Fact · Dictation
1.1percentage points
Noise compensation experiments on the MALACH corpus reduced the word error rate by 1.1 percentage points.
Checked against its source on October 16, 2026
In the source's own words
“led to a 1.1% absolute reduction of the word error rate”
Speech recognition error analysis on the English MALACH corpus, Olivier Siohan et al., Interspeech 2004 (ISCA), 2004, Abstract.
Open the original (DOI 10.21437/Interspeech.2004-171)- What it measures
- The study reports absolute, not relative, reduction.
- What it does not show
- A modest gain on a difficult corpus. Results on other recordings may differ.