Expressive TTS

Expressive TTS is a technology that lets a computer read text aloud while adding emotion and intonation to it. Even with the same script, a news anchor reading it flatly sounds completely different from an actor performing it with feeling. Most existing synthetic voices leaned toward the anchor style, and Expressive TTS is an attempt to move closer to the actor style.

This difference matters because of how the technology gets used. A voice that only reads text in a flat tone can't bring audiobook narration or a game character's lines to life. For the content to work, it needs to sound sad in sad scenes and urgent in urgent ones. Expressive TTS is used exactly for this purpose: automatically generating voice content that needs an emotional arc.

That said, there are failure cases where the intonation becomes exaggerated, so a review step where a human actually listens to and polishes the final output is still necessary.

The article introduces a feature added in audio.cpp 0.5 as "Expressive TTS (DramaBox)." This is different from technology that transfers a voice's identity to another speaker (cross-lingual voice transfer) — one makes the same speaker sound more convincing, while the other changes the speaker itself. The article treats the two as separate entries.