. Author manuscript; available in PMC: 2021 Mar 18.

Published in final edited form as: IEEE/ACM Trans Audio Speech Lang Process. 2019 Aug 12;27(11):1839–1848. doi: 10.1109/taslp.2019.2934319

TABLE VI.

PESQ scores for different two-stage and single-stage DFNs, LSTMs and BLSTMs in simulated reverberant conditions.

T₆₀ (s)		0.3		0.6		0.9
TIR (dB)		−12	−6	−12	−6	−12	−6	Average
Unprocessed		1.40	1.51	1.49	1.51	1.56	1.56	1.50
DFN	Single-stage Mapping	1.77	2.07	1.59	1.91	1.46	1.77	1.76
	Single-stage Masking	1.97	2.29	1.76	2.07	1.62	1.91	1.94
	Mapping+Mapping	1.84	2.15	1.65	1.95	1.52	1.80	1.82
	Mapping+Masking	1.95	2.28	1.75	2.04	1.62	1.89	1.92
	Masking+Mapping	1.96	2.27	1.73	2.04	1.58	1.87	1.91
	Masking+Masking	2.06*	2.40*	1.82*	2.14*	1.67*	1.97*	2.01*
LSTM	Single-stage Mapping	1.97	2.23	1.77	2.04	1.63	1.90	1.92
	Single-stage Masking	2.12	2.42	1.88	2.18	1.73	2.02	2.06
	Mapping+Mapping	1.97	2.23	1.76	2.04	1.63	1.90	1.92
	Mapping+Masking	2.08	2.37	1.85	2.13	1.70	1.97	2.02
	Masking+Mapping	2.05	2.33	1.81	2.09	1.66	1.94	1.98
	Masking+Masking	2.14	2.46*	1.88	2.19	1.72	2.02	2.07
BLSTM	Single-stage Mapping	2.05	2.31	1.85	2.13	1.71	1.99	2.01
	Single-stage Masking	2.18	2.47	1.97	2.26	1.81	2.10	2.13
	Mapping+Mapping	2.14	2.42	1.89	2.18	1.68	1.99	2.05
	Mapping+Masking	2.17	2.48	1.93	2.23	1.74	2.06	2.10
	Masking+Mapping	2.22	2.48	1.94	2.24	1.76	2.06	2.12
	Masking+Masking	2.26*	2.57	2.00*	2.31*	1.82*	2.14*	2.18*