Ad Buster 2.0 PRO is a portable Windows application that monitors TV audio through a microphone and automatically reduces volume during loud commercials, then restores it smoothly when programming resumes. The tool requires Broadlink RM series IR blasters for volume control and runs entirely offline — no cloud, accounts, or telemetry. Users download a zip, unzip, and run Start.bat to launch the detection engine, ML classifier, and local Flask server on port 5000.

The architecture splits into three modules coordinated by CEPA Logic, a proprietary contextual reasoning layer. The Ad Buster detection engine computes RMS, peaks, drift, and spike metrics in real time. Aduster ML runs two on-device models — model.pkl for short-term classification and model_deep.pkl for long-term pattern recognition — validating whether audio segments represent ads or normal content. CEPA then applies human-like filtering: deadzone enforcement, anti-spam limits, cooldown timers, and a 2× up/down safety constraint before queuing VOL_UP or VOL_DOWN commands.

Vol Master, a lightweight Flask server, translates HTTP commands into IR signals sent to the Broadlink device, which blasts them to the TV. Only VOL_UP and VOL_DOWN are automated; MUTE and POWER remain manual. The pipeline — detect → validate → decide → execute — operates fully locally, a combination the author claims is unique among Windows tools.

For developers, the project demonstrates a practical edge-ML deployment: scikit-learn models serialized as pickle files, real-time DSP feature extraction, and a local REST API bridging inference to hardware control. The CEPA framework's event-based reasoning — treating audio as behavioral patterns rather than raw amplitude — offers a template for context-aware automation beyond volume leveling.