notesum.ai

Published at November 5

cs.SD

cs.AI

cs.LG

eess.AS

Released Date: November 5, 2024

Authors: Pourya Jafarzadeh¹, Amir Mohammad Rostami¹, Padideh Choobdar¹

Aff.: ¹Lab of Artin, TOSAN TECHNO

Dataset	Model	Weighted Accuracy	Unweighted Accuracy
	Proposed Method (Hubert)	91.66 ± 1.2	90.48 ± 1.1
	Proposed Method (Wav2Vec2)	83.34 ± 1.3	80.95 ± 1.1
	Ensemble softmax regression [48]	51.50	-
	eGeMAPs feature set [15]	42.40	-
	DCNN + CFS + MLP [13]	66.90	-
SAVEE	CNN [33]	73.6	-
	Proposed Method (Hubert)	92.82 ± 1.5	93.78 ± 1.6
	Proposed Method (Wav2Vec2)	97.67 ± 1.2	98.02 ± 1.3
	DCNN + CFS + MLP [13]	73.50	-
	MFCC + Pitch + Energy + ZCR + DWT [28]	-	85
	CNN [37]	80	79
	GResNets [55]	-	68.48
	CNN+BLSTM [24]	69.4%	68.10
RAVDESS	Bagged SVM [6]	-	75.69
	Proposed Method (Hubert)	97.83± 1.4	98.21 ± 1.3
	Proposed Method (Wav2Vec2)	99 ± 0.2	99 ± 0.3
	Ensemble softmax regression [48]	82.40	-
	eGeMAPs feature set [15]	76.90	-
	OpenSmile features + ADAN [54]	83.74	-
	RESNET MODEL + Deep BiLSTM [42]	85.57	-
	Complementary Features + KELM [14]	84.49	-
	ADRNN [34]	85.39	-
	DCNN + DTPM [56]	87.31	-
	DCNN + CFS + MLP [13]	90.50	-
	CNN + LSTM [57]	-	95.89
	MSFs [52]	-	85.5
	CNN [59]	-	85.2
EMODB	Fuzzy C-Means [12]	-	92.2
	Proposed Method (Hubert)	98.36 ± 1.2	98.33± 1.4
AESDD	Proposed Method (Wav2Vec2)	98.36 ± 1.1	98.33± 1.1
	Proposed Method (Hubert)	83.77 ± 1.7	71.38 ± 1.4
SHEMO	Proposed Method (Wav2Vec2)	95.52± 2.1	91.21± 1.3