Direct the Performance: Emotion and Speed Controls

Until this week, Polyvox let you choose who speaks and what language they speak in, but not how the line lands. A birthday message and a horror movie trailer came out with the same energy. Today we are adding two controls that change that: speaking speed, and emotion.

Speed is a slider from 0.6x to 1.5x, and it works in all 44 languages. Emotion is a set of presets, neutral, calm, content, angry, sad, and scared, that guide the delivery the way a director gives an actor a note before the take. Both are set per clip, so you can generate the same script three different ways and keep whichever one sounds right.

If you never touch either control, nothing about your clips changes. The model already reads emotional subtext out of the words you wrote and paces well-punctuated writing naturally, which is why the default stays on Auto. These are refinements for the clips where the automatic read is not quite what you had in mind, not settings you have to configure before every generation.

The most useful thing to know about emotion is that it amplifies what your script already implies, it does not override it. Tagging a line like 'I am so excited to see you' as sad fights the words and usually produces something muddy. Write the line the way it should be said, then use the preset to push the delivery further in that direction. The six presets we exposed are the ones with the most training data behind them, which makes them the most reliable of the fifty-odd tags the model accepts.

Emotion is English-only for now. The underlying tags are still in beta and only supported for English, and we would rather hide a control than ship one that silently does nothing in 43 of our 44 languages. So the emotion picker appears when you select English and steps aside otherwise, while speed stays available everywhere. When emotional guidance expands to more languages, it will show up on its own.

For speed, small adjustments beat dramatic ones. Around 0.9x gives explainers and tutorials room to breathe and makes dense scripts easier to follow. Somewhere between 1.1x and 1.2x adds urgency, which works well for social clips where the first two seconds decide whether anyone keeps listening. You can push all the way to either end of the range, but the further you get from natural pace, the more likely you are to hear artifacts, so treat 0.6x and 1.5x as effects rather than defaults.

Voice style is part of Pro and Unlimited, alongside downloads, the full language list, and longer scripts. On the free plan you will still see the controls in the studio with a note about what they do, so you can judge whether they are worth $10/month before you decide. Everything the free plan did yesterday, it still does today.

Try it on something you have already made. Take a clip that came out flat, drop the speed to 0.9x, tag it calm, and generate it again. The gap between a clip that sounds machine-read and one that sounds performed is usually smaller than you would expect.