Fact · Accessibility and health
80.9percent of recognition errors
In the Apple stuttering study, 80.9 percent of the speech recognizer's errors were extra inserted words, and only 1.6 percent were missed words.
Checked against its source on October 16, 2026
In the source's own words
“Among all errors the ASR system made in Phase 2, 80.9% were word insertions, 17.5% substitutions, and only 1.6% deletions.”
From User Perceptions to Technical Improvement: Enabling People Who Stutter to Better Use Speech Recognition (full text), Colin Lea et al., Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (arXiv preprint), 2023, Section 4 (error analysis).
Open the original (DOI 10.48550/arXiv.2302.09044)- What it measures
- Phase 2 of the study covered 41 people who stutter with moderate or severe dysfluency. The recognizer was a consumer-grade ASR system.
- What it does not show
- The breakdown covers Phase 2 only and one consumer recognizer, and the authors tie most insertions to part-word and whole-word repetitions.