Peer-reviewed research

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 · DOI 10.48550/arXiv.2302.09044

Read the original

Read on October 16, 2026

What it is: The full text of the 2023 study of people who stutter and voice assistants or dictation. It contains the error breakdown, the endpointer truncation rates and the survey results that the abstract does not give.
How much weight it carries: Peer-reviewed at CHI 2023. This record points to the full-text PDF of the same paper as the catalogued abstract record, so the two describe one study. The error-type breakdown is from the second phase, which covers 41 participants with moderate or severe dysfluency.

Facts taken from this source

In the same study, the baseline endpointer cut off an average of 23.8 percent of utterances from people who stutter.
“the baseline endpointer model truncates on average 23.8% of utterances” (Section 4.2.1 (endpointer))
Of the people who stutter who were familiar with voice assistants, 61 percent named being cut off before they finished talking as a reason they used them less often.
“voice assistants cut off my speech before I finish talking’ (61.0%, n=36/59)” (Section 3 (survey))
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.
“Among all errors the ASR system made in Phase 2, 80.9% were word insertions, 17.5% substitutions, and only 1.6% deletions.” (Section 4 (error analysis))
The 2023 study says that in dictation, the user finishes by clicking a button again, instead of relying on an endpointer.
“they finish by manually clicking a button again, instead of relying on an endpointer” (Section 2.1, Dictation Systems)
The 2023 study notes that many dictation systems do not use an automatic endpointer, so people who stutter were cut off less often in dictation than with voice assistants.
“many dictation systems do not use an automatic endpointer” (Section 3 (survey))
The 2023 study describes an endpointer as a model that detects whether the user has finished speaking.
“detects whether the user has finished speaking” (Section 2.1, Speech Recognition Systems)
The authors found a strong correlation between the rate of inserted words and part-word repetitions, which ties many of the extra words to repeated sounds.
“The high rate of word insertions has a strong correlation with part-word repetitions” (Section 4 (error analysis))
Some smartphones let you hold a button while you speak and release it when you finish, which avoids automatic end-of-speech detection.
“Some smartphones (e.g., Apple iPhone) offer an alternative to automatic endpointing, where the user can push down a button (physical or virtual) while speaking and only release it when done.” (Section 4.2.1 (endpointer))

Pages that rest on this source

Peer-reviewed research

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 · DOI 10.48550/arXiv.2302.09044

Read the original

Read on October 16, 2026

What it is: The full text of the 2023 study of people who stutter and voice assistants or dictation. It contains the error breakdown, the endpointer truncation rates and the survey results that the abstract does not give.
How much weight it carries: Peer-reviewed at CHI 2023. This record points to the full-text PDF of the same paper as the catalogued abstract record, so the two describe one study. The error-type breakdown is from the second phase, which covers 41 participants with moderate or severe dysfluency.

Facts taken from this source

In the same study, the baseline endpointer cut off an average of 23.8 percent of utterances from people who stutter.
“the baseline endpointer model truncates on average 23.8% of utterances” (Section 4.2.1 (endpointer))
Of the people who stutter who were familiar with voice assistants, 61 percent named being cut off before they finished talking as a reason they used them less often.
“voice assistants cut off my speech before I finish talking’ (61.0%, n=36/59)” (Section 3 (survey))
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.
“Among all errors the ASR system made in Phase 2, 80.9% were word insertions, 17.5% substitutions, and only 1.6% deletions.” (Section 4 (error analysis))
The 2023 study says that in dictation, the user finishes by clicking a button again, instead of relying on an endpointer.
“they finish by manually clicking a button again, instead of relying on an endpointer” (Section 2.1, Dictation Systems)
The 2023 study notes that many dictation systems do not use an automatic endpointer, so people who stutter were cut off less often in dictation than with voice assistants.
“many dictation systems do not use an automatic endpointer” (Section 3 (survey))
The 2023 study describes an endpointer as a model that detects whether the user has finished speaking.
“detects whether the user has finished speaking” (Section 2.1, Speech Recognition Systems)
The authors found a strong correlation between the rate of inserted words and part-word repetitions, which ties many of the extra words to repeated sounds.
“The high rate of word insertions has a strong correlation with part-word repetitions” (Section 4 (error analysis))
Some smartphones let you hold a button while you speak and release it when you finish, which avoids automatic end-of-speech detection.
“Some smartphones (e.g., Apple iPhone) offer an alternative to automatic endpointing, where the user can push down a button (physical or virtual) while speaking and only release it when done.” (Section 4.2.1 (endpointer))