For me even though in 0.82 the transcripts started working again, they are completely off, where they used to be pretty accurate. The Whisper models aren't working but the Parakeet v2 is, and it is the one that is generating these inaccurate . Also using HyprLLM for the intelligence bit.