LaunchKit · 2026
Back to Skills

jetson-cuda-voice

Fully offline, CUDA-accelerated local voice assistant pipeline for NVIDIA Jetson.

0
314 downloads
by @nikil511

Setup & Installation

openclaw skills install @nikil511/jetson-cuda-voice

Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:

npx clawhub install jetson-cuda-voice

Version History

v1.1.0Feb 24, 2026 - Version note by skill developer:

v1.1.0: Add setup.sh one-command installer (embeds systemd services inline). Fix manage.sh hardcoded devices — now uses VOICE_MIC/VOICE_SPEAKER env vars. Remove unused json import. Fix fragile test-tts heredoc. Remove cmake from runtime requires. Clean up SKILL.md: quick start section, fixed file structure, removed missing systemd/ dir reference.

What This Skill Does

Offline voice assistant pipeline for NVIDIA Jetson devices. Chains wake word detection, voice activity detection, GPU-accelerated speech-to-text, LLM inference, and neural TTS. Only the LLM call requires internet access.

Running STT and TTS fully on-device eliminates round-trip latency and keeps audio data off third-party servers, which cloud-dependent pipelines cannot offer.

When to use it

  • Hands-free home automation control on a Jetson-based device
  • Private voice queries without sending audio to cloud STT services
  • Multilingual voice assistant that auto-detects spoken language
  • Always-on desk assistant with LED visual state feedback
  • Embedded voice interface for robotics or kiosk projects

Example Workflow

Here's how your AI assistant might use this skill in practice.

INPUT

User says 'Hey Jarvis, what's the weather like today?'

AGENT
  1. 1openWakeWord detects the wake phrase and triggers a two-tone beep with blue LED
  2. 2_measure_ambient() samples 480ms of room noise to set dynamic VAD thresholds
  3. 3transcribe_stream() captures speech via VAD, sends audio to whisper.cpp CUDA server, returns transcript
  4. 4ask_llm() sends transcript plus 20-turn conversation history to OpenRouter API and receives response
  5. 5Piper TTS synthesizes the response offline and aplay outputs it while LED shows cyan then off
OUTPUT

Spoken answer played through ALSA speaker, LEDs return to off state

Requirements

Accounts, API keys, or tools you or your AI assistant may need to set up while using this skill.

OPENROUTER_API_KEY environment variableOpenRouter account (or self-hosted OpenAI-compatible LLM endpoint)NVIDIA Jetson device with JetPack installedReSpeaker USB Mic Array v1.0 or compatible ALSA mic