A sub-band-based feature reconstruction approach for robust speaker recognition

EURASIP Journal on Audio, Speech, and Music Processing

Table 1 Recognition performance of FB, 2-SB reconstruction, and marginalization in the presence of different types of noise (unit: %)

		0 dB	5 dB	10 dB	15 dB	20 dB
Babble	FB	82.60	84.02	87.01	87.49	89.49
	2-SB	82.96	85.30	87.84	89.30	91.25
	Marginalization	63.71	64.88	66.44	67.69	69.75
Factory1	FB	76.52	82.10	87.40	88.00	88.33
	2-SB	80.01	82.25	87.70	89.13	90.23
	Marginalization	67.54	68.10	68.33	68.51	69.40
Pink	FB	75.12	80.78	84.83	87.55	89.11
	2-SB	76.66	82.43	87.40	89.62	90.27
	Marginalization	67.79	68.54	69.00	69.06	69.91
White	FB	77.10	83.02	84.40	87.16	89.71
	2-SB	78.81	83.60	86.09	88.70	89.82
	Marginalization	68.00	68.92	70.21	70.77	71.21
Destroyer-engine	FB	76.51	82.81	86.43	86.29	88.47
	2-SB	82.16	86.60	86.52	87.92	88.68
	Marginalization	66.64	67.26	68.30	68.48	69.97
Average	FB	77.57	82.55	86.01	87.30	89.02
	2-SB	80.12	84.04	87.11	88.93	90.05
	Marginalization	66.74	67.54	68.46	68.90	70.05