Fact · Speech recognition
12.94percent
In the formatted accuracy of on-device speech recognition: punctuation and capitals counted dataset, the 25 published figures for “Whisper large-v3-turbo” have a median of 12.94 percent, ranging from 5.68 to 71.96.
Calculated from 25 data points on October 19, 2026
- What it measures
- Each figure is one row of the dataset: a published measurement for a stated group, traced to its source. The median is the middle value of the 25 figures for this group.
- How it is calculated
- Median of 25 values (5.68, 6.13, 7.1, 7.45, 7.65, 7.71, 8.99, 9.28, 9.88, 10.09, 10.48, 12.58, 12.94, 13.3, 13.99, 14.21, 16.1, 16.9, 17.71, 19.28, 20.35, 20.52, 25.26, 26.71, 71.96) = 12.94.
- What it does not show
- FLEURS sentences are read aloud from Wikipedia, with the source's punctuation; optional commas and style choices count as errors for every engine alike, so compare engines within a language rather than reading the number as the editing a person needs.
A statistic of the dataset Dictation Punctuation and Capitalization Accuracy, calculated by code from its published rows.
Calculated from
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Bulgarian sentences by 17.71 errors per 100 tokens, against 15.87 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Czech sentences by 13.30 errors per 100 tokens, against 11.49 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Danish sentences by 16.90 errors per 100 tokens, against 14.57 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written German sentences by 7.10 errors per 100 tokens, against 5.08 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Greek sentences by 16.10 errors per 100 tokens, against 11.96 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written English sentences by 7.71 errors per 100 tokens, against 3.92 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Spanish sentences by 5.68 errors per 100 tokens, against 2.53 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Estonian sentences by 20.52 errors per 100 tokens, against 19.91 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Finnish sentences by 9.88 errors per 100 tokens, against 7.73 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written French sentences by 9.28 errors per 100 tokens, against 6.62 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Croatian sentences by 14.21 errors per 100 tokens, against 11.34 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Hungarian sentences by 19.28 errors per 100 tokens, against 17.21 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Italian sentences by 7.65 errors per 100 tokens, against 2.73 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Lithuanian sentences by 26.71 errors per 100 tokens, against 25.03 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Latvian sentences by 20.35 errors per 100 tokens, against 18.64 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Maltese sentences by 71.96 errors per 100 tokens, against 72.35 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Dutch sentences by 8.99 errors per 100 tokens, against 6.59 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Polish sentences by 10.48 errors per 100 tokens, against 7.70 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Portuguese sentences by 7.45 errors per 100 tokens, against 3.28 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Romanian sentences by 13.99 errors per 100 tokens, against 11.79 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Russian sentences by 6.13 errors per 100 tokens, against 4.59 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Slovak sentences by 12.58 errors per 100 tokens, against 11.51 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Slovenian sentences by 25.26 errors per 100 tokens, against 23.51 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Swedish sentences by 12.94 errors per 100 tokens, against 9.95 for the words alone.
- In our October 2026 benchmark, counting punctuation and capital letters as a reader sees them, Whisper large-v3-turbo's text differed from the written Ukrainian sentences by 10.09 errors per 100 tokens, against 6.83 for the words alone.