What it does

SUSI Linux is the client-side glue that listens for the hotword “Susi,” pipes your voice through speech-to-text engines, sends the result to a SUSI.AI server, and speaks the answer back. It runs on desktop Linux or headless Raspberry Pi hardware, with a simple GTK interface for the former. Think of it as the local ears and mouth for a larger AI ecosystem.

The interesting bit

The project is built as a finite state machine in Python, and it treats cloud dependency as a bug to be routed around. Google services are the default, but lose internet and it automatically falls back to offline PocketSphinx for recognition and Flite for speech synthesis. The README’s roadmap ambition is “washing machines and car systems”—which may be aspirational, but the architecture (swappable STT/TTS engines, local server option, config-driven) is genuinely aimed at embedded ubiquity rather than just another Echo clone.

Key highlights

• Pluggable speech stack: Google, IBM Watson, Bing, PocketSphinx, or DeepSpeech for STT; Google, Watson, or Flite for TTS
• Hotword detection via Snowboy or PocketSphinx, with optional physical wake button
• Designed to pair with a self-hosted SUSI.AI server for “maximum privacy. No cloud needed”
• Installed via SUSI Installer with systemd service files and a susi-config CLI tool
• Verbose logging mode for development via -v / -vv flags

Caveats

• The README explicitly marks two release blockers as currently failing: switching between online and offline modes, and booting offline without a connection
• DeepSpeech integration is listed as “WORK IN PROGRESS”
• Manual installation requires configuring wrapper scripts to point at your config.json path

Verdict

Worth a look if you’re building a custom voice assistant on a Pi and want control over your speech pipeline without writing the state machine from scratch. Skip it if you need something that reliably works offline today—the README is honest that it doesn’t yet.