Table 5 SER after HEQ evaluated by multi-style trained recognition models
From: Stereo-based histogram equalization for robust speech recognition
| Â | 16 | 64 | 256 |
|---|---|---|---|
Noisy unprocessed | 27.49 | ||
THEQ | 27.42 | ||
QHEQ | 27.45 | ||
PHEQ | 26.88 | ||
Soft HEQ | 16.42 | 15.8 | 17.41 |
Soft HEQ then VTS | 20.27 | 19.47 | 19.4 |
Soft HEQ then SPLICE | 16.29 | 16.29 | 25.88 |