Linux Voice Assistant 1.1.17 was released on 27 October 2026. The project, maintained under the Open Home Foundation's OHF-Voice organization, turns a Linux computer such as a Raspberry Pi into a voice satellite for Home Assistant. It talks to Home Assistant over the ESPHome protocol, so Home Assistant sees it as an ESPHome device. Version 1.1.16 came out three days earlier, on 24 October.

A microphone that no longer dies silently

The most important fix is for a failure that was hard to notice. When the microphone stream failed, for example after the computer woke from sleep, a USB interruption or an audio server restart, the audio thread logged an error and stopped. The rest of the process kept running, so the satellite still showed as online in Home Assistant even though it could no longer hear anything. Docker's restart policy never kicked in.

Now the satellite reopens the microphone straight away, and if that fails, tries once more after two seconds. If the second attempt also fails, the whole process exits with an error, so Docker or systemd can restart it. The author notes that this was tested with unit tests only, not with a real failing microphone.

Other changes

  • A sound when the pipeline ends without a reply. If the assistant hears nothing after the wake word, times out or hits an error, it now plays a short, lower-pitched sound before going idle. Before, the session ended silently and you could not tell whether it was still listening. The sound is on by default and can be changed or turned off with --pipeline-ended-sound.
  • Fixed MAC address. Home Assistant's ESPHome integration identifies devices by MAC address, which the satellite takes from its network interface. The new --mac-address option (MAC_ADDRESS in Docker) keeps the same device in Home Assistant when the computer switches between cable and Wi-Fi, and gives several instances on one machine their own identities.
  • Continued conversation with two wake words. With two wake words mapped to two different pipelines, a follow-up question used to go to the first pipeline. The satellite now remembers which wake word started the conversation.
  • Lower memory use. Wake word models that are no longer selected are unloaded.
  • Smaller Docker image, down from about 1.46 GB to about 600 MB. The author tested the build on amd64 only, without real audio or an arm64 build.

If you run a Linux Voice Assistant satellite, especially on a machine that sleeps or uses a USB microphone, this update is worth installing. If you start it with Docker or systemd, make sure a restart policy is set, otherwise the new exit on a dead microphone just stops the satellite.