Prosody, Emotion & Control

'I didn't say he stole the money' is seven different sentences depending on which word carries the stress. Prosody — pitch contour, energy, duration, rhythm, pauses — carries meaning the transcript is blind to. Emotion lives in the same channel: same words, cold vs warm delivery, opposite social meaning.

This is the actual content of TTS control: the words are fixed by the input; everything that makes speech human is extra input the model must get from somewhere.