Speech to text online: Estonian, English, Russian and more

How to use it

1. Choose the language: English, Estonian and Russian are at the top of the list, 97 more below – type a few letters of a name to find it. Or leave “Detect automatically”.
2. Press the microphone. The first time, the browser asks for access to it – allow it. Start speaking when “Listening…” appears.
3. Say a phrase and stop the recording: the text appears in a few seconds. The next recordings are added to the end of the text in the order you made them.
4. Correct the text by hand if needed, copy it with one button or listen to it.
5. To transcribe a ready recording – a voice message, a lecture, an interview or a podcast – choose an audio file below the microphone or drag it onto the tool. The text appears when the file is recognized; you can download it as .txt or as .srt subtitles.

What the recognition can do

Speech is recognized by Whisper large-v3, one of the most accurate open speech recognition models, and Estonian speech by a version of Whisper that Tallinn University of Technology (TalTech) trained further on Estonian. Both run on this site's own server, not in a third-party cloud, and the recording or the file is used only for the recognition – it is not stored.

• Punctuation and capital letters are added automatically, every sentence starts on a new line.
• 100 languages; the more widespread the language, the fewer the mistakes.
• Copes well with accents – useful when English or Estonian is not your first language.

Estonian speech to text

Dictate an email or a message in Estonian and see how the words are spelled: long vowels, double consonants, ä, õ, ö and ü. It is a quick pronunciation check too: if the recognized text is what you meant to say, people will understand you. Numbers in Estonian come out in words: “kakskümmend kolm”, not “23”. It works the other way round as well: paste an Estonian or English text, choose its language and press “Play audio” to hear it read aloud.

Tips for accurate recognition

• Choose the language yourself when you know it: on short phrases automatic detection sometimes gets it wrong.
• Speak in complete sentences and stop the recording after a few of them: a short recording is recognized faster, and a mistake in it is easier to fix.
• Avoid background noise – music, TV, people talking nearby. A headset works better than a laptop microphone.

Frequently asked questions

Is it free?

Yes, with no sign-up and nothing to install: it works in the browser on a computer or a phone.

Why does the first recording take longer?

After half an hour without requests the server unloads the speech model. It starts loading again as soon as you press the microphone, so usually you don't notice; a very short first recording may still wait a few seconds – the page tells you so. A recording that was not sent can be sent again or dropped: it is not lost.

Which files can I upload?

Audio files: MP3, M4A, OGG, OPUS, WEBM, WAV, FLAC, AMR and more – up to 150 MB and up to 3 hours. Voice messages from WhatsApp and Telegram work too. An hour of speech is recognized in about two minutes. Video files are not accepted: save the sound track of the video first. The file is used only for the recognition and deleted right after it.

Want to talk, not just dictate? The same speech recognition works in the voice AI chat for Estonian conversation practice: you speak, the tutor answers aloud and points out your mistakes.