Fact · Speech recognition
1011word vocabulary
Carnegie Mellon's Harpy system recognized speech from male and female speakers on a 1,011-word document retrieval task, with a semantic error of 5 percent.
Checked against its source on October 17, 2026
In the source's own words
“It recognizes speech from male and female speakers using a 1011-word-vocabulary document retrieval task. Semantic error is 5%”
Multi-System Approach to Speech Understanding, Raj Reddy, NASA Technical Reports Server (NTRS), 1993, Results paragraph.
Open the original- What it measures
- Reddy reports semantic error, which measures errors in the meaning of the result. It is not a word error rate, so it cannot be compared directly with later word error rates.
- What it does not show
- One system on one limited document retrieval task, reported by a member of its team.