Automatic transcription does something particular and dangerous in Arabic: when it is unsure, it produces fluent, grammatical, entirely invented text. Not garbled output you would notice — a clean sentence that reads perfectly and was never said. A monitor skimming a transcript for a quote has no way to tell the difference, which is precisely why so many newsrooms gave up on the tools and went back to a person with headphones.
So this studio routes around the failure rather than hoping it will not happen. Where the risk of invention is high, the audio is transcribed first in the language it can hold most reliably, and the Arabic is produced from that transcript as a translation — a step that cannot invent content, because it only has the transcript to work from. It is slower by a fraction of a second and considerably more honest.