Japanese voice typing on Mac: kanji, kana and keigo

Yes, you can voice type in Japanese, and Apple's built-in dictation (音声入力) supports it. Apple's dictation does add 。and 、for you, and you can turn that off in Keyboard settings. But it does not remove the えっと or the false starts (words you begin, stop, and then say again). It does not always choose the right homophone (a word that sounds the same as another word) from context. And it does not turn a casual spoken sentence into です・ます for a client. Lipo is our app, and it does all of that. You hold a key, speak Japanese the way you say it to a colleague, and release the key. Clean Japanese text arrives in Mail, Slack, LINE, Notion or any other text field (a box where you can type). Formal mode gives you keigo (polite business Japanese), and translation gives you English instead. Japanese voice typing is also called Japanese voice dictation, and this guide uses both.

Why is Japanese voice typing hard?

Start with kana-to-kanji conversion. Kana are the phonetic characters (hiragana and katakana), and kanji are the Chinese characters. When you type Japanese, you type kana, and the IME (the input software) offers kanji candidates. For はし it offers 橋 (bridge), 箸 (chopsticks) or 端 (edge), and you choose one. When you speak, nobody chooses, so the app must decide from context. Common homophones like きかん (期間, 機関, 器官, 帰還) and こうせい (構成, 校正, 厚生, 公正) appear in every business email. A wrong choice changes the meaning.

Segmentation means finding where one word ends and the next begins. Japanese has no spaces, so the model must find the word boundaries before it can convert anything. 「きょうはいしゃにいく」 is either 今日は医者に行く (going to the doctor today) or 今日歯医者に行く (going to the dentist today). Only context separates them.

Register (the level of formality) is the biggest problem for work. Spoken Japanese to a friend uses だ・である or plain form. Written Japanese to a client uses です・ます at minimum, and often full keigo, with 尊敬語 for the reader's actions and 謙譲語 for your own. 「明日行く」 becomes 「明日伺います」. Nobody speaks that way into a microphone, so an app that transcribes exactly what you said gives you text you cannot send.

Loanwords (words taken from another language) go in katakana: ミーティング, クライアント, デッドライン. When a model hears "client" said with a Japanese accent, it may write the word in Latin letters or hear a Japanese word instead. Product names are the opposite problem, because Slack, Figma and GitHub stay in Latin letters in most Japanese writing. So the app must know which English words to convert and which to leave.

Punctuation is the last problem. Japanese uses full-width 。and 、, and quotes are 「」, not "". Apple's dictation inserts 。and 、for you in Japanese. The quote marks are not part of that, so you still add those yourself.

How to voice type in Japanese with Lipo

You keep the IME you already use for typing.

1. Install Lipo from uselipo.com. It needs macOS 15.2 or later. There is a free trial.
2. During onboarding, pick Japanese as your dictation language. You can change it later under Dictation, then Language. Lipo can also auto-detect the language (guess which language you are speaking), but choosing Japanese gives better results.
3. Grant the microphone and accessibility permissions (settings where you allow the app to use part of your computer). The accessibility permission lets Lipo paste into other apps.
4. Click into any text field. Mail, Slack, LINE, Notion, Gmail in Safari and a code editor all work. The cursor is the blinking line that shows where text will go.
5. Hold the hotkey and speak. A hotkey is a key you press to start recording, and the default is the right Option key. When you release the key, Lipo transcribes your speech (turns it into text), cleans it up, and pastes the finished text where your cursor is. Cleaning up means Lipo removes filler words (sounds like えっと that carry no meaning) and false starts, adds punctuation, and fixes small mistakes. Lipo does not type letter by letter, and it does not show text while you speak.

Here is a message with the mess real speech has in it. Six different problems, all in one message.

You say:

「えっと、来週の水曜、あ、木曜でしたっけ、木曜のリリースなんですけど、その、テストが、テストがまだ二件残ってて。あと、在庫が三千個って言ってましたけど、実際は二千八百です。それいがいは問題ないです。資料は今日中に共有しますんで、確認お願いします。」

Lipo pastes:

木曜のリリースですが、テストがまだ2件残っています。あと、在庫は3,000個と言っていましたが、実際は2,800個です。それ以外は問題ないです。資料は今日中に共有しますので、確認お願いします。

Six things happened there:

• The correction was kept. You said Wednesday, corrected yourself, and said Thursday. The message says Thursday.
• The stutter is gone. 「テストが、テストが」 is written once.
• Filler is gone. 「えっと」 and 「その」 do not appear.
• Numbers are written as digits. 「二件」 became 2件, 「三千個」 became 3,000個 and 「二千八百」 became 2,800個. Lipo never changes the value of a number, only the way it is written.
• The right kanji was chosen from the meaning. 「いがい」 is 以外 and 意外 in speech. The sentence means "apart from that", so Lipo wrote 以外. This is the single most common Japanese dictation error, and it is a meaning problem, not a sound problem.
• A half sentence was finished. 「残ってて」 became 「残っています」, matching the です・ます you were already using. Casual mode keeps your register, it does not raise it. The full-width 。and 、are where a Japanese writer would put them, and リリース and テスト stay in katakana.

