JARVIS V1 — Voice Assistant

JARVIS V1 is the foundational voice assistant: wake word detection, audio capture, Whisper transcription, Ollama inference, command classification (system vs conversational), executor for system actions, and Edge-TTS output. Modular core architecture.

Voice assistants typically rely on cloud APIs. This explores fully local inference with explicit module boundaries for transparency and extensibility.

The repository evidence points to these observable boundaries and technologies:

• Modular Python package (core/, audio/, utils/)
• Whisper for speech-to-text
• Ollama (Llama 3.1) for local LLM inference
• Edge-TTS for speech output
• pygame for audio I/O
• Wake word detection + command recording

• Wake word activation → audio recording → Whisper transcription
• Brain classifies: system command vs conversational response
• Executor runs system commands (open apps, search, type)
• Speaker outputs TTS response
• Config-driven settings (model, voice, sensitivity)

• Whisper latency on CPU-only machines
• Ollama model loading time for first request
• Audio device compatibility across platforms
• Wake word false positives/negatives

• Voice activation and command execution functional
• Whisper + Ollama + Edge-TTS pipeline working
• System commands (open, search, type) executing
• Conversational mode maintaining context

Local voice AI is viable but latency budget is tight — Whisper base.en ~1-2s, Ollama ~2-5s. Streaming TTS would improve perceived responsiveness.