Microphone mismatch
-------------------

Training Set 1, 16 kHz, utt. detection.

Test Set 1 : 14.0%
Test Set 8 : 52.7%
------------------
Relative degradation: 276%

Training Set 1, 8 kHz, utt. detection.

Test Set 1 : 16.2%
Test Set 8 : 37.4%
------------------
Relative degradation: 131%

Average degradation: (276 + 131)/2 = 204% <******


Model-mismatch
---------------

Training Set 1, 16 kHz, utt. detection.

Perfectly matched condition:
Test Set 1 : 14.0%

Mis-matched conditions:
Test Set 2  : 56.6%
Test Set 3  : 57.2%
Test Set 4  : 54.3%
Test Set 5  : 60.0%
Test Set 6  : 57.7%
Test Set 7  : 62.9%
Test Set 8  : 52.7%
Test Set 9  : 74.3%
Test Set 10 : 74.3%
Test Set 11 : 67.5%
Test Set 12 : 75.6%
Test Set 13 : 71.9%
Test Set 14 : 74.7%
-------------------
Average     : 64.4%

Relative degradation: 360%

Training Set 1, 8 kHz, utt. detection.

Perfectly matched condition:
Test Set 1 : 16.2%

Mis-matched conditions:
Test Set 2  : 49.6%
Test Set 3  : 62.2%
Test Set 4  : 58.7%
Test Set 5  : 58.2%
Test Set 6  : 61.5%
Test Set 7  : 61.7%
Test Set 8  : 37.4%
Test Set 9  : 59.7%
Test Set 10 : 69.8%
Test Set 11 : 67.7%
Test Set 12 : 72.2%
Test Set 13 : 68.3%
Test Set 14 : 67.9%
-------------------
Average     : 61.2%

Relative degradation: 278% <********

Average degradation: (360 + 278)/2 = 319%