I think there is general agreement that IPA is crucial for learning pronunciation of a language like English, because there are no strict rules of pronunciation, and plenty of exceptions.
IPA is close to roman alphabet (with plenty of special characters). But for learners of English whose L1 is not written using roman alphabet (Japanese, Thai, etc), this adds additional load: they need to learn not a single new alphabet, but two.
I was surprised when I learned that in Thailand, when learning English pronunciation they do not rely much on IPA, but more on sound recordings. Which is problematic, because it was proven that it is hard for learners to distinguish between sound differences in L2 which do not exist in L1. Like well-known problems for L1=Japanese to distinguish between L and R.
When I was learning English decades ago, I had very little access to sounds of native English (beyond music), and relied heavily on IPA. I admit that my L1 uses roman alphabet, and IPA was a bit simplified (/i/ merged with /ɪ/, /e/ with /ɛ/, /u/ with /ʊ/, /o/ with /ɔ/, /ɑ/ with /ʌ/) I think - I do not remember the details now.
Is there any evidence that "sounds without IPA" approach may succeed? Or it requires of many more hours of exposure to the sounds of native speech?