Clone Any Voice & Dub Videos with Voice-Pro

// Before you start

What you need

Windows 10 or 11 PC
NVIDIA GPU (GTX 1060 or better recommended)
At least 20 GB free disk space (models are large)
Stable internet connection for first-run model downloads
Git installed (https://git-scm.com/download/win)

Step 1 of 7

Clone the repository to your PC

2 min

This downloads all the project files to a folder on your computer. Open a Command Prompt (press Windows key, type 'cmd', hit Enter) and run the command below. It creates a folder called 'voice-pro' wherever your prompt is currently pointing — your home folder is fine.

Terminal · mac

$ git clone https://github.com/abus-aikorea/voice-pro.git && cd voice-pro

What you should see

A folder named 'voice-pro' appears. Your prompt now shows you are inside it.

This might happen

Git is not recognised as a command

Install Git from https://git-scm.com/download/win, then open a fresh Command Prompt and try again.

Step 2 of 7

Run the installer to set up Python and all dependencies

10–20 min

The project ships a Windows batch file that handles everything: it downloads a portable Python 3.12, installs uv (a fast package manager), and pulls in PyTorch and all other libraries. You do not need to install Python yourself. Just double-click the file or run the command below from inside the voice-pro folder. The first run takes the longest — subsequent starts are fast.

Terminal · mac

$ start.bat

What you should see

A terminal window shows download and install progress. When it finishes it will either open the web UI automatically or print a local URL like http://127.0.0.1:7860

This might happen

The installer stalls or errors partway through

Delete the 'installer_files' folder inside voice-pro, then run start.bat again. Downloaded AI models in the 'model' folder are kept, so only the environment is rebuilt.

Step 3 of 7

Open the web UI in your browser

1 min

Once start.bat finishes, your default browser may open automatically. If it does not, look in the terminal for a line that says 'Running on local URL' and copy that address into your browser manually.

Terminal · mac

$ http://127.0.0.1:7860

What you should see

A Gradio web interface loads with tabs for Transcription, Translation, TTS, Voice Cloning, and more.

This might happen

The page shows 'This site can't be reached'

The server is still starting. Wait 30 seconds and refresh. If it still fails, check the terminal for a different port number and use that instead.

Step 4 of 7

Transcribe or download a YouTube video

5 min

Go to the 'YouTube' or 'Transcription' tab. Paste a YouTube URL or upload a local audio/video file. Choose your Whisper model size (large-v3-turbo is a good balance of speed and accuracy). Click the Run button. Voice-Pro will download the audio, separate speech from background noise using Demucs, and produce a transcript. The first run also downloads the Whisper model, which adds a few minutes.

Terminal · mac

$ Paste your YouTube URL into the URL field on the YouTube tab, select a Whisper model, then click Run.

What you should see

A transcript appears in the text box below the player. Subtitle files (.srt, .vtt) are available to download.

This might happen

YouTube download fails with an error about yt-dlp

The video may be age-restricted or region-blocked. Try a different video, or upload a local file instead.

Step 5 of 7

Clone a voice and generate speech

5–10 min

Go to the 'Voice Cloning' tab (F5-TTS or CosyVoice). Upload a short audio clip of the voice you want to clone — 10 to 30 seconds of clean speech works best. Type the text you want that voice to say. Click Generate. The first time you use a model it downloads automatically (F5-TTS is around 1–2 GB; CosyVoice2-0.5B is around 9 GB — that one can take over an hour on a slow connection).

Terminal · mac

$ 1. Go to the Voice Cloning tab.

$ 2. Upload your reference audio clip.

$ 3. Type your target text.

$ 4. Click Generate.

What you should see

An audio player appears with the cloned voice speaking your text. You can download the .wav file.

This might happen

Output audio sounds robotic or mismatched

Use a cleaner reference clip with no background music. Clips under 10 seconds or with heavy noise produce poor results. Try trimming to a single clear sentence.

Step 6 of 7

Translate and dub a video into another language

10–30 min depending on video length

Go to the 'Dubbing' or 'Translation' tab. Upload your video or use one you already transcribed. Select the source language and the target language. Voice-Pro will translate the transcript using Deep-Translator and re-synthesise the speech in the target language using Edge-TTS or your cloned voice. The result is a dubbed video file you can download.

Terminal · mac

$ 1. Go to the Dubbing tab.

$ 2. Upload your video file.

$ 3. Set source and target languages.

$ 4. Click Run.

What you should see

A dubbed video file is generated and available for download. A progress bar shows each stage (transcription → translation → TTS → merge).

This might happen

Translation fails or returns partial results

The free Google translation endpoint rate-limits heavy use. Voice-Pro retries automatically and keeps original lines for any that fail. For reliable large-scale use, add your own Azure Translator key in the settings tab.

Step 7 of 7

Shut down and restart cleanly

1 min

Close the browser tab, then press Ctrl+C in the terminal window to stop the server. Next time you want to use Voice-Pro, just navigate back into the voice-pro folder and run start.bat again. It will start much faster because everything is already installed.

Terminal · mac

$ Press Ctrl+C in the terminal, then close the window.

What you should see

The terminal exits cleanly. No uninstall is needed between sessions.

// Status

cooked. baked. worked.

A locally running web app where you can transcribe audio, clone voices from short clips, translate content into 100+ languages, and export dubbed video files — all without sending data to a paid API.

// the honest bit

The honest part

Voice-Pro is Windows + NVIDIA GPU only — Mac and Linux are untested and likely broken. The CosyVoice model is 9 GB and can take over an hour to download on a slow connection. Active development is paused while the maintainers work on a separate project, so bugs may not be fixed quickly. Voice cloning quality depends heavily on the quality of your reference audio clip. The free translation endpoint can be rate-limited; for production use you need your own Azure keys.