Deep Learning-Based Amplitude Fusion for Speech Dereverberation

Table 5

PESQ scores of various methods under different levels of reverberation.

Type		—	Mapping	Masking			Masking + masking			Mapping + masking
Method		MIX	MAP	IRM	IAM	DCC	IAM + IRM	IRM + DCC	IAM + DCC	MAP + IRM	MAP + IAM	MAP + DCC
Far field	S	1.61	1.82	1.71	2.13	2.07	1.91	1.97	2.14	1.99	2.10	2.12
	M	1.19	1.49	1.25	1.47	1.51	1.35	1.39	1.51	1.45	1.57	1.59
	L	1.17	1.45	1.22	1.44	1.48	1.32	1.35	1.47	1.38	1.51	1.53
Near field	S	2.14	2.08	2.26	2.74	2.51	2.52	2.58	2.62	2.62	2.62	2.59
	M	1.40	1.91	1.52	1.85	1.78	1.70	1.75	2.06	2.06	2.14	2.13
	L	1.37	1.89	1.47	1.85	1.82	1.67	1.72	1.95	1.95	2.06	2.08
AVE	1.48	1.77	1.57	1.91	1.86	1.74	1.79	1.91	1.91	2.00	2.01