TipToi Translation (t3)

Usage

• Once you've followed the setup instructions below you can run the application
• Run the following from repository's root
  • python -m t3 'Rekorde im Tierreich.gme' workdir
  • This will translate Rekorde im Tierreich.gme and store the translated GME and intermediate files in the workdir
  • Run python -m t3 -h to see all available options
• Alternatively run the application from a Docker container (see instructions below)

Development

Manual setup

• Clone this repo with submodules: git clone --recurse-submodules git@github.com:jtomori/t3.git
• sudo apt install sox ffmpeg
• pip install numpy typing_extensions
• pip install -r requirements.txt
• Store SeamlessExpressive models in the SeamlessExpressive folder in repository's root
• Compile libtiptoi.c: gcc tip-toi-reveng/libtiptoi.c -o libtiptoi

Docker

• GPU inference requires NVIDIA Container Toolkit
• Build image with docker build -t t3 .
• Run container with docker run --runtime=nvidia --gpus all --volume ./SeamlessExpressive:/app/SeamlessExpressive --volume ./gme:/app/gme --volume ./workdir:/app/workdir --rm --name t3 t3 gme/name_of_file.gme workdir
  • Make sure that gme, SeamlessExpressive, workdir directories are present in your current directory
  • workdir will contain translated GME file along with intermediate files, CSV report
  • Omit --runtime=nvidia --gpus all for performing a CPU inference

Tests & code checks

• python tests.py
• ./checks.sh

Releases

v1.1 - 2025-03-12

• Finished setup for running GPU (or CPU) inference from a Docker container

v1.0 - 2025-03-11

• Initial release