J.A.R.V.I.S.

The server for a voice-controlled AI assistant that listens, processes, and responds like Tony Stark's AI assistant, J.A.R.V.I.S., from the MCU.

Speed

On average, J.A.R.V.I.S. responds in about 2-4 seconds on CPU after the user finishes talking.

How It Works

1. Wake Word Detection - Streams microphone over websocket to server, and listens continuously for "Hey Jarvis" using OpenWakeWord
2. Speech-to-Text - Transcribes your voice input using Faster Whisper with Whisper Distil, and Silero VAD
3. AI Processing - Generates near-instant responses using Cerebras and a J.A.R.V.I.S. optimised system prompt
4. Text-to-Speech - Send the response from Piper TTS to client

Technologies

• OpenWakeWord - Wake word detection using hey_jarvis_v0.1.tflite
• Faster Whisper - Speech recognition using distil-small.en
• Silero VAD - Voice activity detection
• Cerebras - AI inference using qwen3.8-27b
• Piper - Neural text-to-speech using jarvis-medium.onnx