Skip to main content

Advertisement

Table 13 Experimental results on the simulated reverberant data by using NN training dataset ā€˜5s.20uā€™

From: Single-channel dereverberation by feature mapping using cascade neural networks for robust distant speaker identification and speech recognition

Method Frame selection Acoustic model Digit accuracy (%)
Set A Set B Avg. set A Avg. set B Avg.
Office EvHall InCar LivingR Lounge JPstyR ConfR BathR
Proposed (24 NNs) skip1 8-1-0 Clean 98.6 66.3 98.8 89.8 90.1 95.2 92.7 91.5 88.4 92.4 90.4
Multi-rt 97.4 95.6 97.8 97.4 92.1 97.3 96.1 94.5 97.1 95.0 96.1
Cln+mlt-rt 97.6 94.2 98.0 97.0 93.5 97.7 96.5 94.2 96.7 95.5 96.1
Proposed (24 NNs) + CMN skip1 8-1-0 Clean 97.7 67.8 98.4 86.0 93.3 93.4 92.4 86.6 87.5 91.4 89.5
Multi-rt 96.6 94.7 96.8 96.2 94.7 96.5 95.9 93.6 96.1 95.2 95.7
Cln+mlt-rt 97.1 94.7 97.1 95.2 94.1 96.2 96.2 92.5 96.0 94.7 95.4