What is the best speech to text for Japanese?

For most people, the simplest choice is the built-in tool on the device they already use. On a Mac, Apple's Dictation lists Japanese (Japan) for on-device processing ¹. On Windows, voice typing supports Japanese ², but it needs an internet connection ³. If you need a free model that runs on your own computer, Whisper is the best-known option. We did not find a published test that compares Whisper with the built-in engines on the same Japanese audio, so test your own speech before you rely on any of them.

Start with the built-in tool, and switch only when it falls short

For short messages, the built-in tool is usually enough. It needs no setup beyond choosing Japanese as the dictation language, and you can check each sentence as it appears. The table compares the three options that cover most everyday use.

The built-in options differ most in one respect: whether they work without a connection. The Mac option is designed for on-device processing ⁴, while Windows voice typing will not work offline at all ³. If you write Japanese on a Windows PC away from a connection, plan on an offline app.

Whisper is free and runs offline, but it takes setup and memory

Whisper's code and model weights are released under the MIT license ⁵, which allows you to use the software without a per-minute fee. The price is effort and hardware. The largest model needs about 10 GB of video memory for full-size use ⁶. Smaller models need less memory, but the smallest model in the table below, untuned Whisper-Tiny, scored 32.7 percent ⁷ on a different test from the large model's result. Setup also takes more work than turning on a built-in setting, so Whisper suits people who want an offline tool and are willing to install it.

Google Cloud Speech-to-Text is a different kind of option. Developers call it from their own code, and it offers 60 free minutes a month for its standard model ⁸. Above that, Google charges by the minute, at a standard rate of 0.016 dollars per minute ⁹, so it is not a free dictation app for everyday use.

The published Japanese accuracy numbers come from different tests

Accuracy for Japanese is reported as a character error rate, which counts the character-level edits needed to match the reference text. The two results below are the ones we could open that test Whisper directly.

The first row is a small model before training on Japanese. The same study found that fine-tuning on Japanese data lowered the score to 14.7 percent ⁷. The second row is a harder test, with spontaneous speech full of fillers and hesitations, not the short, deliberate sentences most people dictate. The two rows use different test sets, so they do not rank Whisper against the built-in tools. They show that model size and training change the score a lot.

Which option to choose for your situation

1. On a Mac, start with Apple Dictation for short messages. Check the Dictation settings on your own Mac, because the language lists change between macOS releases.
2. On Windows, use voice typing when you have an internet connection. It works in any app that has a text box.
3. For offline or long recordings, consider Whisper. Use it when you can install and run it. For a one-off file, the free file transcription in the browser avoids the setup.
4. Test whichever you choose. Read about 200 characters of your own Japanese aloud. Count the characters that differ from what you meant, and divide by the length of the passage. The guide to Japanese, Chinese and Korean speech to text describes the three-step test in full.

Homophones are a common Japanese error. The same sound can be written with different kanji, and a dictation engine chooses one from context, so read each sentence before you send it. For the term, see homophone.

Yap, which we make, runs dictation on the device for Mac and Windows. We have not stated its Japanese support here, so check the list on our download page before you choose it for Japanese.

Related answers

• How do you voice type in Chinese?
• How do you voice type in Korean?
• How fast do people type in Japanese?
• Free online dictation in your browser

This answer is part of the guide Japanese, Chinese and Korean speech to text in the Languages section.