Generates subtitles from audio using whisperX (faster-whisper-large-v3)
This model is cold. You'll get a fast response if the model is warm and already running, and a slower response if the model is cold and starting up.