Speech Production Tools¶
Making dialogue inside the editor: a transcript from a recording, a recording from a line of text, a voice built from a few reference clips, and an actor's read moved onto that voice. All of it local, all of it in one window.
Everything here is optional. FaceCue bakes recordings you already have, and nothing on this page is a step you have to take. It exists for the parts of a project where the audio is not ready yet: placeholder dialogue while writing, a voice for a character nobody has cast, a transcript for a line you were given without one.
Open it from FaceCue ▸ Cue Baker Tools ▸ Speech Production Tools.
The Five Tools¶
| Tab | What it makes |
|---|---|
| Recognize Speech | A transcript, from a recording |
| Synthesize Speech | A recording, from text |
| Convert Speech | A recording in one voice, from a recording in another |
| Create Voice | A voice, from reference clips |
| Registered Voices | The catalogue the other tabs pick from |
They are separate jobs that chain. A voice you create is a voice you can synthesize with, and a line you synthesize is a line you can bake, so a character can go from nothing to speaking without leaving the editor.
Every Tab Works the Same Way¶
The window has one shape, and learning it once covers all five.
Input is what you are starting from: a clip, a folder, some text, a CSV. Voice appears wherever a voice is involved, and picks which one. Pipeline is the model and the hardware it runs on. Output is where the result is written and what happens if something is already there. Results is what came out, with the same Show In and Add To routes the baker's results carry.
If that reads familiar, it is the Cue Baker's layout. The two windows are meant to be siblings.
Models Install On Demand¶
Nothing here ships with the package, and nothing here phones home once it is installed. Each tool downloads what it needs the first time you use it, then runs on your machine.
Which model, and how big, depends on the tool. The window says what it is fetching and how far along it is, and a tool whose model is missing says so instead of failing partway.
They run on the same hardware the baker uses. The execution provider you chose for baking is the one these tools run on, so a machine set up for CUDA bakes and synthesizes on CUDA. See Backends.
Single and Batch¶
Every tool takes one item or many, and the choice is a toggle, not a different tool. One clip or a folder of them. One line of text or a CSV of them.
Batch is the point of the window, not a convenience bolted on top. A hundred placeholder lines is the case these tools exist for, and the output naming is built so a batch drops straight into a bake afterwards.
The Settings Pane¶
The window has a settings pane beside the body, and most of what is in it is not this window's.
Those settings belong to the project. Number handling is the clearest case: what a number becomes in words is the same question for a line you synthesize and a line you bake, so both windows read and write one file. Editing it here changes what the baker does, and the two cannot drift apart.
The rest follows the same rule the baker's does. It is there for when you have a specific problem, each control documents itself in the editor, and the useful default is to leave it alone. See The Settings Pane.
Languages¶
Speech generation covers a different set of languages from baking, and a smaller one. Both are listed together, with a checker, in Language Support.
The short version: a language you can bake is not automatically a language you can synthesize, and the reverse is also true. If you are planning to generate dialogue in a particular language, check it there before building a pipeline around it.