HiveMind Microphone Satellite

hivemind-mic-satellite is the smallest HiveMind satellite. It runs only a microphone plugin and a VAD (voice activity detection) plugin on the device.

Audio streams from this device to a HiveMind server that runs hivemind-audio-binary-protocol. The server handles wakeword detection, speech-to-text, intent processing, and text-to-speech synthesis. The satellite receives the synthesized speech audio and plays it back locally. No local STT or TTS models are needed, so the satellite can run on cheap, low-power hardware such as a Raspberry Pi Zero.

Satellite spectrum: where does processing happen?

Server requirements

The default hivemind-core does not include audio processing. Install hivemind-audio-binary-protocol on the server to enable server-side wakeword, STT, and TTS.

Why mic-satellite, and when not to use it

mic-satellite exists for one reason: device resources. With only a microphone and VAD on-device, it runs on cheap hardware, such as a Raspberry Pi Zero or a recycled phone, with no local models. The hive owns everything else: wakeword, STT, intent, and TTS. It gates them behind the same access-key authentication as the rest of the mesh. The hive operator chooses the engines, models, and voice for every connected satellite. A satellite cannot override them. This is the same ownership model as voice-relay. See its docs for the "HiveMind as a service" framing.

There is a trade-off. Because there is no local wakeword, the satellite streams every detected voice segment upstream (gated by VAD, not by a wakeword). This continuous audio stream uses more bandwidth and puts the full STT load on the server for all speech, not just commands. This works well for a homelab with a handful of personal devices, where on-device resources are the binding constraint. It does not scale for HiveMind-as-a-service across many tenants, because streaming raw audio per client costs too much. For a service-style deployment, prefer voice-relay: a local wakeword means audio leaves the device only after activation.

Install

pip install hivemind-mic-satellite

Requires Python 3.10 or later.

Quickstart

1. On the hive (server): create an access key for this device:

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

2. On the satellite device: set the identity:

hivemind-client set-identity \
  --key <access_key> \
  --password <password> \
  --host <hive-host-or-ip>

3. Run:

hivemind-mic-sat

Or pass credentials directly without storing them:

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

Minimal configuration

The satellite shares the standard OpenVoiceOS config file ~/.config/mycroft/mycroft.conf. At minimum you need a microphone plugin and a VAD plugin:

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

See docs/configuration.md for all options, plugin selection, and audio device tuning.

Supported plugins

Features handled server-side, not on this device

• STT (speech-to-text)
• TTS (text-to-speech) synthesis
• Wakeword detection
• Continuous listening, hybrid listening, sleep mode, recording mode
• Multiple wakewords
• Dialog transformer plugins

Documentation

Full documentation is in docs/:

• Overview and satellite spectrum
• Getting started
• Configuration reference
• Architecture (advanced)
• Deployment (systemd, Raspberry Pi)
• Testing (e2e suite, mocked hardware)
• Troubleshooting

Related

License

Apache-2.0: see LICENSE.

Transformer pipelines

The satellite can run OVOS transformer plugins client-side, configured in this device's mycroft.conf:

• audio_transformers — applied per chunk to microphone audio before it is streamed to the server (e.g. denoise).
• tts_transformers — applied to received TTS audio before playback (e.g. per-device sound effects).

{
  "tts_transformers": {
    "ovos-tts-transformer-sox-plugin": {"pitch": 300}
  }
}

Loading is opt-in: a plugin only runs if named in its section. Avoid double-processing: if the HiveMind server also enables the same pipeline (hivemind-audio-binary-protocol runs audio transformers server-side), the audio gets processed twice — enable each plugin on exactly one side.