Peer-reviewed research

Speech and Speech Recognition during Dictation Corrections

Keith Vertanen, Interspeech 2006 (ICSLP), Pittsburgh, 2006

Read on October 16, 2026

What it is: A study of how people correct errors while dictating to a computer. Users respoke corrections in a simulated dictation interface, and the paper measures how often those spoken corrections were misrecognized in turn.
How much weight it carries: Peer-reviewed conference paper from the Cambridge Cavendish Laboratory, hosted by the author. The interface was simulated and the recognizer is from 2006, so the rates are historical. The paper's observation about correction time is the part that generalizes.

Facts taken from this source

When people correct dictation errors by speaking again, the spoken correction is often misrecognized as well, which costs time.
“often spoken corrections are themselves misrecognized, costing the user time and testing their patience” (Abstract)
Vertanen reports that when dictating with a large-vocabulary recognizer, correcting errors takes up most of the task time.
“When using a large vocabulary continuous speech recognizer to dictate text, correcting errors dominates task time” (Introduction)