I built this after months of being able to read French but failing to understand spoken French. Turned out my brain was merging /u/ and /y/ into one sound. ABX minimal pair training (Lively, Logan & Pisoni 1991) is the established way to fix this but no consumer tool implements it cleanly with audio-only training (no orthographic bias). So I built one. French and English are live, more languages coming. Free demo on the homepage.