Fast Local Transcription

faster-whisper backend; pick model_size=small for a balance of speed/accuracy.