Preprint
Sorry, I Didn’t Catch That: How Speech Models Miss What Matters Most
Kaitlyn Zhou et al., arXiv, 2026
Read the originalRead on October 17, 2026
- What it is
- A 2026 study of how 15 speech models from OpenAI, Deepgram, Google and Microsoft transcribe short, high-stakes utterances: US street names spoken by US participants.
- How much weight it carries
- Preprint, not peer reviewed as of checking. It tests one task, street names, so its error rate does not generalize to all speech.
Facts taken from this source
- In a 2026 study, 15 speech models from OpenAI, Deepgram, Google and Microsoft had an average transcription error rate of 44 percent on US street names spoken by US participants.
“We evaluate 15 models from OpenAI, Deepgram, Google, and Microsoft on recordings from linguistically diverse U.S. speakers and find an average transcription error rate of 44%.” (Abstract)