| name | deploy-dia |
| description | Deploy the Dia neural text-to-speech model on reComputer Jetson for expressive, multi-speaker audio generation. Installs custom aarch64 PyTorch/torchaudio/triton wheels, clones the Dia repo, patches pyproject.toml, and launches a Gradio WebUI. Requires Jetson with 8GB+ RAM and JetPack 6.1+. |
Deploy Dia TTS on reComputer Jetson
Dia is an expressive neural speech generation model that produces natural multi-speaker audio from text. This skill deploys it on Jetson with a Gradio web interface.
Execution model
Run one phase at a time. After each phase:
- Relay all output to the user.
- If output contains
[STOP] → stop immediately, consult the failure decision tree.
- If output ends with
[OK] → tell the user "Phase N complete" and proceed to the next phase.
Prerequisites
| Requirement | Detail |
|---|
| Jetson device | 8GB+ RAM (e.g. reComputer J4012 with Orin NX 16GB) |
| JetPack | 6.1 or later |
| Python | 3.10 with pip |
| Network | Internet access for downloading wheels and cloning repo |
Phase 1 — Download and install aarch64 PyTorch wheels (~5–10 min)
Download the following wheels from the Seeed shared drive:
torch-2.7.0-cp310-cp310-linux_aarch64.whl
torchaudio-2.7.0-cp310-cp310-linux_aarch64.whl
triton-3.3.0-cp310-cp310-linux_aarch64.whl
Install them:
pip install torch-2.7.0-cp310-cp310-linux_aarch64.whl
pip install torchaudio-2.7.0-cp310-cp310-linux_aarch64.whl
pip install triton-3.3.0-cp310-cp310-linux_aarch64.whl
Verify:
python3 -c "import torch; print(torch.__version__); print('CUDA:', torch.cuda.is_available())"
[OK] when torch imports and CUDA is available. [STOP] if wheel install fails.
Phase 2 — Clone Dia and patch dependencies (~2 min)
git clone https://github.com/nari-labs/dia.git
cd dia
Edit pyproject.toml to comment out the torch, torchaudio, and triton dependency lines (lines 19–22) since we installed custom Jetson wheels:
sed -i 's/^\(\s*"torch.*\)/#\1/' pyproject.toml
sed -i 's/^\(\s*"torchaudio.*\)/#\1/' pyproject.toml
sed -i 's/^\(\s*"triton.*\)/#\1/' pyproject.toml
Install Dia and fix numpy:
pip install -e .
pip install numpy==1.26.4
[OK] when pip install -e . completes without error. [STOP] if dependency resolution fails.
Phase 3 — Launch Dia Gradio WebUI (~2 min to start, model download on first run)
export GRADIO_SERVER_NAME=0.0.0.0
cd dia
python app.py
Open a browser and navigate to http://<jetson-ip>:7860.
[OK] when the Gradio interface loads and you can generate audio from text.
Failure decision tree
| Symptom | Action |
|---|
| Wheel install fails — platform mismatch | Confirm you are on aarch64 with Python 3.10: python3 --version and uname -m. |
import torch fails after install | Check pip target: python3 -m pip show torch. Ensure no conflicting system torch. |
| CUDA not available in torch | Verify JetPack 6.1+ is installed: cat /etc/nv_tegra_release. Check nvcc --version. |
pip install -e . fails on dependency | Ensure torch/torchaudio/triton lines are commented out in pyproject.toml. |
| Gradio WebUI not accessible remotely | Confirm GRADIO_SERVER_NAME=0.0.0.0 is set. Check firewall: sudo ufw allow 7860. |
| OOM when generating audio | Close other GPU processes. 8GB is minimum — larger models may need 16GB. |
| Model download hangs on first run | Check internet connectivity. Retry or manually download model weights. |
Reference files
references/source.body.md — Full Seeed Wiki tutorial with screenshots, demo video, and sample dialogue text (reference only)