Im using it for non english speach. It (pro small model) works much better than ios/google/bing voice to text. Even the english auto translation is usable( though i would recommend to transcribe in the original language and then use either chatGPT or other tool to translate - quite a big improvement ) * the auto english translation average results is not the fault if the app but the model and its size Given only the 1/3 if the training set non english, one has to use the small model. Id love to see bigger models as well, ideally the large one. But i understand that there are SOME challenges.