Peer-reviewed research
Comparing Human and Machine Errors in Conversational Speech Transcription
Andreas Stolcke and Jasha Droppo, Proceedings of Interspeech 2017, ISCA, 2017
Read the originalRead on October 17, 2026
- What it is
- A 2017 Microsoft study that compares the errors of a conversational telephone speech recognizer with the errors of a commercial transcription vendor on the same audio.
- How much weight it carries
- Peer-reviewed. It covers conversational telephone speech on one benchmark, so the error patterns may differ for dictation or other audio.
Facts taken from this source
- 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.
“the most frequent substitution, deletion and insertion error types of both outputs show a high degree of overlap” (Abstract)
- In a 2017 comparison of machine and vendor transcripts on telephone speech, the automatic recognizer tended to confuse filled pauses such as uh with backchannels such as uhhuh, and humans tended not to make that error.
“the automatic recognizer tends to confuse filled pauses ("uh") and backchannel acknowledgments ("uhhuh"). Humans tend not to make this error” (Abstract)