It doesn't save a recording of someone speaking; it's 15 seconds of a transcription.