Blazingly fast whisper transcriptions with Inference Endpoints