Transcribe Japanese Audio to Text — Free & Private, No Sign-up

The test of a Japanese transcript is whether it looks like something a person wrote. Spoken Japanese has to come back as the right mix of kanji, hiragana and katakana, with kanji chosen by meaning, loanwords in katakana, and sentences closed with the 。 full stop. A transcript that gets this right can go straight into 議事録 or a subtitle file; one that does not is barely worth correcting.

This page opens with Japanese already chosen and runs the Whisper model inside your browser, so the recording of a meeting, a seminar, an interview or a lecture is processed on your own machine and a confidential discussion is never sent to a server. There is no charge per minute. The model is fetched once and reused after that when the browser still has it. Plain text is the default output, with SRT and VTT exports when you need timestamps.

Where the output needs checking: homophones, where 機会 and 機械 sound identical and the model picks from context; personal and place names, which get the most common kanji rather than the right one; and fast, overlapping conversation, which loses accuracy. Clear, single-speaker audio is what it does best; a Kansai or Tohoku speaker is transcribed, but with more errors than standard speech. Try a short clip first before committing a two-hour file.