Table 5 Effect of environment identification method for two-step environment-dependent DAE. WER of Dev. dataset (%)
From: Environment-dependent denoising autoencoder for distant-talking speech recognition
Use reverberation | SimData | RealData | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|
estimated by MSLP | Room 1 | Room 2 | Room 3 | Ave. | Room 1 | Ave. | ||||
Near | Far | Near | Far | Near | Far | Near | Far | |||
No | 5.16 | 7.23 | 7.32 | 16.79 | 7.05 | 20.35 | 10.65 | 42.30 | 41.90 | 42.10 |
Yes | 5.01 | 8.24 | 6.78 | 11.78 | 7.54 | 13.72 | 8.85 | 32.13 | 33.42 | 32.78 |