This is a spaced repetition app I learned to help me learn 汉字. To get it working, you will need a .env file with the following keys:
OPENAI_API_KEYSPEECH_KEY-- Azure TTS keySPEECH_REGION-- Azure TTS region
This app uses the optimal character learning order computed by Loach and Wang in doi:10.1371/journal.pone.0163623. This order is based on usage frequency subject to the constraint of learning character components before they appear in other characters. A problem with this learning order is that it can create (by chance) clusters of characters with similar pronunciations which can make them hard to distinguish when learning. A future improvement would involve recomputing the character order subject to a constraint that nearby characters not be too similar in pronunciation.
The app is currently designed around the idea of teaching 汉字. However there are many characters which have multiple meanings with their own pronunciation. For example, 相 can mean "one another" when pronounced xiang1 but "appearance" when pronounced xiang4. An improvement would be to teach character-pronunciation pairs instead of just characters.
Currently the app assumes there is only one user and is designed to be run locally. Adding session support and switching to a more robust database could allow Hanzi Friend to be hosted, which would also have the benefit of letting you practice on any device.
Because of the character order approach the app will sometimes attempt to teach radicals that aren't used by themselves, antiquated words, etc. A better approach would filter out antiquated meanings and only quiz the meaning of radicals.
It would be good if you could add words that you learn in the wild into the rotation ahead of time.
- Always marks pronunciation answers wrong on multi-character words or characters with multiple pronunciations.
- AI sometimes generates phrases which do not contain the target word.
- AI sometimes generates malformatted output.
- Sometimes includes "no character found" in character decompositions. These could be removed or alternatively a high-lift alternative would be to make SVG implementations of some of the shapes.
- Code cleanup