TABLE 7.

Performance evaluation in unseen SNR condition for noisy speech (traffic noise at 0 dB) mixed with alert signal. The neural network models were trained at −5 dB and +10 dB SNR, tested at −10 dB and +15 dB respectively.

SNR dB	Method	Objective measures
		SAR	SDR	SIR	CSII
	Unprocessed	−19.82	−19.32	7.21	0.2303
−10	Conv.	−19.88	−20.27	5.21	0.2309
	Conv + Separation	−15.77	−15.93	9.92	0.3981
	Unprocessed	−2.32	−2.36	22.27	0.439
+15	Conv.	−2.64	−2.66	22.05	0.4511
	Conv + Separation	−1.71	−1.77	23.59	0.5334