The Speaker Settings panel controls how each speaker in your project sounds: their details, which voice is used, and audio settings that help the AI voice match the original speaker.
How do I open Speaker Settings? ⚙️
Open your project and click the speakers icon (second icon) in the left-hand toolbar of the editor. Each speaker has a tab at the top of the panel; use the search icon to find one, or + to add one.
1. Speaker details 🎙️
For each speaker, add:
Name: for example "John" or "Sarah". CAMB.AI suggests names based on how speakers look in the video (for example "Host" or "Male guest in black blazer"); you can change them.
Gender.
At the top you can add new speakers or switch between existing ones, rename a speaker, or delete one. See How to add, change or remove a speaker.
2. Which voice is used? 🗣️
You can choose the voice reference for each speaker, per target language:
Voice from original video/media: CAMB.AI learns the speaker's voice from your video and generates the dub in that voice.
Voice reference: use a voice from your Voice Library, either a ready-made voice or your own custom voice.
How do I choose a voice for a speaker?
1. In the speaker's tab, open Select languages* and pick the language(s) you want to set a voice for. Choose Select All to use the same voice in every language.
2. Click Select Voice (or the Voice Reference row for a language). The voice picker opens.
3. Pick a voice. The tabs show Explore (ready-made voices), My Voices (voices you created), Workspace (voices shared in your team) and Saved (voices you bookmarked). Search by name or filter by Language, Gender, Age and Category, and press ▶ to preview a voice. Voice from original media is always at the top of the list.
4. Regenerate the speaker's dialogues to hear the new voice. This uses credits.
If you don't choose a voice, CAMB.AI uses Voice from original media. Choose Clone & Save to Library to save a speaker's voice to your Voice Library so you can name it and reuse it in other projects. More: What is the Voice Library? and How to create a custom voice.
3. Audio settings (MARS8-Pro projects)
ℹ️ These settings only work with MARS8-Pro. You’ll still see them on projects that use other models, such as MARS8-Instruct, but changing them has no effect there. To use them, switch the language’s model to MARS8-Pro under Settings → Project Model, then regenerate.
If your video has complex audio (background noise, crowds, music, a poor microphone) or the voice doesn't sound right, fine-tune each speaker here. Change a setting, click Confirm, then regenerate the speaker's dialogues to hear the result. Each regeneration uses credits.
Quick reference
Setting | When to use | Recommended value |
Clean Reference | Public settings, background cheering or singing, poor microphone | On for noisy or poor-quality audio; Off if the audio is already clean |
Acoustic Quality Boost | The output voice sounds hazy or muffled | On if the output is hazy; Off if it's already clear |
Stability | Make speech more expressive, or reduce hallucinations (odd sounds or words) | Towards 0 for more expressive; towards 1 to remove hallucinations |
Speaker Similarity | How closely the voice copies the original speaker | 0.7 (default) for most speakers; 0–0.3 if you don't have permission to clone the voice |
Accent Boost | Balance between keeping the original speaker's voice and sounding like a native speaker of the target language | 0 (default) = Better voice match; move towards 1 = Stronger native accent. Raise it gradually, e.g. 0.3, then 0.7 |
Speaking Rate | Speech feels rushed, or pauses are too long | Start at 1.0; above 1.0 if rushed, below 1.0 if pauses are long (0.8 works well for audiobooks and simple narration) |
More detail on each setting
Clean Reference ("Remove noise from reference") isolates the main speaker from the environment when the video was filmed somewhere noisy (singing, cheering, low-quality microphones).
Acoustic Quality Boost is your main tool for clarity: turn it on to sharpen a hazy or muffled voice.
Stability controls how robotic or human the voice sounds. Lower values allow more natural inflection and emotion; higher values make the voice steadier.
Speaker Similarity: 0–0.4 makes the speaker sound like a different person; 0.7 is the sweet spot for most speakers; 1.0 is for special cases where the output doesn't sound enough like the original. Use lower values if you don't have explicit permission to clone a voice.
Accent Boost is a slider from Better voice match (0) to Stronger native accent (1). At 0 the dub stays as close as possible to the original speaker's voice, including their own accent (for example, a French speaker dubbed into English keeps a French accent). Moving it up makes the voice sound more like a native speaker of the target language (for English, an unaccented American or British accent), with a little less of the original voice character. Even 0.3 can remove much of the source accent, so raise it step by step.
⚠️ Good to know
Noisy or distorted input audio can reduce how natural the voice sounds, whatever the settings.
A speaker's settings apply to all target languages, but each speaker can have different settings. How much each needs depends on how much clear speech that speaker has in the video.
Intermediate values give you a balance between settings. Regenerate after each change to hear the effect (this uses credits).
Still not happy with the timing rather than the voice? See How to improve your dubbed output.