When I'm fairly or completely new to a language, I always turn audio on when learning. It helps me to get the correct pronunciation into my head.
Later, I don't really need it anymore because I start "subvocalising" the words automatically, and if phase 1 went well, I should subvocalise without too many pronunciation errors.
This worked quite well for me in Spanish. The stories have very natural audio, I think there are real people speaking instead of auto-generated TTS. The sentences are quite ok, too, and even the turtle mode or clicking tiles is ok. Well, obviously these last two modes will never teach you the intonation of sentence (its "melody"), but at least I can learn the words properly.
I checked it in English and German, and it seems similar there. Turtle mode and tiles don't sound great, let alone natural, but I can stand listening to them.
Now, with French, stories are fine, sentences normally too, but tile clicking or turtle mode is so horrible that I would rather have it muted. The thing is, you just can't say an isolated "d'", "l'" or "j'", so "d'or" becomes "de or", "l'homme" "le homme", and "j'ai" "je ai". Similarly letters may be pronounced or silent in a single word, but it can be the opposite in a sentence context, and it's really hard to get a feeling for it when I hear it the wrong way half of the time.
I wasn't able to find any good way to deal with this up to now. Typing instead of clicking tiles isn't available in lower levels within the app, and turning the volume up and down all the time is distracting. Would there be any better solutions?