Baked

On a baked line the Prosody driver knows the words as well as the voice. It knows where each syllable falls, which ones the language stresses, and where each phrase ends, so it can place the emphasis on the syllable that carries it and make a question look like a question.

Emphasis & Signal

A baked line offers both drive modes.

Emphasis, the default, is the one to use. It finds the emphasised syllables in the line and gives each one a brow lift and a head nod together, sized by how much the syllable stands out. On top of that it poses the head and brows on questions and winds the head down toward each phrase end. Everything below this section is about Emphasis.

Signal follows the voice moment to moment instead, the way a live line does, reading the voice from the bake. It does not look at the words, so it has no questions or phrase endings. It is there for a character you want to answer the voice more loosely, and it is set up the same way as on a live line, see Live.

Where the Emphasis Comes From

Two things at once, and neither on its own would be enough.

How the line was delivered. Loudness, duration and pitch, each read as contrast against the rest of the line, never as an absolute. A syllable is emphasised because it stands out, not because it crossed a threshold.

How the line is built. Which syllables carry stress in the first place, from the structure of the words themselves.

Take the delivery alone and a shout is all emphasis. Take the structure alone and every sentence is emphasised identically regardless of how it was said. Together they find the syllable that was both stressed and delivered as though it mattered.

Every Comparison Stays Inside the Line

This is worth stating on its own, because it is what makes the reading portable.

Every distinction Emphasis makes is made inside the clip, against the clip. A word only stands out against the rest of its own line, never against another recording or a profile of the speaker. So a whisper and a shout each find their full range of emphasis, each heard on its own terms.

So the reading a bake produces is the whole reading. It is finished when the bake is, it travels with the clip, and it plays the same on every character and every machine.

One Clip Is One Yardstick

The other side of that rule is that the clip decides what counts as ordinary. Two clips of the same words can read slightly differently if one is longer, or delivered more evenly, because each is measured against itself. A very short line, with little to measure against, is given a calmer read on purpose rather than an exaggerated one.

The case to know about is a clip whose delivery changes inside it: a whisper that becomes a shout, a calm sentence and a furious one in a single recording. Measured against the whole clip, the quiet half reads as quiet, and its emphasis comes out smaller than it would have alone. If a line looks underplayed and you know its audio is uneven, this is the first thing to suspect.

What to do about it has no single right answer. You can split the recording so each part is measured against itself. You can have each syllable compared with the syllables just around it, under Emphasis ▸ Contrast. Or you can leave it, on the grounds that the quiet half really was quieter, which is often the more truthful performance.

Questions & Phrase Endings

A question does not end like a statement, and the face knows it. On a baked line, the face asks along with the voice:

• The head lifts and tilts a little at the end of a question, like someone waiting on an answer.
• The brows raise on a yes or no question, joining the lift into one questioning look.
• The brows gather on a question built on a question word, such as who, what or where. It needs a Brow Furrow blendshape in the character's prosody rig map, and a rig without one simply goes without.
• The head winds down toward the end of each phrase and holds still through the pause, the way a speaker settles between thoughts.

Each is sized on its own, and a character can raise one brow instead of two for a more quizzical question. Keep the furrow light, since it easily reads as puzzled rather than asking.

All four start on, and each has its own switch, under Question and Head Dip Settle. None of them runs on a live line, because a live line has no words to find a question in.

Tuning It

Three things, and they are the ones worth touching. Each is in its own section under Settings.

How much the brows raise, under Brow. The most visible, and the easiest to overdo. A brow that moves on every accent reads as surprise, not emphasis. Between emphases the brows can rest a touch below neutral, so each lift stands out against a settled face.

How far the head nods, under Head Dip. Small numbers go a long way here, because the head is large and the eye is sensitive to it.

How the nod moves. Engine picks between Authored, where each emphasis plays the same shaped rise and return, and Spring, where the head and brows are pulled toward each emphasis and settle back on their own, so no two returns are quite alike.