> If left the OverflowError propagated I would catch it at the caller place (_read_float() is used only once) and reraise as aifc.Error.

Ah, if there's a dedicated exception type already then yes, agreed that raising aifc.Error (with a suitable message) makes much more sense than raising an OverflowError or a ValueError.

> if there are files with an exponent of 0x7fff in wild

I hope there aren't, and I'd consider any such file to be broken/corrupted. That is, unless there's a common practice of using NaNs or infinities for the frame rate. I suppose it's conceivable that people us this as a placeholder meaning "I don't know the sample rate". No idea whether that's true in practice: I'm not a frequent user of this format.