HiveMind Voice Satellite Documentation

July 31, 2026 · View on GitHub

HiveMind Voice Satellite (hivemind-voice-sat) is a full-stack OVOS satellite. Every audio-processing stage runs on this device. The satellite captures the microphone locally, VAD filters silence, a wakeword engine listens for the trigger phrase, STT converts speech to text, and the satellite forwards that text to the hive. Responses arrive as text and TTS speaks them locally. No raw audio leaves the device.

Satellite spectrum

This table shows where each satellite sits on the local-vs-remote processing axis.

SatelliteMicVADWakewordSTTTTSWire payload
HiveMind-clin/an/an/an/an/atext only
hivemind-mic-satelliteyesyesn/aserverserverraw audio stream
HiveMind-voice-relayyesyesyesserverserverpost-wakeword audio
HiveMind-voice-satyesyesyesyesyestext utterances

When to choose voice-sat:

  • The device has enough CPU/GPU to run STT and TTS models locally (Raspberry Pi 4/5, x86, or any device with ≥2 GB RAM to spare).
  • Bandwidth to the hive is limited or metered.
  • Audio privacy is a requirement. No spoken audio leaves the device.
  • You want the hive to remain lightweight and serve many clients.

When to choose a thinner satellite:

  • The device is a minimal microcontroller or SBC with little compute.
  • You want centralized STT/TTS model management on the server.

Pages

PageAudience
Getting startedFirst-time setup, pairing, first run
ConfigurationCLI flags, config file, plugin selection
ArchitectureOn-device pipeline, wire protocol, trade-offs
Deploymentsystemd service, Raspberry Pi, audio hardware
TroubleshootingCommon failure modes and fixes
TestingTest tiers, the bus-client 2.x stack, running e2e