Skip to main content

Table 1 SDR improvements for each method under each simulated noise condition

From: Deficient-basis-complementary rank-constrained spatial covariance matrix estimation based on multivariate generalized Gaussian distribution for blind speech extraction

Methods Babble noise Station noise Traffic noise Cafe noise
ILRMA 6.1/– 6.2/– 4.7/– 6.4/–
IVE 6.2/5.8 6.2/5.8 5.4/5.2 6.4/5.9
BSSA 6.8/– 6.9/– 5.7/– 7.2/–
MWF1 6.1/– 6.9/– 5.8/– 7.0/–
MWF2 6.9/– 7.2/– 5.6/– 7.4/–
MNMF 2.7/2.7 3.5/3.5 2.6/2.6 2.8/2.8
ILRMA + MNMF 6.5/6.5 7.0/7.0 6.2/6.2 8.1/8.1
FastMNMF 1.7/1.7 2.6/2.5 2.9/2.8 2.6/2.6
ILRMA + FastMNMF 6.5/6.2 6.6/6.6 5.4/5.4 7.3/7.3
Fixed-basis RCSCME
 \(\rho =0.5\) 8.7/7.7 10.2/9.5 8.3/7.8 10.1/9.3
 \(\rho =1\) 8.6/7.7 9.9/9.4 8.0/7.7 9.9/9.2
 \(\rho =2\) 8.5/7.6 9.8/9.2 7.9/7.5 9.8/9.1
Proposed basis-optimizing RCSCME
 \(\rho =0.5\) 9.3/8.8 11.5/11.0 9.4/9.2 11.1/10.5
 \(\rho =1\) 9.1/8.6 11.2/10.8 9.1/9.0 10.9/10.4
 \(\rho =2\) 9.0/8.6 11.1/10.7 9.0/8.9 10.8/10.4
  1. Each term represents “best-iteration score [dB]/after-200-iteration (except for IVE; after-4000-iteration for IVE) score [dB].” Bold text indicates the maximum value