05 Voice Roles & Cloned Voices

This chapter is about whose voice speaks the target language in the dub. There are four sources; you only need to pick one:

• Voice Clone: takes this speaker's timbre from the original audio on the spot and uses it to speak the target language. The closest to "the original video" of the four, and the default.
• Built-in Characters: the voices that ship with the app, ready to use and grouped by language.
• Custom Role: record a clip yourself (or have someone else record it) and upload it, then reuse it whenever you want. This is the route if one person should keep the same voice across several videos.
• ByteDance Online Voices: a set of voices served from the cloud. This section only appears when your account can fetch the online voice list; when the list cannot be fetched the whole section is missing, which does not mean anything is broken.
• This section's filter row is a category multi-select: tick the categories you want, or keep "All Categories" (the default) to show every category.

1. What the "Voice Roles" page shows

Click Voice Roles in the left bar. From top to bottom the page is three sections whose titles you click to expand or collapse: Custom Role (with a count to the right of the title), System Role, and one extra section when online voices are available. The Custom Role section and the online voices section each have their own filters above the list:

• Language: a multi-select dropdown — tick several languages, or pick "All Languages".
• Status: All / Enabled / Disabled.

The System Role section has no filters — it only lists its voices grouped by language, and there are many of them, so just keep scrolling.

Every custom role takes one row: name, language, an Enabled/Disabled switch, Preview Audio, Delete.

2. Creating a custom role

Click "+ New Voice Role" at the top right of the Custom Role section and fill in four things:

1. Name: it cannot be empty, cannot be longer than 128 characters, and cannot contain illegal characters; no two roles may share a name within the same language, and a duplicate reports "A voice with the same name already exists".
2. Language: which language this voice will serve later. The dropdown holds 19 languages and has no Traditional Chinese — for Traditional Chinese use the Simplified Chinese entry.
3. Voice file: click "Browse" to choose a local audio file; mp3 / wav / m4a are supported. There is no recording feature — record it somewhere else first and save it as a file.
4. Status: Enabled / Disabled, with Enabled as the default.

Below the form is a reference text card, headed "Please generate audio based on the reference text below": once you pick a language it gives you a sample passage to record. The point is that the closer what your reference audio says is to this passage, the steadier the cloned voice. When recording, aim for 10–30 seconds, a quiet room, one speaker, and no background music.

After you click "OK" the role is usable immediately (this completes synchronously — there is no "generating / waiting for training" state, and nothing retries after a failure). Nothing can be edited afterwards: to change the audio you delete the role and create it again, and deletion asks you to confirm and is irreversible.

3. Preview Audio

When you choose Built-in Characters, the Browse Voices… button in the form opens a voice library dialog you can pick from (before you pick anything, that field reads "No voice selected"). Custom Role has no such dialog — you choose one of the roles you created straight from the dropdown.

Each row has a play button, and clicking it plays that role's own reference audio. Only one plays at a time — clicking the same button again stops it. When the audio cannot be fetched, you get "Preview failed".

One thing to flag: in the New Project form, the Preview Audio button next to "Custom Role" does nothing when you click it (the feature at that spot is not wired up in the form). To audition a role, do it on the "Voice Roles" page — it amounts to the same thing.

4. How these voices are chosen inside a project

The Target Voice field in the order form is a choice of three: Voice Clone / Built-in Characters / Custom Role. Choose "Custom Role" and a second dropdown appears for picking the exact role. Note a few things:

• Expert Video Translation does not have this field in the order form — it always runs in two fixed stages, and the voice is set in the second stage (the editor's left panel), see 04 Subtitle & Dubbing Editor.
• Once you turn on Multi-speaker mode, Built-in Characters and Custom Role are put away: the app then assigns a voice per detected speaker, and you change each speaker's voice in the editor instead of choosing one voice for everything.
• The dropdown in the editor lists only roles that are "Enabled" and whose language equals this project's target language. Nine times out of ten that is why your own role is missing: you switched it to Disabled, or you picked the wrong language when you created it.

Voice cloning is opened up by plan. When your plan does not support it, the option reads "Voice cloning not supported in your current plan", it is disabled and marked "(N/A)", and Built-in Characters is used automatically.

Professional voices need an account. While you are signed out, the Professional entry in the voice-quality dropdown is greyed out and reads Log in to unlock; the local voices keep working, so you can build a project and listen to it before you register. Sign in to any free account and Professional voices become available.

5. Emotion voices and dialects

If you set Speech Synthesis Quality to Emotion (Apple Silicon only), the dubbing follows the per-sentence emotions from the editor and can also switch dialect. The emotion choices in the editor are limited to the ones the engine knows: happy, excited, sad, depressed, angry, fear, hate, surprised and neutral.

Dialect is only offered for Chinese target languages: Mandarin (default), Cantonese or Sichuanese. The cloned voice is kept — the dialect changes the accent, not the speaker.

Emotion mode needs a reference with a verbatim transcript, so it works with the original voice (Voice Clone) or a built-in character. Custom roles have no transcript and are disabled while Emotion is selected.