Japanese Speech to Text: Get the Names Right
Japan's 20th and 21st most common surnames are both pronounced Saitō: 斉藤 and 斎藤. 齋藤 and 齊藤 are Saitō too. The audio won't tell a speech model which one is in your meeting. You can.
WhisperAI transcribes Japanese audio and video. Before you start, type the names and terms you'll hear into Important words or names, spelled the way you want them. Then choose Japanese and upload. Files can be up to 5 GB, and your first 5 minutes each month are free, no card needed.
Why does speech to text get Japanese names wrong?
Because most Japanese words can be written more than one way, and that often includes names. Speech researchers call Japanese a language "lacking orthography": one word has no single fixed spelling. A model can hear さいとう right and still write the wrong kanji.
Those are often the words that matter most in a transcript: a client's name, a company, a product.
How to transcribe Japanese audio to text
Start with the list. In the upload panel, open Advanced Transcription Instructions and type your names and terms into Important words or names, separated by a regular comma (,), the kind on an English keyboard:
齋藤, 髙橋, 株式会社ミドリ, サブスクリプション, 決算説明会
The Japanese commas 、 and , don't split the list, so five names typed with 、 count as one. You can add up to 100 terms, each up to 80 characters.
Keep Transcription Style on Standard. With any other style, or with Advanced instructions filled in, only the first 50 terms are sent.
Then choose Japanese, not Auto-detect, which can come back unsure and ask you to transcribe again. If speakers switch into English, open Multi-Language (with English) in the language menu and pick Japanese. Upload MP3, M4A, WAV, MP4, MOV and other common formats.
The list goes to the speech engine with your file, and the engine's documentation says that's what it's for: "highly specific names, brands, acronyms, or niche industry jargon".
Japanese speech to text accuracy: what to check
Japanese has no spaces between words, so accuracy is counted by character. OpenAI's Whisper paper scores Japanese that way. That score misleads in both directions. A harmless alternative spelling counts as an error, while a wrong kanji in a name costs only a character or two.
So when you try any tool on your own audio, go straight to the names and terms. That's where a transcript is right or wrong for you.
Japanese audio transcription for meetings
Recording a meeting in Japanese? WhisperAI transcribes it live from your browser tab and microphone, with speakers labeled as they talk and no bot joining the call. Japanese is one of 18 live languages, and live transcription is on every plan.
On Business Pro, the AI summary and meeting recap are written in Japanese when the transcript is.
Japanese transcription online: exports and price
Download the transcript as TXT, DOCX, PDF, SRT, VTT or JSON. Word and PDF files use a Japanese font. Translate it into English or any of 87 languages from the supported list.
Premium is $14.99 a month for 120 minutes, which a weekly one-hour meeting outgrows. Business Pro is $24.99 a month with no minute limit.
Japanese transcription questions
Should I type names in kanji or kana?
The way you want them in the transcript. Type 齋藤 for kanji or さいとう for kana.
Can WhisperAI add furigana or romaji?
No. The transcript is in kanji and kana, the way Japanese is written.
Transcription guides and best practices
Transcribe in 100+ languages
WhisperAI transcribes Japanese and more than 90 other languages in the same editor, with the same exports.
Other widely-used pages: English, Spanish, French, German, Portuguese, Korean, Chinese, Arabic, Russian, Hindi, Italian, and 80+ more languages including Swedish, Norwegian, Danish, Finnish, Greek, Malay, Filipino and beyond.
See every supported language, with transcription, realtime and translation coverage.
Start transcribing Japanese today
Sign up free, drop in your first Japanese file, and have a usable transcript in minutes.