Make Voiceovers and Dubs on Your Own Computer

Install VoiceStudio, make a one-line voiceover test, then turn a video you made into a dubbed version in another language. The voice work happens on your own computer.

What this helps you do

VoiceStudio is a free, open-source desktop app for voiceovers, voice cloning, and video dubbing. You type a line and it speaks it in the voice you pick. Or you give it a video and it transcribes the speech, translates it, and generates a new voice track timed to the original.

This page is for you if you make videos and want a voiceover or a second-language version without paying per minute or uploading your recordings to a cloud service. It also explains why your first try feels slow, so you don't give up on it.

Before you start

• A supported computer. An Apple Silicon Mac on macOS 13.3 or newer, a Windows 10 or 11 PC, or a Linux PC. The install table lists every option.
• About 10 GB of free disk space for the app, its helper software, and the voice models.
• Intel Macs can't run the voice engine locally. On an Intel Mac the app opens, but it needs to connect to another computer to do the work.
• A short test clip for dubbing, 30 to 60 seconds of you talking, plus the text for one voiceover line.
• Permission for any voice you clone. Use your own recording or one from someone who said yes.

Terms to know

• Voice model: the AI file that turns text into speech. VoiceStudio downloads it the first time you use it.
• Voiceover: a spoken track made from text you type.
• Dubbing: replacing the speech in a video with a translated version in another language.
• Voice cloning: making new speech that sounds like a voice from a clean reference recording.
• Acceleration: using your computer's graphics chip to make voices faster. VoiceStudio uses Metal on Apple Silicon Macs and CUDA on NVIDIA graphics cards. Without either, it still works on the main processor, just slower.
• Translation engine: the part that translates your transcript. Engines marked Local run on your computer. Engines marked Online send the text to a web service.

Step-by-step

Install the app

1. Download the installer for your system from the latest release page. Use the .dmg on a Mac (pick the arm64 file), the .exe on Windows, or the .AppImage or .deb on Linux.
2. If you prefer one command, open Terminal on a Mac or Linux and paste:

curl -fsSL https://voicestudio.sh/install | sh

On Windows, open PowerShell and paste:

irm https://voicestudio.sh/install.ps1 | iex

3. Open VoiceStudio. On a Mac, if you see "VoiceStudio is damaged and can't be opened," that is macOS blocking an unsigned app, not a broken download. The Mac install guide has the one-line fix. The shell installer above handles it for you.

Make a test voiceover

4. Open Voice cloning. Pick the bundled voice, or add a clean recording of your own voice.
5. Type one short line and click generate.
6. The app asks before it downloads a voice model. Say yes and let it finish. You should see a progress bar with the total size and file count.
7. Play the result. Then type a second line and generate again. This one comes back much faster because the model is already on your computer.

Dub a video you made

8. Open Dub from the sidebar. Drop in your test clip.
9. Leave the spoken language on Auto and set how many people talk in the clip. Wait for the transcript to appear.
10. Choose your target language and a translation engine marked Local. Argos is the default and runs on your computer. NLLB-200 gives better offline quality but downloads about 2.4 GB.
11. Click translate and read every line. Fix anything that sounds wrong, then click generate.
12. Preview the dub, then export it as an MP4. The dubbed language plays first by default.

Copy this

Keep this checklist next to you for your first dub. Replace the bracketed parts.

First dub checklist
1. Clip: [30 to 60 seconds of video I made]
2. Spoken language: Auto
3. Number of speakers: [HOW MANY PEOPLE TALK]
4. Translation engine: Argos (Local, Fast) or NLLB-200 (Local, Heavy)
5. Target language: [LANGUAGE]
6. Translate, then read and fix every line before Generate
7. Preview with sound, check names and numbers
8. Export MP4 with the dub as the default track
9. Voices used: mine or [NAME], who gave permission

Things to know

• The first run is slow on purpose. It is downloading the voice model. After that, speed depends on your computer. Apple Silicon Macs and PCs with NVIDIA graphics are fastest.
• Keep translation local if privacy matters. The Online engines, such as Google Translate and DeepL, send your transcript text to that service. The Local engines don't.
• It's free. The app is open source under the AGPL-3.0 license and has no paid cloud. Some voice models carry their own licenses, which the Model Manager shows for each model.
• Only clone voices you have permission to use. The project asks for this too.
• A free Hugging Face account is optional. Adding its token in setup can speed up model downloads and unlocks some larger models.

If something goes wrong

• Mac says the app is damaged: run the quarantine fix from the Mac install guide, or install with the Terminal command above.
• Mac says it isn't supported or asks for Rosetta: you installed the Intel build. Download the arm64 .dmg instead.
• A model download stalls: cancel it and try again, or add a free Hugging Face token in Settings. The model download guide covers mirrors and slow networks.
• Generation is very slow: your computer is likely running on the main processor. Try a shorter clip first.
• The translation reads badly: switch to NLLB-200, then edit the lines by hand before you generate.

Final check

You pass when a typed line plays back in the voice you picked, your test clip exports as an MP4 that speaks the new language in time with the video, and every voice you used is yours or one you have permission for.