From source

Requirements

GPU	NVIDIA. With 8 GB, everything at maximum quality; with 6 or 4 GB, cloned voice with lighter models; with 2 GB or no GPU, subtitles only. Translay chooses automatically based on your machine (see Performance).
Python	3.11 or newer for the server; the voice engine uses its own environment with Python 3.11.
uv	To create the voice engine environment.
ffmpeg	On the PATH: downloads, thumbnails and export.
Node.js 20+	On the PATH: yt-dlp uses it for YouTube’s JavaScript challenge.
Disk	~15 GB for models and environments, plus whatever your videos take up.

Note

Would you rather not install any of this? Use Docker.

On Windows the simplest option is the desktop app, which installs all of this by itself. Running from source is useful for development, for Linux, or to use Translay from another machine’s browser.

Steps

Every published version is on Releases, with its notes; the code is always on GitHub.

1. Clone the repository and install the server dependencies:

   git clone https://github.com/alenj0x1/translay.git
   cd translay
   pip install -r requirements.txt

   On Linux, if your system already has CUDA 12 and cuDNN 9, you can remove the nvidia-* lines from requirements.txt.

2. Create the voice engine environment (OmniVoice with torch for CUDA 12.8):

   Windows

   uv venv -p 3.11 .venvs/omnivoice
   uv pip install -p .venvs/omnivoice/Scripts/python.exe --torch-backend=cu128 omnivoice deepfilternet soundfile

   Linux

   uv venv -p 3.11 .venvs/omnivoice
   uv pip install -p .venvs/omnivoice/bin/python --torch-backend=cu128 omnivoice deepfilternet soundfile

3. Start it:

   python run.py

   On Windows, iniciar.bat also works. http://127.0.0.1:8765 opens.

The first time, the models are downloaded from Hugging Face (Whisper large-v3-turbo, Opus-MT, OmniVoice and WeSpeaker, about 6 GB) into the user cache. After that it starts in seconds and stays in idle mode until there is something to process.

All startup options: Command line.

What gets stored

Everything that is yours (library, transcripts, voices, preferences, hashed passwords) lives in data/, which is not tracked by git. With --data-dir (or TRANSLAY_DATA) you can use another folder, for example the same one the desktop app uses. Details in Data on disk.