[abound]

Current LLMs are multimodal -- you could record yourself singing + beatboxing, upload it to an LLM and ask it to translate it to Strudel code.

My best guess is that it won't work well, but we'll never know until someone tries it.

[paulgb]

My experience is that the best LLMs can produce syntactically valid strudel but don't understand the actual semantics of the "mini notation" language, unfortunately.