i was wondering about the 'Detect Speech Emotion', does this detect it based on the transcript or based on the actual voice?