So we're looking at a benchmark of about two and a half seconds to take the audio and generate a transcription and kick it back and then another eight seconds to generate text around it. So 11 seconds total in this setup and that's probably for a 22nd piece of text.