If you uploaded the audio via the web site or app and got terrible results I'd bet Argon wasn't the model doing the transcribing. The only way to know for sure is to try again with the API.