Deploy a streaming speech recognition (ASR) and voice synthesis (TTS) service on your edge device — Jetson Orin, RK3576, RK3588, or a Pi 5.
What you'll get:
Requirements: SSH access to device · Internet to pull Docker image and download models
Deploy the speech service to your edge device. Models auto-download on first start.
Deploy over SSH to a Jetson Orin. Runs on the GPU, supports multiple languages and voice cloning. Needs at least 7.5 GB of free disk.
recomputerjetson-zh-enen0zh_en mode: 0-9, en mode: 0-10 (default 8 = bf_isabella)
0.850.5 = slow, 1.0 = normal, 1.5 = fast
| Issue | Solution |
|---|---|
| SSH connection failed | Verify IP and credentials. Try ssh username@ip from your computer |
| Image pull slow | The image is ~2 GB compressed. Ensure stable internet on the device |
| Service not starting | Check logs: ssh user@ip "cd openvoicestream && docker compose logs" |
| Health check fails | First startup takes ~40 seconds for model warmup. Wait and retry |
| Out of memory | Ensure Jetson has 8GB+ RAM and no other GPU tasks running |
| NVIDIA runtime missing | Install: sudo apt install nvidia-container-toolkit && sudo systemctl restart docker |
Service is running at http://<device-ip>:8621. Quick test:
curl http://<device-ip>:8621/health
# Expected: {"asr": true, "tts": true, "streaming_asr": true}
curl -X POST http://<device-ip>:8621/tts \
-H "Content-Type: application/json" \
-d '{"text": "Hello, I am your voice assistant.", "sid": 0}' \
--output test.wav
Try the deployed speech service directly from this page. Enter the device IP address, then use the panels below to test speech recognition and voice synthesis.
Press and hold the Record button to speak. Your speech will be recognized in real-time and the transcribed text will appear on screen.
Type any text and click Generate to hear it spoken.
| Issue | Solution |
|---|---|
| Microphone not working | Allow microphone access when prompted by your browser |
| ASR shows no results | Verify the service is running: curl http://<ip>:8621/health |
| TTS playback silent | Check browser audio is not muted. Try a shorter text first |
Congratulations! Your local voice service is running.
http://<device-ip>:8621/health in your browser — all fields should show truecurl command above| Endpoint | Method | Description |
|---|---|---|
/health | GET | Service health check |
/asr/stream | WebSocket | Real-time streaming speech recognition |
/tts | POST | Text-to-speech (returns WAV) |
/tts/stream | POST | Streaming text-to-speech (returns raw PCM) |
/asr | POST | Offline speech recognition (upload WAV file) |