On the Whisper Large V3 Turbo experiment, labeling filled pauses lowered the word error rate by 0.7 percentage points.

Checked against its source on October 17, 2026

What it measures: The two word error rates come from the same abstract sentence, one for each labeling scheme on the same Whisper model. The gap is a difference of the two published rates, not a new measurement.
How it is calculated: 6.2 - 5.5 = 0.7 percentage points, using the two word error rates in the same abstract sentence
What it does not show: A difference of two rates from one experiment. It inherits that experiment's reference transcripts and its learner-English corpus.

Calculated from