Written Japanese mixes kanji for most content words, hiragana for grammar and katakana for loanwords and emphasis, and it does not put spaces between words. A transcript therefore involves decisions that do not exist in English: where one word ends, which words to write in kanji, and which of several characters with the same reading is meant. The text you get here is written the way Japanese is normally printed, with 、 and 。 for commas and full stops. It is not romaji and it has no furigana.
Because there are no spaces, a timestamp on each segment is the practical way to move between the text and the recording. If you are studying the language, copying a segment into a dictionary or reading tool works well, since the text is real Japanese script and not an approximation in Latin letters. Text search in a browser also behaves differently without word boundaries: searching for a two-character word will also find it inside longer compounds.
Japanese has a small set of sounds and a very large number of words that share them. Kōshō alone can be 交渉, 考証, 高尚, 公証 and more; kikan may be 期間, 機関, 器官 or 帰還. A speech model chooses between them from the surrounding sentence, and in ordinary conversation, news-style speech and clear lectures it mostly chooses well. Errors of this kind look alarming on the page, because the wrong kanji is a different word, but they are quick to spot for anyone who reads Japanese and knows the subject.
Personal names are the case that cannot be solved by listening. Satō is almost always 佐藤, but a given name such as Yūki can be written in dozens of ways and nothing in the audio says which. Place names outside the big cities, company names and the titles of works have the same problem. Expect names to come out in a plausible but possibly wrong spelling, or in kana, and correct them from a participant list or a business card. Numbers and counters are usually written in Arabic digits, which is worth checking for dates, prices and amounts.
A business meeting in Japanese is long sentences in keigo, softened endings and a steady stream of aizuchi, the はい, ええ and そうですね that listeners add while someone else is talking. Short interjections that overlap with the main speaker are often dropped, which generally makes the transcript easier to read, and fillers such as えーと and あのー are kept only in part. If you need a strictly verbatim record with every hesitation, plan to add them by hand while listening.
English enters Japanese as katakana: ミーティング, スケジュール, コンセンサス. These are written in katakana as spoken. Product names and acronyms may appear in Latin letters. Standard Tokyo Japanese is recognised best. Kansai speech from Osaka and Kyoto is common enough in broadcasting that it transcribes well. Stronger regional dialects, such as those of Tōhoku, Kagoshima or Okinawa, and very casual speech between friends, are where the text departs most from what was said.
Japanese files uploaded here are usually one to two hours long: seminars and university lectures, online meetings saved from Zoom or Teams, interviews, and talks, podcasts and long conversations from YouTube, which language learners transcribe to read along. Guests get 10 transcriptions of up to 3 hours each. The free allowance is 5 hours in total across those 10 recordings; a free account keeps the full transcripts. AI chat with the transcript, automatic summaries and PDF or Word export are in the paid plans, and subtitles can be translated into other languages.
Recordings are processed on our own servers with our own speech models and are not sent to a third-party speech API. For internal meetings and client interviews, that removes one step from the approval most companies require before audio can be uploaded anywhere.