Näin muutat nauhoitteen tekstiksi
Napsauta latauskenttää ja valitse tiedosto: MP3, M4A, WAV, WhatsApp-ääniviesti tai MP4- tai MOV-video. Ääni erotetaan videosta automaattisesti. Voit myös nauhoittaa suoraan selaimessa mikrofonilla tai liittää YouTube- tai TikTok-linkin. Kieleksi on jo asetettu suomi, joten automaattisen tunnistuksen ei tarvitse arvata.
Valmis litteraatti on jaettu aikaleimattuihin kappaleisiin. Soitin korostaa juuri puhuttavan rivin, ja riviä napsauttamalla kuulet sen uudelleen ja voit korjata tekstin suoraan. Lopuksi vie litteraatti muodossa TXT, DOCX, PDF, SRT, VTT tai CSV. Voit kokeilla ilman rekisteröitymistä, ja ilmaisella tilillä litteraatti tallentuu.
Finnish writing conventions the model can miss
Finnish spelling follows sound closely, so most remaining errors are about how words are put together on the page. These are the points to check.
- Compound words (yhdyssanat): työterveyshuolto, asiakaspalvelu and projektipäällikkö are one word. A split compound is the most common written error.
- Colon before endings: abbreviations, numbers and foreign names take a colon before a case ending, as in SRT:ksi, 5:n, EU:ssa and 2020-luvulla with a hyphen.
- Lowercase words: months, days, nationalities and languages are lowercase (marraskuu, maanantai, suomalainen, ruotsi).
- Commas: a comma goes before että, joka, koska and kun when they start a clause, and fast speech often drops it.
- Foreign names inflected in speech, such as Facebookissa or Teamsissä, may be spelled several ways in one transcript.
Numbers, dates and money
Edited Finnish text uses a decimal comma (3,5), a space between thousands (12 000) and the euro sign after the amount with a space (15 €). Dates are written 4.11.2026, times klo 14.30, and ordinals with a dot (3. kerros).
Spoken numbers are long and inflected, so kolmellasadallakahdellakymmenellä is easy to mishear and may come out as words. Convert the figures to digits where your style needs them, and check each one against the audio when amounts or dates matter.
Years and decades come with their own conventions: vuonna 2026, 1990-luvulla and 2020-luvun alussa, with a hyphen between the digits and the ending. Ordinals in speech, such as kolmas or kahdeskymmenes, are written with a dot after the digit (3., 20.). Phone numbers and personal identity codes are better written as digits and checked character by character.
Spoken Helsinki slang and nicknames for places, such as Hesa, Stadi or Tampesteri, may appear as said or be changed to the official name. For a verbatim research transcript keep them, and for a report use the official form so readers outside the region can follow.
Checklist before you publish
In Finnish research the level of detail is often agreed before transcription: sanatarkka (verbatim, with fillers and repetitions) or yleiskielistetty (converted to standard language). Decide which you need, then check:
- Compounds are joined and colon endings are correct.
- Months, days and nationalities are lowercase.
- Spoken forms like mä oon are kept or standardized consistently, matching the chosen level.
- Names of people, places and organizations are spelled correctly.
- Euro amounts, dates and klo times use one format.
- Swedish or English passages are checked by ear, since the file was transcribed as Finnish.
Typical Finnish recordings
Common uploads include gradu and väitöskirja interviews, focus groups, university luennot, kunnanvaltuusto meetings that municipalities stream, workplace palaverit that need a muistio, local podcasts, and family recordings used for sukututkimus. Remote interviews recorded in Teams or Zoom can be uploaded as the MP4 file the meeting app saves, and the audio is taken from the video. Lectures are best recorded close to the lecturer, since a recorder at the back of a hall picks up more room noise than speech.
Export DOCX for a muistio or thesis appendix, CSV for a timed spreadsheet when coding interviews, and SRT or VTT for subtitles. Timestamps and speaker names can be switched on or off per export, and speaker labels come with paid plans or a Minute Pack. Recordings are never used to train AI models.