. Author manuscript; available in PMC: 2015 Jan 14.

Published in final edited form as: IEEE/ACM Trans Audio Speech Lang Process. 2014 Dec;22(12):1849–1858. doi: 10.1109/TASLP.2014.2352935

TABLE II.

Performance Comparisons Between Various Targets and Systems on −5 dB Mixtures. “MC-IRM” stands for Multi-Condition Training (on all five noises) and uses IRM as the Training Target

Target/System	Factory1			Babble			SSN			Engine			Oproom
Target/System	STOI	PESQ	SNR	STOI	PESQ	SNR	STOI	PESQ	SNR	STOI	PESQ	SNR	STOI	PESQ	SNR
Mixture	0.54	1.29	−5.00	0.55	1.42	−5.00	0.57	1.48	−5.00	0.57	1.41	−5.00	0.59	1.40	−5.00
IBM	0.66	1.49	6.63	0.63	1.50	3.98	0.72	1.45	8.71	0.78	1.53	13.24	0.77	1.81	12.24
TBM	0.65	1.33	5.19	0.62	1.32	3.08	0.72	1.45	8.71	0.77	1.52	6.16	0.76	1.60	6.38
IRM	0.67	1.75	8.27	0.63	1.64	4.39	0.73	1.87	10.81	0.80	2.17	15.66	0.79	2.19	15.33
FFT-MAG	0.66	1.73	5.45	0.62	1.50	3.80	0.72	1.76	5.18	0.76	2.02	6.09	0.74	2.01	5.84
FFT-MASK	0.68	1.77	7.59	0.65	1.65	5.52	0.74	1.87	7.58	0.78	2.16	9.73	0.77	2.15	9.89
GF-POW	0.67	1.80	8.23	0.62	1.63	5.98	0.72	1.85	8.62	0.76	2.06	9.83	0.74	2.14	9.31
MC-IRM	0.69	1.80	9.52	0.64	1.65	5.08	0.74	1.88	11.40	0.78	2.12	14.97	0.77	2.16	14.79
ASNA-NMF	0.60	1.55	5.62	0.57	1.53	4.21	0.64	1.61	5.69	0.70	1.84	7.04	0.68	1.81	7.08
SPEH	0.51	1.56	4.13	0.50	1.38	3.07	0.57	1.68	4.34	0.62	1.85	5.73	0.58	1.88	5.98