Fact · Voice and AI
13.63percent word error rate
On a dysarthric speech test set, the recognizer alone had an overall word error rate of 13.63 percent: 0 percent on sentences it transcribed without error, and 21.98 percent on sentences that had errors.
Checked against its source on October 17, 2026
In the source's own words
“The ASR baseline achieves an overall WER of 13.63% (0% on the NoErr group and 21.98% on the Err group).”
Towards Robust Dysarthric Speech Recognition: LLM-Agent Post-ASR Correction Beyond WER, Xiuwen Zheng et al., arXiv (ICASSP 2026), 2026, Results, Table 1.
Open the original- What it measures
- The SAP-Hypo5 test set has 2,647 utterances. The authors split it into 1,080 sentences the recognizer got right (NoErr) and 1,567 it got wrong (Err), and report every correction method separately for each group.
- What it does not show
- Dysarthric speech is harder to recognize than typical dictation, and the test uses one recognizer. The split shows where correction has room to help, not how a dictation app would perform.