EURASIP Journal on Audio, Speech, and Music Processing

Table 5 The average PESQ scores increment and segSNRs increment with and without postfiltering for the typical DNN-based speech enhancement methods under the assigned types of noise

From: Low-complexity artificial noise suppression methods for deep learning-based speech enhancement algorithms

Objective metrics	ΔPESQ				ΔsegSNR
Noise type	White	Babble	f16	Factory	White	Babble	f16	Factory
CRN	1.02	0.53	0.71	0.74	6.63	4.58	4.35	4.96
SPP-MMSE	1.08	0.55	0.75	0.75	7.39	5.22	5.00	5.75
SPP-proposed-1	1.16	0.61	0.89	0.81	7.83	5.58	5.69	6.07
SPP-proposed-2	1.16	0.64	0.88	0.85	7.85	5.59	5.59	6.18
SPP-proposed-3	1.17	0.6	0.86	0.81	8.00	5.72	5.68	6.34
DCN	0.83	0.54	0.75	0.68	5.37	4.82	5.17	5.34
SPP-MMSE	0.90	0.56	0.79	0.69	5.77	5.31	5.95	5.99
SPP-proposed-1	0.99	0.61	0.90	0.72	6.48	5.71	6.69	6.32
SPP-proposed-2	0.98	0.65	0.89	0.79	6.44	5.76	6.55	6.41
SPP-proposed-3	1.02	0.59	0.87	0.74	6.52	5.76	6.61	6.56
GRN	1.02	0.6	0.77	0.67	6.64	5.01	5.21	5.57
SPP-MMSE	1.07	0.61	0.8	0.67	6.89	5.28	5.8	5.86
SPP-proposed-1	1.18	0.69	0.90	0.72	7.65	5.77	6.50	6.22
SPP-proposed-2	1.18	0.71	0.88	0.78	7.64	5.81	6.39	6.35
SPP-proposed-3	1.20	0.67	0.89	0.73	7.78	5.86	6.51	6.47
DARCN	1.06	0.71	0.92	0.86	6.42	5.29	5.76	5.76
SPP-MMSE	1.15	0.71	0.95	0.88	6.88	5.72	6.48	6.43
SPP-proposed-1	1.17	0.80	1.02	0.92	7.32	5.98	6.88	6.65
SPP-proposed-2	1.13	0.76	0.95	0.90	6.89	5.59	6.42	6.30
SPP-proposed-3	1.20	0.78	1.01	0.93	7.52	6.12	6.98	6.92

Back to article page