How accurate is speech recognition in Russian?

Russian is recognized well by on-device engines: about 5 wrong words per 100 on read speech. In our October 2026 benchmark on a Mac with an M2 Pro, Whisper large-v3-turbo had a word error rate of 4.59 percent ¹ and Parakeet TDT 0.6B v3 had 4.74 percent ², on 100 Russian recordings from the FLEURS test set. We have no Apple SpeechTranscriber result for Russian, so the comparison here is between those two open engines.

Whisper and Parakeet are tied in Russian

The two engines differ by about 0.15 points, far less than the noise in 100 recordings, so neither is more accurate in any way you would notice. Parakeet is somewhat faster at about 21 times faster than real time ³, against about 14 times for Whisper ⁴. Both keep far ahead of speech, so choose on features and on what your app supports. The speed dataset covers every language.

Our measurement of the engines in Russian

Lower is better in this table.

The transcripts behind each row are on the on-device benchmark page.

Published results agree on 5 to 6 percent for FLEURS

NVIDIA reports 5.51 percent for Parakeet on FLEURS Russian ⁵, and OpenAI's paper gives 5.6 percent for the older Whisper large-v2 ⁶. Those match our measurement to within about a point. The test set changes the result: on CoVoST 2 Russian, Parakeet had 3.00 percent ⁷, lower than on FLEURS. The recordings and the text rules differ, so numbers from different test sets should not be lined up against each other.

All published rows are on the published accuracy page.

Russian ranks fifth of 25 languages

Taking the best engine in each language, Russian is fifth of the 25 languages in the benchmark. Italian (2.52 percent ⁸), Spanish, Portuguese and English (3.92 percent ⁹) are ahead. Russian is just ahead of German (4.84 percent ¹⁰) and well ahead of Ukrainian (5.65 percent ¹¹), a close relative. Russian is among the languages with plenty of training audio, which probably helps, though the benchmark does not test that.

What to expect when you dictate in Russian

In a quiet room, expect roughly 5 errors per 100 words from Whisper or Parakeet, a short editing pass. Apple lists Russian (Russia) among the Mac's Dictation languages ¹² and in its on-device list ¹³, and Windows 11 voice typing lists Russian as one language with no region ¹⁴. Since we did not measure Apple's model in Russian, try the built-in tool first and move to Whisper or Parakeet if you see too many errors on names, numbers or technical terms. Our page on Russian voice typing covers the practical steps.

How to choose an engine for Russian

Because the two open engines are tied, pick the one your app already supports and test it on your own speech. Read a page of your usual text, once through each engine if you can, and count the corrections. Your microphone, your pace and your vocabulary (technical terms, names, English loanwords) will move the result more than the 0.15-point gap in the table. If you dictate long documents, favor the engine that handles punctuation the way you like. If you dictate short messages, favor the faster one. The published accuracy page shows how cloud services compare on the same kind of test.

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, mixed Russian and English, and domain vocabulary are not in the test. Word error rate also counts inflected forms as wrong words, which matters in a language with many word endings. 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 Ukrainian and Portuguese.