Ukrainian is written with its own Cyrillic alphabet. It has і, ї, є and ґ, which Russian does not, and it has no ы, э, ё or ъ. Shared letters do not always stand for the same sound: Ukrainian и and г are pronounced differently from the Russian letters of the same shape. A Ukrainian recording transcribed here comes out in Ukrainian orthography. It is not written with Russian letters, and it is not transliterated into Latin script.
Two details are worth knowing when you read the result. The letter ґ, the hard g, is rare and distinguishes a few pairs such as ґрати, bars, and грати, to play; speakers do not always keep the two sounds apart, and the transcript may show г where a dictionary has ґ. And the 2019 spelling rules changed some common words, so both проєкт and the older проект are in circulation. If your organisation follows one of them, a search and replace settles it.
Ukrainian uses an apostrophe inside words to show that the next vowel starts with a clear y sound after a hard consonant: п'ять, м'ясо, об'єкт, комп'ютер, сім'я. It is part of the spelling, not punctuation, and the transcript includes it. The practical catch is that several different apostrophe characters exist, straight and curly, and software treats them as different. If a search for a word with an apostrophe finds nothing in the text you pasted elsewhere, that is the usual reason.
Ukrainian spelling is close to pronunciation, with a few exceptions that have to be settled from grammar. Unstressed е and и sound nearly alike, so мене, me, and мине, will pass, differ mainly by context. The reflexive ending -ться is pronounced like -ця. The pairs у and в, і and й alternate depending on the neighbouring sounds, and these are written as the speaker said them. Ukrainian also has a vocative case, so a person addressed in a meeting appears as Олено, Андрію or пане Тарасе; search for a name by its stem to find every form.
Many recordings from Ukraine are not in one language from beginning to end. Surzhyk, everyday speech that mixes Ukrainian and Russian words and grammar, has no standard spelling: шо for що, да for так, щас for зараз. The model writes such speech in Ukrainian letters. Sometimes it keeps the word as it was said and sometimes it substitutes the standard Ukrainian one, so the text may be a little cleaner than the speech. If you need a strictly verbatim record, for linguistic research or for testimony, compare it with the audio.
Clear switches are a separate case: an interviewer asks in Ukrainian and the guest answers in Russian, or a speaker changes language partway through. The language is detected automatically and no setting is needed, but mixed recordings are where detection is least certain, and a stretch of one language can come out in the spelling of the other. We do not promise the same accuracy here as for a recording in one language. Recordings that are plainly in Ukrainian throughout do not have this problem.
Standard Ukrainian grew out of the central dialects around Kyiv and Poltava, and speech from that tradition, which is also the language of broadcasting and education, is recognised best. Western speech from Lviv or Ivano-Frankivsk in a normal register is transcribed well too, with regional words such as файно kept as spoken. Strong Transcarpathian or Hutsul dialect, with vocabulary shared with neighbouring languages, is harder and needs more correction.
Many names have a Ukrainian and a Russian form, Олександр and Александр, Микола and Николай, Олена and Елена, and people use either. The transcript writes the form that was spoken; it does not convert names, and it does not produce Latin spellings such as Oleksandr or Kyiv. If you need those for an English document, they have to be added by hand. Workplace recordings, especially from IT teams, are full of English terms said the Ukrainian way: дедлайн, мітинг, апдейт. These are usually written in Cyrillic, while brand and product names may appear in Latin letters.
Ukrainian uploads are often interviews, recorded by journalists, researchers and people collecting oral history, along with online lectures, team meetings from Zoom or Google Meet, long YouTube conversations and voice messages saved from a messenger. Guests get 10 transcriptions of up to 3 hours each, and the free allowance is 5 hours in total across those 10 recordings. A free account draws on the same allowance; it keeps the full transcripts and adds an editor and a history. Translation, AI summaries, AI chat and export to PDF or Word are in the paid plans.
Recordings are processed on our own servers with our own speech models, and the audio is not passed to a third-party speech API. For interviews with people who spoke on condition of confidentiality, that keeps the answer to where the recording went short.