What this is: a test of what today's technology can hear. A speech-sound model runs right on this iPad and lists the sounds it heard, then lines them up with the sounds the word should have.
Recordings never leave this device. Only the typed word goes out, for the model voice and its list of sounds.
Honest limits: the model learned mostly from adult voices, so it's less accurate with kids and with speech that's still developing. It often misses subtle differences (a slushy S, a slight lisp) and mixes up close sounds like TH / F / V. Vowels are shown in grey but not judged. It's a practice toy, not a test. His speech therapist's ear is always the final word.
The sound checker needs a one-time download of about 320 MB (use Wi-Fi). After that it works right away, even offline.