. Author manuscript; available in PMC: 2020 Jul 13.

Published in final edited form as: IEEE Access. 2019 Jun 12;7:78421–78433. doi: 10.1109/access.2019.2922370

TABLE 3.

Performance evaluation in unseen SNR condition for speech degraded by machinery noise. The neural network models were trained at −5, 0 and+5dB SNR and tested at−2.5,+2.5 and+7.5dB respectively.

Objective Measure	Method	SNR= −2.5 (dB)	SNR=2.5(dB)	SNR=7.5(dB)
PESQ	Noisy speech	1.1496	1.1819	1.2468
	CNN-DAE	1.19	1.6437	1.6883
	Proposed	1.2679	1.7711	1.8441
LSD	Noisy speech	18.5031	17.6985	16.3696
	CNN-DAE	14.9512	14.4864	14.3056
	Proposed	14.0192	13.7728	13.6160
SegSNR	Noisy speech	4.3988	5.4868	7.1029
	CNN-DAE	6.0734	6.8436	7.2104
	Proposed	6.4612	6.9063	7.2141