For a long email, switch to Toggle mode in settings. Press the hotkey once to start and again to stop.

Does Lipo write keigo?

Formal mode does, and Casual mode keeps whatever register you spoke in. In Casual mode, Lipo fixes homophones and punctuation, removes the filler words, and leaves your register alone. In Formal mode, Lipo rewrites the text in a professional style, which for Japanese means business keigo. You say: 「明日行くね」. Lipo writes: 明日伺います。 You say: 「これ見といて」. Lipo writes: こちらをご確認いただけますでしょうか。 You say: 「ごめん、遅れる」. Lipo writes: 申し訳ございません、少々遅れます。

Pin Formal to Mail and Casual to LINE or Slack, and the same hotkey produces the right register in each app.

There are some limits. Formal mode writes standard business keigo, so read a 詫び状 (formal letter of apology) for a major client before you send it. Homophone choice works on ordinary vocabulary and gets weaker on specialist terms. In a printing company 「こうせい」 means 校正, and in a government office it means 厚生. The dictionary exists for those cases. Lipo keeps product names (Slack, Figma) in Latin letters, which matches most Japanese business writing. If your company writes スラック, add that spelling to the dictionary. If a sentence comes out wrong, tell us.

Speak Japanese, paste English

Many Japanese speakers write English to overseas teams, a parent company or a supplier. You think of the sentence in Japanese, translate it in your head, and then type English. Lipo lets you skip the middle step.

In Lipo's settings, open the Translation tab and set "Lipo pastes" to English. Then choose when translation happens:

• Always translate. Every dictation comes out in English.
• Translate with this hotkey. Your normal hotkey pastes Japanese, and a second hotkey pastes English.

With translation on, Lipo transcribes your Japanese, cleans it up, translates the cleaned text, and then pastes it. If you translate raw speech with the えっと still in it, you get English that is fluent but empty.

When you write in a second language, you often say less than you mean, because you are thinking about grammar. In Japanese you know exactly how much to soften a request. When you speak Japanese and Lipo translates it, the meaning and the tone stay.

Here is a whole email, spoken the way people actually speak one.

You say (Japanese, Formal mode, output language set to English):

「サラさん、いつもお世話になっております。あの、納品の件なんですけど、その、来週の月曜はちょっと厳しくて、水曜まで、水曜までいただけないでしょうか。代わりに、先に仕様書だけお送りするか、テスト環境を先に開けるか、どちらでも対応できます。ご都合いかがでしょうか。」

Lipo pastes:

Hi Sarah,

Thank you as always. On the delivery: next Monday will be difficult for us. Could we have until Wednesday? In the meantime we can either send you the specification first or open the test environment early, whichever suits you. Please let me know which you would prefer.

Read that against what you said. You spoke a greeting, so you get a greeting. Lipo will not invent a greeting, a subject line or a sign-off you did not say, so the message stops where your speech stopped instead of adding "Best regards".

「いつもお世話になっております」 has no English equivalent. It became a natural opening line rather than a word-for-word translation, and 「ちょっと厳しくて」 became "will be difficult" rather than "a little severe". The two options stayed inside the sentence: Lipo turns three or more parallel items into a list, but leaves two of them in the prose, which is what a person would write. 「水曜まで、水曜まで」 was written once, and 「あの」 and 「その」 are gone.

The translation docs explain every setting. The reverse direction works too. Set "Lipo pastes" to Japanese, speak English, switch to Formal mode, and Lipo pastes Japanese with keigo. Ask a native colleague to read a delicate message, but a routine 「資料をお送りします」 email is fine. The dictate and translate guide shows more examples. Translation has a monthly word limit on the smaller plan; see pricing for plans and limits.

For names that keep coming out wrong, use the dictionary. Add a word spelled the way you want, and Lipo uses that spelling in every language and after translation. You can set your surname with the kanji you use (斎藤, 斉藤 or 齋藤), or 校正 instead of 構成.

What about Wispr Flow, Willow Voice, Aqua Voice and superwhisper?

Here is how the four compare with Lipo on the things that matter for Japanese.

A long language list means the app can recognize those languages. It does not mean it is equally good in all of them.

