Currently Pump uses comparison to a reference audio for prompt detection. It would be nice to be able to write a test case where we compared a prompt to expected text not a wav file.