184 · Audio Tools
Talk instead of typing: notes, drafts, messages and ideas.
Choosing the language yourself is faster and more reliable than detecting it, especially for short recordings.
Runs on your device with an open-source AI model (40–240 MB, depending on accuracy, downloaded from this site the first time). Your files are never uploaded.
Speak clearly at a normal pace in a quiet room, and say punctuation naturally with pauses rather than words. Choosing the language instead of “Detect automatically” helps with short recordings.
Text in a few seconds
English text
It uses Whisper, an open-source speech model, in its small “tiny” size so it can run in a browser. Clear speech in common languages works well; strong accents, noise, music and overlapping speakers cause mistakes, so check the result before using it.
No. The AI model is downloaded from this site once and runs on your device. The audio is never uploaded.
