Gave it a try. After a few minutes I felt more like I was recognising the samples than I was recognising the notes. Not sure what you can do about that short of physically modeling an instrument.
There are some libraries that make it easy to simulate instruments. E.g. tone.js https://tonejs.github.io/
It should be possible to generate unique-ish variants at runtime.
limut implements WebAudio and WebGL, and FoxDot-like patterns and samples: https://github.com/sdclibbery/limut
https://glicol.org/ runs in a browser and as a VST plugin
"Using the Web Audio API to Make a Modem" (2017) https://news.ycombinator.com/item?id=15471723
gh topics/webaudio: https://github.com/topics/webaudio
awesome-webaudio: https://github.com/notthetup/awesome-webaudio
> Currently includes the following built in exercises:
> [...]
> 7. Interval recognition - the very popular exercise almost all app has. Although I do not recommend using it as I find it inaffective in confusing, since the intervals are out-of-context.
Will probably add a "randomize instrument used per round" setting or something to really dial it in. I added a randomize velocity option but didn't test it much