GitHub - AdBusterOfficial/Adbuster--WinApp: A small Windows tool that lowers loud TV ads automatically and restores normal volume smoothly.
Pangram verdict · v3.3
We believe that this document is mainly AI-generated, with some AI-assisted content.
AI likelihood · overall
AIArticle text · 1,438 words · 6 segments analyzed
AdBuster 2.0 PRO – TV Volume Stabilizer
Offline audio stabilization engine powered by ML/AI and CEPA Logic.
⚠ CEPA Logic is covered by an additional proprietary license included in this repository. See: CEPA_LOGIC_LICENSE.txt
A small Windows tool that lowers loud TV ads automatically and restores normal volume smoothly.
Latest Windows build: AdBuster_2.0_PRO_Test.zip
Portable Windows application — no installation required. Download → Unzip → Run Start.bat
✨ Overview AdBuster PRO is a lightweight Windows application that monitors TV audio in real time and automatically reduces the volume when loud commercials appear. When normal audio returns, the volume is restored smoothly and naturally. No cloud. No accounts. No telemetry. No data collection. Everything runs locally on your PC or laptop.
🚀 Quick Start
Download the latest release from the Releases page. Unzip the folder anywhere on your PC. Run Start.bat to launch all modules automatically. (Optional) Configure your Broadlink IR device for volume control. Start watching TV — AdBuster will detect loud ads and stabilize audio in real time.
🎯 Key Features
🔊 Real‑time loudness detection 📉 Automatic volume reduction during commercials 🔄 Smooth volume restoration after ads 📡 IR blaster support (Broadlink RM series) ⚙ Configurable thresholds and reaction times 🖥 Runs silently in the background 🔐 Fully local — no internet required
⭐ Why This System Is Unique AdBuster 2.0 PRO is not a typical audio tool. It is the only Windows application that combines: 🔹 Real‑time audio analysis (DSP) A dedicated audio engine extracts loudness, stability, spikes and long‑term patterns in real time — fully offline.
🔹 CEPA Logic — human‑like decision framework A proprietary contextual reasoning layer that interprets audio events the same way a human would: detects disruptive spikes avoids over‑correction restores volume smoothly adapts to context and time of day Note: The public release of AdBuster 2.0 PRO includes the CEPA Core (Perception → Behaviour → Action). The full CEPA Logic v2 framework described in the documentation is proprietary and not included in the test build. 🔹 Offline machine learning (FREE + PRO) Two independent ML engines:
model.pkl — lightweight, short‑term classifier model_deep.pkl — PRO‑level long‑term pattern recognizer
Both run 100% locally. 🔹 Broadlink IR automation AdBuster is the only Windows tool that:
detects loud commercials decides what to do sends IR commands automatically stabilizes TV volume without user interaction
🔹 Fully offline architecture All processing happens on the user’s device. 🔹 Unique CEPA + ML + IR pipeline This combination does not exist in any other Windows application or smart‑home tool. AdBuster 2.0 PRO stands alone as a complete, intelligent, privacy‑first audio stabilization system.
🛠 How It Works
A microphone listens to the TV audio. When a sudden loudness spike is detected (typical for ads), AdBuster PRO decides whether to lower the volume. The app sends a command to the local Flask server (VolMaster). VolMaster forwards the IR command to the Broadlink device. Broadlink sends the IR signal to the TV. When audio returns to normal, the app restores the previous volume level.
All logic is processed locally — no cloud, no external servers.
📐 Architecture Diagram (High‑Level) Below is a simplified architecture diagram showing how the main components of AdBuster PRO interact. This diagram does not reveal internal algorithms or implementation details.
