javichur/fitness-voice

Leverages TensorFlow for offline hotword detection ("coach") to minimize data transmission, processes body pose recognition entirely client-side to preserve privacy, and uses wit.ai for hands-free voice command parsing only after wake-word activation. The architecture prioritizes local inference—webcam frames and pose data remain on-device while AI-generated synthetic voices deliver real-time coaching feedback without requiring manual input triggers.

No commits in the last 6 months.