Fact · Transcription
5.5percent word error rate
Fine-tuning Whisper Large V3 Turbo on labeled filled pauses gave 5.5 percent word error rate, against 6.2 percent when hesitations were removed.
Checked against its source on October 17, 2026
In the source's own words
“fine-tuning Whisper Large V3 Turbo with the "Extra" scheme yielded a 5.5% WER, an 11.3% relative improvement over the "Pure" scheme (6.2% WER).”
Acoustically Precise Hesitation Tagging Is Essential for End-to-End Verbatim Transcription Systems, Jhen-Ke Lin et al., arXiv (SLaTE 2025 workshop), 2025, Abstract.
Open the original- What it measures
- The authors ran this experiment after the challenge deadline, with the same corpus and the same two labeling schemes.
- What it does not show
- Word error rate here is measured against the verbatim reference the authors used, which includes the hesitations. A different reference would give different numbers.