┌────────────────────────────┐ │ MICROPHONE │ │ (raw audio signal) │ └──────────────┬─────────────┘ │ audio frames ▼ ┌────────────────────────────┐ │ AUDIO ENGINE │ │ RMS / mean / std / range │ │ smoothing / GUI level │ └──────────────┬─────────────┘ │ features ▼ ┌────────────────────────────┐ │ ML CONTROLLER │ │ (ml_controller.py) │ │ - collects RMS history │ │ - calls ml_pro_update() │ │ - sends RMS to Aduster ML │ │ - toggles ad_mode │ └──────────────┬─────────────┘ │ RMS history ▼ ┌────────────────────────────┐ │ ADUSTER ML │ │ (model.pkl / deep.pkl) │ │ - long-term pattern check │ │ - AD / NORMAL classifier │ └──────────────┬─────────────┘ │ AD / NORMAL state ▼ ┌────────────────────────────┐ │ CEPA ENGINE │ │ - contextual analysis │ │ - deadzone / stable zone │ │ - spike detection │ │ - anti-spam limits │ │ - VOL_UP / VOL_DOWN │ └──────────────┬─────────────┘ │ decision ▼ ┌────────────────────────────┐ │ AdBuster App Logic │ │ - pending_cmds queue │ │ - cooldown timers │
│ - 2x up/down safety limits │ │ - sends HTTP commands │ └──────────────┬─────────────┘ │ HTTP GET /send?cmd=... ▼ ┌────────────────────────────┐ │ VolMaster Server │ │ (Flask API, port 5000) │ └──────────────┬─────────────┘ │ IR command ▼ ┌────────────────────────────┐ │ Broadlink IR Blaster │ │ (RM Mini / RM Pro series) │ └──────────────┬─────────────┘ │ infrared signal ▼ ┌────────────────────────────┐ │ TV / Audio Device │ │ (volume changes applied) │ └────────────────────────────┘
📞 IR Command Policy AdBuster’s automation layer uses only two IR commands:
VOL_UP — increase volume VOL_DOWN — decrease volume
These are the only commands generated by CEPA, the ML controller, and the app logic. The MUTE and POWER commands are manual‑only (UI buttons) and are never used by the automation pipeline. Automatic volume changes are always constrained by:
a 2× up/down safety limit cooldown anti‑spam CEPA validation
🧩 Architecture Overview — 3 Modules + CEPA Brain AdBuster 2.0 PRO is built from three independent modules that work together in real time. At the center sits CEPA, the decision‑making brain that connects everything. 🧮 AdBuster — Detection Engine Captures audio from the microphone Computes RMS, peaks, dynamics, drift and spikes Builds short‑ and long‑term loudness history Generates structured audio events for CEPA Handles smoothing, deadzone, stable zone, cooldowns and safety limits Role: real‑time detection + runtime logic 🧿 Aduster ML — Machine Learning
Engine Uses on‑device models (model.pkl and model_deep.pkl) Classifies AD / NORMAL behaviour based on RMS history Validates CEPA’s interpretation Does not send IR commands Does not control volume Role: pattern validation + long‑term statistical context 🧰 VolMaster — IR Control Server Local Flask server that communicates with Broadlink RM devices. Sends VOL_UP / VOL_DOWN IR commands. Compatible with any IR‑based TV or audio device. Exposes /send?cmd=... and /status endpoints. Role: execution layer responsible for IR command delivery. 🧠 CEPA — The Decision‑Making Brain CEPA (Contextual Event Pattern Analysis) is not a separate module. It is the internal intelligence layer inside AdBuster 2.0 PRO. CEPA receives: audio events from AdBuster AD / NORMAL state from Aduster ML recent actions, cooldowns and safety limits from the app logic CEPA performs: contextual pattern analysis human‑like filtering anomaly detection stability evaluation safety enforcement CEPA decides: VOL_DOWN VOL_UP PASS BLOCK AdBuster then forwards the final command to VolMaster, which executes it via IR. In simple terms: AdBuster detects → Aduster ML validates → CEPA decides → VolMaster executes. CEPA Logic — Technical Foundations CEPA is a lightweight analytical framework for interpreting audio as a sequence of contextual events rather than raw signal frames. It operates on four core concepts:
Contextual Event Pattern A structured representation of how audio events evolve over time, including their direction, stability, transitions, and contextual relationships. CEPA treats events as semantic units rather than isolated DSP measurements.
Behaviour‑Based Audio A signal interpretation model focused on behaviour (stability, drift, anomalies, resets) instead of amplitude or spectral values. This enables deterministic detection of unexpected behaviour in real‑world audio.
Event‑Level Inference A decision‑making layer that evaluates relationships between events rather than individual samples or frames. Inference is based on event sequences, priorities, and contextual state transitions.
Contextual Audio Logic A rule‑based logic system that interprets event patterns in the context of signal history, system state, and previous decisions. This allows CEPA to produce stable, predictable reactions without machine‑learning models.
🔄 CEPA Logic Diagram (Decision Flow) Below is a simplified decision‑flow diagram showing how CEPA (Contextual Event Pattern Analysis) interprets audio behavior and makes human‑like volume control decisions. This diagram represents CEPA’s internal logic only. CEPA interprets loudness behavior in context, rather than reacting to raw classification signals — making the system stable, human‑like, and resistant to natural volume fluctuations. ┌────────────────────────────────────────────┐ │ CEPA INPUTS │ │ - RMS trend (short / long) │ │ - sudden spikes │ │ - drift up / drift down │ │ - AD / NORMAL mode │ │ - recent volume actions (history) │ │ - cooldown & safety state │ └───────────────────────────────┬────────────┘ │ ▼ ┌────────────────────────────────────────────┐ │ PERCEPTUAL CONTEXT │ │ - is volume stable? │ │ - is change gradual or sudden? │ │ - is this a spike or real increase? │ │ - is this a natural fluctuation? │ │ - is this AD mode (react faster)? │ └───────────────────────────────┬────────────┘ │ context_state ▼ ┌────────────────────────────────────────────┐ │ HUMAN‑LIKE FILTERS │ │ - ignore micro‑fluctuations │ │ - ignore short spikes │ │ - apply deadzone (no reaction zone) │ │ - apply stable zone (PASS) │ │ - apply hysteresis (avoid oscillation)