Skip to main content

Table 15 Distant-talking speaker identification rates for evaluation data (%)

From: Deep neural network-based bottleneck feature and denoising autoencoder-based dereverberation for distant-talking speaker identification

Method RT60 of test data (s) (RWCP data) Ave.
  0.38 0.47 0.60 0.78 1.30  
(a) Conventional methods
CMN 79.70 76.05 75.55 74.40 75.75 76.29
MCLMS-SS 82.25 79.70 78.75 78.05 81.30 80.01
MSLP-SS 82.85 78.60 78.50 78.00 75.70 78.73
BF-MLP 72.35 69.30 64.05 64.90 63.25 66.70
(b) DNN-based feature transformation methods
BF-DNN 87.90 84.95 82.45 84.00 82.15 84.29
DAE 92.10 89.70 87.60 89.45 88.10 89.39
DAE + BF-DNN 94.20 92.20 90.65 91.95 90.70 91.94