Skip to content
/a/ FaceCue Performance Studio

Voices

Building a voice from a few reference clips.

Building a voice from a few reference clips, and the catalogue the rest of the window picks from. Two tabs, one job: a voice you make here is a voice you can synthesize and convert with.

A voice in FaceCue is a file. Building one produces a .fcv1 bundle, that bundle sits in a folder, and anything that needs a voice reads it from there. There is no account, no service, and nothing that expires.

Create Voice

Give it one or more recordings of somebody speaking, and it builds a voice that sounds like them.

Reference Clips

A reference is an AudioClip from your project or a path to a file on disk, and you can give it more than one.

What makes a good reference is what you would expect: clean speech, one speaker, no music underneath, and a decent stretch of ordinary talking rather than a single word. What it is capturing is the character of a voice, so a reference that is all shouting produces a voice that only knows how to shout.

More than one clip is worth doing when no single recording covers the range you want, and there are two separate decisions about how they are used.

How the clips are joined. Hard butts them together, Crossfade blends the join, and Combined reads them as one continuous take. Crossfade is the safe one when clips were recorded at different times, because a hard join can leave a click that the voice then learns from.

How they form the voice. Several clips can be pooled into a single sense of the speaker's tone, which is what you want when they are the same person in the same conditions and you are simply after more material.

You Need the Rights to the Voice You Are Cloning

Building a voice from a recording of a real person is a decision about that person, not just about your project. Use recordings you own or have permission to use, and treat a cloned voice the way you would treat any other likeness.

Native Language

The languages the speaker in the reference speaks natively. They are what let Language Coach tell the voice's own languages from foreign ones, so a line in one of its own languages is left alone and one in another language can be coached. A voice can also coach its native languages for your other voices.

With Manual, you pick them. Main Language is the language the speaker speaks natively, or the one they speak most if they have more than one. Also Native In is for a speaker who grew up with more than one: tick any other language they speak just as natively. Leave out a language they speak with an accent, since that is where coaching helps. A main language left at Not Set saves the voice without one, and it is never coached.

With Auto-Detect, the main language is found from the reference when the voice is created, as long as the Multilingual ASR Transcript Generator is downloaded. The status line says what was found. If it guessed wrong, or the speaker has other native languages, change it in Registered Voices.

What Comes Out

A .fcv1 bundle, named after the reference clip unless you name it yourself, written to the folder you choose. The native language is saved inside it.

That folder is the one the Registered Voices tab lists, and it is shared with the baker's own voice picker, so a voice you build here shows up everywhere a voice can be chosen without another step.

Registered Voices

The catalogue. It lists the .fcv1 bundles in the voices folder, plus Male Sample, the voice that comes with the speech pack, and it is where you choose which one the window uses.

Selected is the important column. It picks the voice that Synthesize and Convert will use, so this tab is where you go to switch voices rather than hunting for a picker inside each tool.

Male Sample is for trying FaceCue out, not for the characters of a finished game. The voices you ship with are the ones you create.

Changing a Voice's Native Language

Each voice shows its main language, and beside it the others it is also native in. Click either one to change it: the main language is a single pick, and the list beside it takes as many as the speaker has. Male Sample is an English speaker and cannot be changed.

A voice shown as Not Set is never coached. Voices made before this setting existed start out that way, and setting it here is all they need.

The Folder

One field, pointing at where the bundles live. It is shared with the baker, so pointing it somewhere else moves both.

Because a voice is a file in a folder, managing them is file management. Removing a voice means removing its bundle, moving voices between projects means copying bundles, and putting a voice under version control works the way any other asset does.

Keep Voices With the Project

A voice bundle inside the project folder travels with it, gets committed with it, and is there for anyone who clones it. A voice kept somewhere else on your machine is a voice your teammates do not have, and the failure shows up as a synthesis that cannot find its voice.

Where a Voice Goes Next