The sentence that would settle each column
A blank is more useful when it comes with the sentence that would fill it. For each column here there is one short statement a vendor could publish, and asking for that sentence is a faster conversation than asking whether a feature exists. As of 2026-09-12.
| Column | The sentence that would settle it |
|---|---|
| Audio source | Whether speech leaves the model with the picture, or the shot arrives silent |
| Languages | The names of the languages a spoken performance is available in |
| Lip-sync | What a generated mouth is following, and what happens with two speakers |
| Per-character binding | Whether a chosen voice persists between generations, and how |
| Voice source | What a production may hand over, in seconds and megabytes |
Inclusion rule. Every column this register keeps, with the minimum published statement that would fill it. Order. Alphabetical by column.
1A request for one sentence is harder to deflect
Asking whether a product supports voice consistency invites a yes. Asking whether a vendor will write down that a chosen voice persists between generations, and by what mechanism, invites either a sentence or a silence, and both are informative.
A vendor unwilling to put one line in writing has said something that no amount of testing establishes. That is the most useful outcome of a procurement conversation and it takes one email.
2Units are what make a sentence checkable
Three of these five statements contain a quantity, and a quantity is what turns a claim into something that can be wrong. Seconds, megabytes and language names can all be checked; support and capability cannot.
That is why this register grades a figure above a capability and a list above a total. The grade is not a judgement about the product, it is a measure of how much of the answer survives being written down.
3The mouth question needs two halves to be useful
Naming a driver is most of an answer and not all of it. What happens when two faces are in frame is the part that decides whether a conversation can be shot in one generation, and it is asked by almost nobody.
Asking for both halves in one sentence is deliberate. A vendor that answers the first and skips the second has told a production to plan on alternating singles, which is a useful answer even though nobody said it.
4Where the published detail is logged
What is described above is how the techniques work in general. Which models document which of them, in whose words, is kept in the speech controls table with a date on every field.
- How a field gets filled in — the rule behind every column
- Reading a blank cell — and what to do about each kind
A mechanism note rather than a documented field. Nothing here is attributed to a model, and nothing here is a claim about one. The sourced material is on the speech controls table. Related: Asking for silence, Sizing a scene.