Estonian accuracy, honestly
Estonian has more speech data than some small languages, but still much less than English, German or Finnish. The model gives a usable draft on clear recordings, and the draft still needs a review of names, terms and long compound words.
Recording quality matters most. A lecture, a dictated memo or a podcast with good microphones comes out well enough to edit quickly. Phone calls, farm and family recordings with background noise, and several people talking at once give weaker text that needs line-by-line correction.
Estonian has three degrees of length in vowels and consonants, and the difference between them is often heard as two. That shows up as single and double letters swapped, such as kalu and kallu, or linna and lina. Read with that in mind.
Estonian spelling points to check
The transcript is written in the Estonian alphabet. These are the most common fixes.
- Õ, ä, ö and ü are usually right, but õ and ö are sometimes confused in quick speech.
- Single and double vowels and consonants (a/aa, t/tt) depend on length the model may not hear clearly.
- Long compound words may be split into two words or joined where they should be separate.
- Š and ž appear in loanwords and names and can lose their háček.
- Numbers may come out as digits or words. Estonian uses a comma as the decimal mark and a space for thousands, and the model may not follow that.
Estonian, Finnish, Russian and auto-detect
Estonian and Finnish are closely related and sound similar to outsiders. On auto-detect a short clip can be labeled Finnish, and the transcript then follows Finnish spelling. Clips that start with an English or Russian greeting can also be misjudged.
This page preselects Estonian. If you get Finnish or Russian text back, check the setting and run the file again with Estonian selected. In Tallinn and Ida-Virumaa many conversations switch between Estonian and Russian, so choose the language spoken most in each file.
Võro and Seto in southeast Estonia differ a lot from standard Estonian. The model writes them closer to the standard language, so expect heavy editing.
English mixing and typical uploads
Estonia's tech sector mixes in many English terms. They tend to come through in English spelling, sometimes with an Estonian ending attached. If a meeting is mostly English, choose English for that file.
Estonian has no grammatical gender and no future tense, and word order is flexible, so the model rarely gets the grammar badly wrong. Most corrections are about sounds: length, vowel quality and where one word ends and the next begins. A careful listen at normal speed usually catches them, and slower passages can be replayed by clicking the line.
Swedish-speaking Estonians and Finnish visitors also turn up in recordings from the islands and the north coast. If a whole file is in Finnish or Swedish, pick that language instead of Estonian.
Common uploads are university lectures, startup and team meetings, research interviews, journalism, local podcasts and YouTube videos that need Estonian subtitles. Estonians in Finland, Sweden, Canada and the UK upload family recordings and voice notes, and oral history projects record older people describing the war years and life in the Soviet period.
Uploading, exports and next steps
Upload MP3, M4A, WAV, WhatsApp voice notes, or MP4 and MOV video without converting. You can also record in the browser or import a YouTube or TikTok link. The transcript arrives as timestamped paragraphs, and clicking a line plays that moment.
Exports to TXT, DOCX, PDF, SRT, VTT and CSV keep Estonian letters, and you can turn timestamps on or off. To read the content in English, use the Google Translate link or the AI prompt export with ChatGPT, Claude or Gemini. Speaker labels are on paid plans or with a Minute Pack. Files are never used to train AI models.