Subband DCT and EMD Based Hybrid Soft Thresholding for Speech Enhancement

Table 1

Comparison of the SNR, AvgSegSNR, and PESQ improvements of different denoising methods for a high range of SNR values (white noise).

A
Input SNR (dB)	Output SNR (dB)
	WP [3]		DCT [11]	Soft DCT [4]	𝑈EMD (𝜆opt)
0	4.86		6.44	6.42	7.91
5	8.92		10.03	10.07	11.22
10	12.52		13.61	13.95	14.98
15	15.64		17.26	18.05	18.87
25	20.95		24.92	26.78	27.18
30	23.27		28.86	31.32	31.51
B
Input AvgSegSNR (dB)	Output AvgSegSNR (dB)
	WP [3]		DCT [11]	Soft DCT [4]	𝑈EMD (𝜆opt)
−4.111	−1.933		−0.669	−0.317	0.779
−1.341	0.926		2.01	2.246	3.166
2.079	3.666		4.823	5.187	6.078
5.758	6.504		7.83	8.472	9.294
13.837	11.64		14.516	15.902	16.394
18.002	13.71		18.092	19.679	19.998
C
Input SNR (dB)	PESQ
	Input	WP [3]	DCT [11]	Soft DCT [4]	𝑈EMD (𝜆opt)
0	1.06	1.27	1.38	1.36	1.74
5	1.36	1.58	1.78	1.76	2.07
10	1.69	1.95	2.19	2.14	2.39
15	2.04	2.31	2.58	2.52	2.71
25	2.81	2.86	3.32	3.21	3.32
30	3.21	3.06	3.64	3.53	3.66