The clearest evidence for that comes from a competitor. Wispr Flow's own research page sorts its languages into two groups: seven that it says are "trained and tuned to match English-level performance", which are French, German, Hindi, Italian, Portuguese, Spanish and Thai, and "dozens of other major languages" where it promises accurate dictation without making that comparison. Japanese is in the second group. That is their wording, not our opinion of them.

"Not sure" in the table means the company gives a total but does not say in public whether that language is one of them. On totals, Wispr Flow and Willow Voice both say 100+, Aqua Voice publishes a full list of 49, superwhisper depends on the model you choose, and Lipo lists 50+.

Nobody publishes per-language accuracy figures, ourselves included. So speak one real sentence in Japanese, with a person's name and one English word in it, and read what comes back. Ten minutes of that tells you more than any comparison table, including this one. Lipo is built so Japanese gets the same accuracy work as English, rather than a long list where only a few languages work well.

None of the four recognize a language change inside one sentence, so they treat 「Figmaのファイル、Slackで送っといて」 as one language. You cannot predict which product names come out right.

Screenshots and Visual Brief, in Japanese

While you are recording, the small floating window (Lipo calls it the pill) has a screenshot button. Click it, capture the screen, and keep speaking. When you release the hotkey, Lipo pastes the text and attaches the screenshot to the same message. So 「このエラー見て」 arrives with the error in it. The screenshots docs have the details.

Visual Brief goes further. AI tools such as Claude, ChatGPT and Cursor read only text. Visual Brief lets you record your screen while you describe a problem in Japanese. Start it from the Lipo menu bar icon and record your screen or one window for up to 5 minutes. You can say 「このボタンを押すとページがリロードされるんですけど、データが表示されないんです」. Lipo transcribes what you said, analyzes the recording, writes a structured prompt in your output language, and pastes it into the AI tool. It needs the Screen Recording permission, and recordings are deleted from your computer after 7 days. Details are in the Visual Brief docs.

What about Apple dictation and the free options?

Apple dictation on macOS 15 supports Japanese. It runs on-device, which means on your computer, without sending audio to a server. Its kanji conversion works on everyday vocabulary. It does not add 。and 、unless you say 「まる」 and 「てん」. It does not remove えっと or false starts, and it does not change register or translate. It listens for one language at a time, so it cannot handle English product names inside a Japanese sentence.

Tips for better Japanese voice dictation

Speak full sentences. The cleanup step reads the whole sentence for context, and that is how Lipo decides between 橋 and 箸.

Do not say punctuation. Pause at the end of a sentence, and Lipo adds the 。.

If you correct yourself (「月曜、じゃなくて火曜」), Lipo keeps the correction and removes the mistake.

Choose Japanese as your language instead of auto-detect, especially if your sentences contain many English terms.

Use the built-in microphone or a wired headset. Wireless earbuds compress the audio, and pitch accent is the first thing that gets lost.

Frequently asked questions

Can I voice type in Japanese on a Mac?

Yes. Apple's built-in dictation supports Japanese. Apps like Lipo add automatic punctuation, filler removal, keigo through Formal mode, and translation.

Does Apple dictation support Japanese?

Yes. It runs on-device, and its kanji conversion works on everyday words. It does not add 。and 、unless you say them, and it does not remove filler words. It does not change register, and it listens for one language at a time.

How does Lipo choose the right kanji?

Lipo chooses from the meaning of the whole sentence, the same way a reader does. For specialist terms (校正 or 構成 in your industry), add the word to the dictionary.

Can Lipo write keigo?

Yes. Formal mode rewrites casual speech into business Japanese with です・ます and the matching 尊敬語 and 謙譲語. Casual mode keeps whatever register you spoke in.

Does it use full-width punctuation?

Yes. Lipo writes 。and 、and uses 「」 for quotes, the way printed Japanese does. Product names like Slack and Figma stay in Latin letters unless you add a katakana spelling to the dictionary.

How are English loanwords written?

Lipo writes them in katakana (ミーティング, クライアント, デッドライン), the way they appear in Japanese print. Product and company names stay in Latin letters.

Can I speak Japanese and get English text, or the reverse?

Both. Set "Lipo pastes" to English or Japanese in the Translation tab. Lipo transcribes, cleans up, translates, and then pastes. English to Japanese in Formal mode produces keigo. Translation has a monthly word limit on the smaller plan; see pricing.

Does it work in LINE, Slack and Gmail?

Yes. Lipo pastes into any field with a cursor, including the LINE desktop app, Slack and Gmail in the browser.

How much does it cost?

Plans and prices are on the pricing page. There is a free trial.