Run a multilingual dubbing workflow locally from YouTube links
Paste a YouTube link and the tool generates subtitles and creates dubbed files that preserve the original voice. Voice-Pro handles video download, audio separation, recognition, translation, and synthesis in a single web interface. You can run the entire process on your own computer without worrying about cloud service pricing or data transfer.
Development pause and full open-source release
Voice-Pro has paused updates to focus on WeConnect development. The team has released all code as open source, allowing anyone to freely distribute and modify it. The installation method has changed from Miniconda to uv, which speeds up installation, and it supports Python 3.12 and the latest Torch version. You do not need to install CUDA Toolkit or Visual Studio Build Tools separately, as all required dependencies are provided pre-built.
Dubbing studio and voice technology integration
The core of the tool is the 'Dubbing Studio' tab. Here, it downloads YouTube videos, separates background music and voice using Demucs, and extracts text with Whisper-family models. The extracted text is translated into over 100 languages and converted back to speech using models like Edge-TTS, F5-TTS, or CosyVoice. F5-TTS and CosyVoice offer zero-shot cloning, which mimics the original speaker's voice from short samples. The Fun-CosyVoice3 model, which supports Korean and eight other languages, has also been added to expand options.
Local execution environment and requirements
It runs most stably on Windows 10/11 64-bit with an NVIDIA GPU. Mac and Linux are supported but not officially verified. GPU memory should be at least 4GB, with 8GB or more recommended, while system RAM requires at least 4GB and storage space at least 20GB. An internet connection is required, and approximately 10GB of AI models are downloaded on first run. It can run without administrator privileges, and if ffmpeg is missing, it automatically downloads a portable version.
Points to check before use
Development is currently paused, so bug fixes and new features are not being added. Consequently, updates to the latest AI models or libraries are unlikely. It uses free services like Deep-Translator and Edge-TTS by default, which may be blocked or rate-limited on corporate networks. You can switch by setting an Azure API key, but this incurs additional costs. If GPU memory is insufficient, CUDA Out-Of-Memory errors may occur, requiring adjustments such as lowering the noise removal level or changing the calculation method to int type.
