Fact · Dictation
14percent relative gain
The Aurora standardization tests, as cited by Bauer et al., found a 14% average relative gain in recognition of noisy speech sampled at 16 kHz instead of 8 kHz.
Checked against its source on October 16, 2026
In the source's own words
“ASR performance gain of 14 % averaged over a weak, medium, and high recording mismatch between training and test conditions, when recognizing noisy speech sampled at 16 instead of 8 kHz”
Automatic Recognition of Wideband Telephone Speech with Limited Amount of Matched Training Data, Patrick Bauer et al., EUSIPCO 2014, European Signal Processing Conference, 2014, Introduction.
Open the original- What it measures
- The sentence reports earlier Aurora results for noise-robust distributed speech recognition. The authors cite them, so the value is their summary of other work.
- What it does not show
- A relative gain averaged across mismatch conditions, reported secondhand. It is evidence that wideband helps in noise, not a measurement of a specific headset.