Tongue Twister

Tongue Twister Challenge

Try classic Japanese tongue twisters like “Namamugi namagome namatamago”. Read the phrase into your microphone and the Whisper speech recognition AI writes down what it heard. It compares the reading with the phrase and scores you on accuracy and speed, with a rank, a comment and the spots where your tongue got tangled. Everything runs in your browser.

Phrase

Level

How many times

Your turn

Press the button and read the phrase aloud. Recording stops automatically when you finish (up to 30 seconds).

MP3, M4A, WAV, WebM etc. Up to 30 seconds is used.

AI settings
Speech recognition model

Your voice is processed only in your browser and is not sent to our server or to Hugging Face.

  • This is a game. The score is a rough guide based on what the AI heard, and it is not an assessment of pronunciation or speech. The AI sometimes mishears or “corrects” what you said.
  • Accuracy compares the hiragana reading of the AI’s result with the reading of the phrase. Speed is the number of morae (beats) per second from the start to the end of your voice.
  • The phrases are traditional Japanese tongue twisters and lines from the 18th-century kabuki speech “Uirō-uri”, which are in the public domain.
  • Processing runs on your device and can take a while on older smartphones. Results are best in a quiet place.
  • AI model: OpenAI Whisper base / small (Apache-2.0, ONNX versions Xenova/whisper-*). Runtime: Transformers.js (Apache-2.0), ONNX Runtime Web (MIT). Readings: kuromoji.js (Apache-2.0, IPADIC dictionary).

About this tool

A free game for Japanese tongue twisters (hayakuchi kotoba). Read a phrase into your microphone and Whisper speech recognition writes down what it heard. The result is converted to hiragana and compared with the phrase, and you are scored on accuracy and speed. Choose from four levels, from classics like “Namamugi namagome namatamago” to master-level lines from the kabuki speech Uirō-uri. See your rank, a comment and where you stumbled, and save or share your result card. Audio is processed in your browser and never uploaded.

How to use

  1. Choose a level (Easy, Medium, Hard or Master) and pick a phrase, or press Another phrase. You can also choose 3 times in a row.
  2. Press Record and challenge and read the phrase aloud. Recording stops automatically when you finish. Without a microphone, choose an audio file instead.
  3. The first time, wait while the AI model and the reading dictionary load.
  4. Your score, rank, accuracy, time and speed appear, with the parts of the phrase that differed shown in red.
  5. Use Save result image or Share to send your result card.

Useful for

  • Competing with friends or family on Japanese tongue twisters
  • Warming up your diction before a presentation, interview or stream
  • Practising Japanese pronunciation and rhythm in a fun way
  • Playing word games with children

FAQ

How is the score calculated?

The AI’s transcript is converted into hiragana and compared with the reading of the phrase to give an accuracy rate. Speed, measured in morae per second from the start to the end of your voice, is added to make a score out of 100.

Is it an accurate test of pronunciation?

No, it is a game. The AI sometimes corrects a slip into the right words or mishears a correct reading, so the score is only a rough guide.

Is my voice uploaded?

No. Transcription and reading conversion run in your browser. Only the AI model (about 80–325 MB) and the reading dictionary (about 18 MB) are downloaded the first time and kept in your browser.

Can I play without a microphone?

Yes. Record yourself with a voice memo app and choose the audio file (MP3, M4A, WAV, etc.) with the audio file option.

Contact