Turn Japanese speech into text on your own device, written in ordinary Japanese script: kanji, hiragana and katakana mixed as in normal writing, not romaji. Drop a recording or a video and get a plain transcript or timed .srt subtitles. Japanese has no spaces between words and many words share a sound, so the model picks each kanji from context. It is usually right on everyday vocabulary and most often wrong on personal and place names, which can be read several ways. Check those before you publish anything. Nothing is uploaded.
How it works
- Drop a Japanese audio file or video; Japanese is preselected as the spoken language.
- Pick Accurate (≈77 MB) for interviews, regional speech or several speakers, Fast (≈41 MB) for a quick draft. Choose Subtitles under Output before running if you need .srt or .vtt.
- Press Transcribe, play back any line whose kanji looks off, fix it in place, and download .txt, .srt or .vtt.
Questions
- Can I get the transcript in romaji or with furigana?
- No. The output is standard Japanese script, the way Japanese is normally written, and there is no setting for romaji or for readings above the kanji. If you need romaji, run the finished .txt through a separate converter; it is plain UTF-8 text and pastes cleanly.
- Why is a name written with the wrong kanji?
- Because the audio carries only the sound. Many Japanese names and words share a reading, and the model writes the characters it has seen most often with that sound. Common words are usually right; names, rare words and specialist terms are where it guesses. Lines can be edited in place, so fix the characters before you download.
- Why is there a line nobody said?
- Over silence, applause or music, Whisper sometimes writes a stock line it learned from subtitled video, such as ご視聴ありがとうございました (thanks for watching). There is no silence filter here to catch it. Delete the line in place, or cut long silent stretches with the audio trimmer before you transcribe.
- Does my file get uploaded?
- No. The models and all processing run inside your browser on your own device. You can verify this yourself: open your browser’s DevTools, watch the Network tab, and process a file: there are no uploads. After the first visit most tools also work fully offline.
- How accurate is it?
- The models are OpenAI’s Whisper (tiny and base tiers). On clear speech they are very usable; heavy accents, crosstalk and noisy rooms will show errors. The Accurate tier is noticeably better on difficult audio.
- Which languages work?
- Fifteen, picked in the Language setting: English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Polish, Turkish, Japanese, Korean, Chinese, Arabic and Hindi. The decoder is conditioned on the one you pick, so it transcribes that language rather than approximating it in English — which also means picking the wrong one returns confident nonsense. The weights understand 99 languages; these are the ones we measured as good enough to offer.
- How long does it take?
- Transcription runs on your processor, so speed depends on the machine: expect around a quarter of the recording’s length, and less on the Fast model than on Accurate.