Page 33 - 《应用声学》2022年第1期

第 41 卷 第 1 期 韩哲等： 分布式无线声传感网加权预测误差语声去混响方法 29

Ref SN-WPE Cen-WPE DWPE
7
3.0 1.0 6
PESQ 2.5 STOI 0.8 SRMR/dB 5 4

2.0 3
0.6
ͯᎶ 1 ͯᎶ 2 ͯᎶ 1 ͯᎶ 2 ͯᎶ 1 ͯᎶ 2
(a) 450 msຉ־ PESQ (b) 450 msຉ־ STOI (c) 450 msຉ־ SRMR
7
3.0 1.0 6
PESQ 2.5 STOI 0.8 SRMR/dB 5 4

2.0 3
0.6
ͯᎶ 1 ͯᎶ 2 ͯᎶ 1 ͯᎶ 2 ͯᎶ 1 ͯᎶ 2
(d) 650 msຉ־ PESQ (e) 650 msຉ־ STOI (f) 650 msຉ־ SRMR

3.0 1.0 6
PESQ 2.5 STOI 0.8 SRMR/dB 5 4

2.0 3
0.6
ͯᎶ 3 ͯᎶ 4 ͯᎶ 3 ͯᎶ 4 ͯᎶ 3 ͯᎶ 4
(g) 900 msຉ־ PESQ (h) 900 msຉ־ STOI (i) 900 msຉ־ SRMR
3.0 1.0 6
PESQ 2.5 STOI 0.8 SRMR/dB 5 4

2.0 3
0.6
ͯᎶ 3 ͯᎶ 4 ͯᎶ 3 ͯᎶ 4 ͯᎶ 3 ͯᎶ 4
(j) 1100 msຉ־ PESQ (k) 1100 msຉ־ STOI (l) 1100 msຉ־ SRMR
图 5 随混响时间增加各算法在不同位置下的性能对比测试
Fig. 5 Performance comparison of the three methods along increasing reverberation times at different positions

Ref SN-WPE Cen-WPE DWPE
2.6 0.9 5
2.4 0.8
PESQ 2.2 STOI 0.7 SRMR/dB 4 3
2.0
1.8
0.6
10 dB 15 dB 20 dB 25 dB 30 dB 10 dB 15 dB 20 dB 25 dB 30 dB 10 dB 15 dB 20 dB 25 dB 30 dB
(a) 650 msຉ־ PESQ (b) 650 msຉ־ STOI (c) 650 msຉ־ SRMR
0.9
2.4 4
PESQ 2.2 STOI 0.8 SRMR/dB 3
0.7
2.0
1.8 0.6
10 dB 15 dB 20 dB 25 dB 30 dB 10 dB 15 dB 20 dB 25 dB 30 dB 10 dB 15 dB 20 dB 25 dB 30 dB
(d) 900 msຉ־ PESQ (e) 900 msຉ־ STOI (f) 900 msຉ־ SRMR
图 6 不同噪声情况下各算法性能对比测试
Fig. 6 Performance comparison of the three methods with different levels of noise