Willow Esspresif functionality in a container

[GregJohnStewart]

Hey guys! This looks like a fairly interesting project.

I was wondering if it were possible to get the functionality of Willow in a docker container? I realize this is likely a big pull, but the functionality seems pretty great but my usecase would become more complex for the "extra" esp32 hardware.

[hamishcunningham]

IIRR there used to be runtime docker support, and I think the build is still containerised, and some of the services. I remember trying to put everything in containers once but hit latency issues.

If you're asking can you use this without using ESP-BOX then docker won't help I think, and the answer is probably no.

Hth, H

[adityashirsatrao007]

/assign

[adityashirsatrao007]

Hi @GregJohnStewart, just to add to what @hamishcunningham mentioned -

If your use case is primarily about getting the voice assistant/inference capabilities without the ESP32 hardware, you might want to check out the Willow Inference Server which already runs in Docker and provides:

• ASR/STT (speech-to-text via Whisper)
• TTS (text-to-speech)
• LLM integration
• WebRTC support for streaming audio directly from browsers

WIS can work standalone for transcription/TTS tasks or with other clients beyond the ESP32 hardware. The WebRTC client at \https://[your-wis-host]:19000/rtc\ lets you interact with it directly from a browser.

That said, the core Willow firmware is pretty tightly coupled to the ESP32 hardware (audio pipelines, wake word detection on the device, specific board configurations, etc.) so a full port to run in a container would be a significant architectural change.

What specific functionality were you hoping to use? That might help identify whether WIS or another approach would work for your use case.

[GregJohnStewart]

Interesting! I saw that project but had (wrongly, I guess) that it was meant as a second step in data processing to the ESP32 solution.

Since it's not listed here, my only question is does it support wake wording? And I know other projects support intent recognition, didn't know if that was under the umbrella as well

[adityashirsatrao007]

Hey @GregJohnStewart - good questions!

So wake word detection isn't actually handled by WIS - on the ESP32 that's done locally using openWakeWord. For a container-only setup you'd need something separate for that, or just go push-to-talk style.

For intent recognition, WIS has LLM support so you can pipe the transcription through that for parsing commands. The HA integration does this - audio -> WIS -> transcription -> Home Assistant's conversation API.

If you want everything containerized with wake word support, check out Rhasspy (https://github.com/rhasspy/rhasspy3) - it's built for exactly that use case.