Fact · Speech recognition
In a 2017 comparison on conversational telephone speech, the most frequent substitution, deletion and insertion errors of a recognizer and of a commercial transcription vendor showed a high degree of overlap.
Checked against its source on October 17, 2026
In the source's own words
“the most frequent substitution, deletion and insertion error types of both outputs show a high degree of overlap”
Comparing Human and Machine Errors in Conversational Speech Transcription, Andreas Stolcke and Jasha Droppo, Proceedings of Interspeech 2017, ISCA, 2017, Abstract.
Open the original- What it measures
- The study compared a top research recognizer with a standard vendor transcription pipeline on the same audio.
- What it does not show
- One benchmark and one pair of systems. The one notable difference the authors report is that the recognizer confused filled pauses such as uh with acknowledgments such as uh-huh.