	AlexNet + FS + SVM	96.02
IEMOCAP	Satt, Rozenberg & Hoory (2017)	3 Convolution layers + LSTM	68.00
IEMOCAP	Chen et al. (2018)	3-D ACRNN	64.74
IEMOCAP	Zhao et al. (2018)	Attention-BLSTM-FCN	64.00
IEMOCAP	Etienne et al. (2018)	CNN + LSTM	64.50
IEMOCAP	Meng et al. (2019)	Dilated CNN + BiLSTM	74.96
IEMOCAP	Proposed approach	AlexNet + FS + MLP	89.12
IEMOCAP	Proposed approach	AlexNet + FS + RF	86.23