My 2nd grader didn’t do well in his weekly English spelling quiz recently. Out of 12 words, he got about five of them right. I was the one to blame, of course, for some very legitimate reason :p
As usual, I started by searching for some App or website on which the kid can practice spelling words. I didn’t get something flexible enough that we can readily start to use. Some website supports customized word list to read out each word for students to spell, but they are all missing something subtle but critical.
Imagine how the teacher is giving kids spelling quizzes in class. There is usually an example sentence after each word to spell. On one hand an example sentence would help disambiguation, on the other hand, the student gets reinforced the context to use the word.
The ideal spelling quiz App should provide an example sentence for each word to spell. A straightforward way to augment existing website with this feature is to ask user to provide example sentences in addition to the word list, which, I guess, would make the tool much less interesting to use.
To my surprise, at this point, it boils down to a very technical problem. How can we automatically get an example sentence for each word in the quiz list? The answer was obvious nowadays, ChatGPT to the rescue :p
So I put together something over the weekend. The project is at https://github.com/pppoe/spelling-quiz/.
GPT3.5 can easily parse a list of words into the desired format and writing example sentences given words are one of its core tasks by design. The Text-To-Speech API works well for English.
The kid gets a webpage with a list of play buttons.
The ones we have are hosted lively at http://pppoe.github.io/spelling-quiz/.
I was planning to get the Chinese character writing quizzes into this project as well but the part of codes are commented out for now. The problem is OpenAI’s TTS model.
For Mandarin, the synthetic voice sounds nature as human but for some reason, all voice models have very obvious English speaker accent in speaking Chinese. I assume it is harder to collect this kind of training data and cannot help wondering if the accent unintentionally emerges from the large-scale training…