Shallow and deep feature fusion for digital audio tampering detection

EURASIP Journal on Advances in Signal Processing

Table 4 Detection accuracy of fitting coefficient features (%)

Feature	Classifier	Dataset	3 Sines	4 Sines	5 Sines	6 Sines	7 Sines	8 Sines
Proposed feature \(f'\)	SVM	Classical	79.21	79.21	82.18	88.12	88.12	91.09
	SVM	GAUDI-DI	77.29	79.28	86.19	86.59	86.32	83.40
	RF	Classical	87.13	85.15	83.17	86.14	87.13	92.08
	RF	GAUDI-DI	83.27	82.74	86.85	86.59	85.13	86.59
	XGBoost	Classical	88.12	88.12	84.16	89.11	87.13	90.10
	XGBoost	GAUDI-DI	76.49	80.08	81.67	80.48	86.32	85.92
Proposed feature \(\phi\)	SVM	Classical	86.14	78.22	86.14	80.20	81.19	79.21
	SVM	GAUDI-DI	77.95	75.83	73.04	74.63	73.71	74.50
	RF	Classical	90.10	89.11	90.10	90.10	89.11	90.10
	RF	GAUDI-DI	79.42	79.68	80.08	82.74	82.60	81.14
	XGBoost	Classical	91.09	89.11	90.10	87.13	86.14	88.12
	XGBoost	GAUDI-DI	80.61	77.29	79.55	80.48	81.01	81.41

Bold means the best performance of tampering detection in the same experimental setting