Skip to content

Microphone Satellite

Take an old Raspberry Pi Zero, plug in a microphone, and let it do almost nothing but listen. The mic satellite runs just enough to notice when someone is speaking — a microphone and voice-activity detection — and streams that audio straight to hivemind-core. It doesn't even wait for a wake word; the server handles that too, along with the transcription, the skills, and the spoken reply. That makes it the simplest software satellite to run and the least demanding on the device. The trade-off is candid: because it forwards every scrap of detected speech, it's chatty on the network and leans hard on the server, so it's happiest on a small, fast home LAN.

In a nutshell

  • On the device: a microphone and VAD. That's it.
  • On hivemind-core: wakeword, STT, intents, and TTS — which means the server needs hivemind-audio-binary-protocol installed.
  • It streams every VAD-gated segment of speech as binary RAW_AUDIO, so it's bandwidth-heavy — a good fit for a handful of devices on a homelab LAN.
  • Serving many users or running at scale? Reach for voice-relay instead — it holds audio back until a local wake word fires.

When to use it

  • Devices with very limited CPU/RAM (Raspberry Pi Zero, recycled phones)
  • Scenarios where you want zero local model downloads
  • Personal homelab with a handful of devices on a fast LAN

When not to use it

mic-satellite streams every detected voice segment to hivemind-core continuously (VAD-gated but not wakeword-gated). This is bandwidth-intensive and puts the full STT load on the server for all speech activity. It is appropriate for a small homelab. For a deployment serving multiple users, prefer voice-relay — local wakeword means audio only leaves the device after activation.

Install

pip install hivemind-mic-satellite

Requires Python ≥ 3.10.

Server requirements

hivemind-core must have hivemind-audio-binary-protocol installed and configured. On its own, hivemind-core does not handle audio.

See Audio Binary Protocol.

Quickstart

Pre-flight: this satellite sends audio to hivemind-core, so hivemind-core MUST have the Audio Binary Protocol configured — it provides the VAD-gated STT, TTS, and wakeword. Without it the satellite connects but its audio is silently dropped, with no error.

1. On hivemind-core — register a client:

hivemind-core add-client --name my-mic-sat
# note the Node ID, access_key and password printed

2. On hivemind-core — grant the client a message type:

hivemind-core allow-msg "recognizer_loop:utterance" <node_id>

A client whose whitelist is empty is denied everything, including the binary audio this satellite sends. Grant at least one message type or the audio never reaches the server.

3. On the satellite — write the identity file:

hivemind-client set-identity \
  --key <access_key> \
  --password <password> \
  --host <server_host_or_ip>

4. Run:

hivemind-mic-sat

Or pass credentials directly without storing them:

hivemind-mic-sat --key <key> --password <password> --host <host> --port 5678 --siteid kitchen

The full flag set is --key --password --host --port --siteid. Each falls back to the identity file written by hivemind-client set-identity.

Note

mic-satellite has no --selfsigned flag. If hivemind-core uses a self-signed certificate, prefer a plain ws:// LAN connection or a properly issued certificate.

Configuration

mic-satellite reads the standard OVOS configuration at ~/.config/mycroft/mycroft.conf.

Plugin type Config key Notes
Microphone listener.microphone.module Required
VAD listener.VAD.module Required
PHAL PHAL.ovos-phal-... Optional, platform support

Example mycroft.conf excerpt:

{
  "listener": {
    "microphone": {
      "module": "ovos-microphone-plugin-alsa"
    },
    "VAD": {
      "module": "ovos-vad-plugin-silero"
    }
  }
}

How audio flows

It helps to trace one spoken sentence end to end. The device does only the first two steps; the moment there's sound, everything else happens on the server and comes back as audio to play:

[Microphone] → [VAD] → raw audio chunks → [hivemind-core]
                                              ↓
                                    [hivemind-audio-binary-protocol]
                                              ↓
                                    WakeWord → STT → IntentService
                                              ↓
                                    TTS synthesis → audio back to satellite
                                              ↓
                                    [Satellite plays TTS audio]

Audio is sent as BINARY messages with payload type RAW_AUDIO (type 1). hivemind-core's AudioBinaryProtocol.handle_microphone_input() processes the stream.

Next

Set up the server side: Audio Binary Protocol.

Source

Validated against the HiveMind source: