Question · Languages and accents
How accurate is speech recognition in Ukrainian?
On our Mac benchmark, Ukrainian read speech is transcribed with a word error rate of 5.65 percent by Parakeet and 6.83 percent by Whisper large-v3-turbo. That ranks Ukrainian eighth of the 25 languages tested.
Updated October 19, 20263 min read11 cited facts, each traced to its source
Ukrainian is recognized well on a Mac: about 6 wrong words per 100 on read speech. In our October 2026 benchmark on a Mac with an M2 Pro, Parakeet TDT 0.6B v3 had a word error rate of 5.65 percent 1 and Whisper large-v3-turbo had 6.83 percent 2, on 100 Ukrainian recordings from the FLEURS test set. We have no Apple SpeechTranscriber result for Ukrainian, so the comparison is between those two open engines.
Parakeet is about one point better than Whisper in Ukrainian
Parakeet leads by a little more than a point. On 100 recordings that is a modest but plausible edge, and it matches the published picture below. Speed is also on Parakeet's side: about 21 times faster than real time 3, against about 13 times for Whisper 4. Both are quick enough for dictation. The speed dataset covers every language.
Our measurement of the engines in Ukrainian
Lower is better in this table.
| Engine | Word error rate (WER %) | Source |
|---|---|---|
| Parakeet TDT v3 | 5.65 | Calculated 1 |
| Whisper large-v3-turbo | 6.83 | Calculated 2 |
The transcripts behind each row are on the on-device benchmark page.
Published results show the same order, with older Whisper behind
NVIDIA reports 6.79 percent for Parakeet on FLEURS Ukrainian 5, while OpenAI's paper gives 8.6 percent for the older Whisper large-v2 6. Whisper large-v3-turbo, which we ran, is better than that older model. On CoVoST 2 Ukrainian, Parakeet had 5.10 percent 7. Different test sets use different recordings and scoring, so keep comparisons inside one test.
| System | Kind | Test set | Speech | Word error rate (WER %) | Source |
|---|---|---|---|---|---|
| NVIDIA Parakeet TDT 0.6B v3 | Open model | CoVoST 2 | 5.1 | NVIDIA, 2025 7 | |
| Parakeet TDT 0.6B v3 | Open model | FLEURS | Clean read speech | 5.65 | Calculated 1 |
| NVIDIA Parakeet TDT 0.6B v3 | Open model | FLEURS | Clean read speech | 6.79 | NVIDIA, 2025 5 |
| Whisper large-v3-turbo | Open model | FLEURS | Clean read speech | 6.83 | Calculated 2 |
| Whisper large-v2 | Open model | FLEURS | Clean read speech | 8.6 | Radford et al., 2022 6 |
All published rows are on the published accuracy page.
Ukrainian ranks eighth of 25 languages
Taking the best engine in each language, Ukrainian is eighth of the 25 languages in the benchmark. It sits just behind French (5.29 percent with Parakeet 8) and German (4.84 percent 9), and behind Russian (4.59 percent 10), and well ahead of Croatian and Slovenian. It is a mid-resource language that does very well, which shows that a smaller language is not always a harder one.
What to expect when you dictate in Ukrainian
In a quiet room, expect roughly 6 errors per 100 words, a bit more editing than in Russian, Portuguese or Spanish but still light. Parakeet is the first engine to try, with Whisper as the alternative if you prefer its features. We found no published fact that lists Ukrainian among the built-in Mac or Windows dictation languages, so check your own system's language list before relying on it. Claude Code's voice dictation documentation lists 20 languages, from Czech to Ukrainian 11. Our page on dictation languages compares platforms.
How to choose an engine for Ukrainian
Parakeet is the better first choice on the numbers: lower error and higher speed in our test, and a lower figure in NVIDIA's own FLEURS result than Whisper large-v2 had. That is not a reason to skip your own check. Dictate a page of your usual text through each engine you can run and count the corrections, paying attention to names, numbers and any Russian words that slip in. A few percent of difference in the table can disappear once your microphone and vocabulary enter, so let your own text decide. The published accuracy page shows cloud services on similar tests.
What these numbers do not show
The recordings are people reading prepared sentences in a quiet room, scored on one Mac with 100 recordings. Spontaneous speech, regional accents, noise, Russian words mixed into speech (surzhyk) and domain vocabulary are not in the test, and we did not measure Apple's model. For the measure itself, see the guide on how accurate speech recognition is.
Our guide to European languages in speech to text helps you choose an engine, and the languages overview covers other languages. For comparison, read the same question for Russian and Croatian.
Evidence
Every number on this page, with where it comes from. Open a fact to see the exact words in its source.
- 1.
- 2.
- 3.
- 4.
- 5.
- 6.
- 7.
- 8.
- 9.
- 10.
- 11.
The data behind these numbers
- Speech Recognition Accuracy: Word Error Rates8 of this page's figures, 673 data points in all
- On-Device Speech Recognition Benchmark: 25 Languages5 of this page's figures, 56 data points in all
- Speech Recognition Speed: Real-Time Factor Data2 of this page's figures, 66 data points in all
- How Many Languages Speech Recognition Supports1 of this page's figures, 55 data points in all
Go deeper
Related questions
- Does dictation work for both Brazilian and European Portuguese?
- Does speech recognition understand Swiss German?
- How accurate is speech recognition in Bulgarian?
- How accurate is speech recognition in Croatian?
- How accurate is speech recognition in Czech?
- How accurate is speech recognition in Danish?
- How accurate is speech recognition in Dutch?
- How accurate is speech recognition in English?