An Improved Unsupervised Single-Channel Speech Separation Algorithm for Processing Speech Sensor Signals

Table 3

The results of the enhancement methods (SNR, PRSQ, STOI).

		Scene
Method	Index
CNMF + JADE	SNR	13.10	9.20	11.19	8.44	7.03	8.32	5.68
	PESQ	3.02	2.40	2.69	2.05	2.20	2.26	2.35
	STOI	0.95	0.90	0.92	0.88	0.85	0.89	0.74
CNMF + JADE+ spectral subtraction	SNR	5.22	5.36	6.16	4.76	4.07	4.76	3.55
	PESQ	1.97	1.95	1.70	1.89	1.88	1.89	1.27
	STOI	0.79	0.82	0.70	0.79	0.72	0.80	0.75
CNMF + JADE+ Wiener filtering	SNR	-2.24	-2.04	-1.02	-1.96	-2.13	-1.96	-1.62
	PESQ	2.36	2.33	2.96	2.13	2.19	2.13	2.49
	STOI	0.73	0.77	0.72	0.72	0.69	0.72	0.72
CNMF + JADE+ Kalman filtering	SNR	3.27	3.42	3.19	3.05	2.84	4.0	2.05
	PESQ	2.03	2.25	2.05	2.00	1.97	2.00	2.04
	STOI	0.83	0.88	0.69	0.82	0.78	0.83	0.57
CNMF + JADE+ wavelet transform	SNR	12.02	11.10	15.9	8.35	7.23	8.37	7.40
	PESQ	2.49	2.46	1.96	1.70	2.25	2.15	1.90
	STOI	0.82	0.92	0.81	0.84	0.88	0.91	0.76