Fact · Dictation
18.8percent word error rate
In a 2024 baseline test, the OpenAI Whisper model had a word error rate of 18.8 percent on whispered speech.
Checked against its source on October 16, 2026
In the source's own words
“achieving a Word Error Rate (WER) of 18.8% and a Character Error Rate (CER) of 4.24% on whispered speech”
Leveraging Self-Supervised Models for Automatic Whispered Speech Recognition, Aref Farhadipour et al., arXiv (Electrical Engineering and Systems Science, Audio and Speech Processing), 2024, Abstract.
Open the original- What it measures
- The authors fine-tuned WavLM on whispered and normal speech from the wTIMIT and CHAINS datasets, then compared it with the baseline Whisper model on the same whispered speech.
- What it does not show
- Read-aloud test sentences from research corpora, in English dialects, tested in 2024. It does not measure current dictation apps, everyday whispering in a room, or a different microphone distance.