This bundle targets GPU servers with CUDA. The backend prefers GPU and falls back to CPU only if CUDA is unavailable.

python -m venv .venv && source .venv/bin/activate  # Windows: .venv\Scripts\activate
pip install --upgrade pip
# Install PyTorch CUDA 12.1 wheels
pip install --index-url https://download.pytorch.org/whl/cu121 torch torchvision torchaudio
# App deps
pip install -r requirements.base.txt
python app.py

Requires NVIDIA Container Toolkit on the host.

docker build -t voice-asr:cuda -f Dockerfile.cuda .
docker run --rm --gpus all -p 5000:5000 voice-asr:cuda

Notes:

• Models are downloaded from ModelScope on first use.
• FFmpeg is installed in the image and the app also uses imageio-ffmpeg for robust audio handling.
• If the container/host lacks CUDA, the app will automatically fall back to CPU but performance may be lower.