# Voice Docker image (CGO + sherpa-onnx)

Prod compose: `leggnom/unishift_voice` in [../docker-compose.yaml](../docker-compose.yaml).

Build/push from the service submodule (multi-stage Dockerfile downloads Linux sherpa libs):

```bash
cd ../../services/voice
./scripts/publish.sh                  # linux/amd64 → Docker Hub
./scripts/publish.sh --build-only     # local load only
TAG=0.0.2 ./scripts/publish.sh
```

Do **not** build voice with `CGO_ENABLED=0` — TTS/STT become stubs.
