The ratio of processing time to audio duration; lower values indicate faster-than-real-time inference.