Real-Time Detection of Free-to-Air Television Commercials in Audio Using Feature Extraction and Machine Learning

Detecting television commercials is essential for the advertising, broadcasting, and content analysis industries. In that sense, traditional methods, which rely on audio-visual cues or embedded markers, have evolved with machine learning (ML) and artificial intelligence advances. However, while highly accurate, deep neural networks (DNNs) are computationally intensive and often unsuitable for real-time detection. This paper proposes a real-time method for detecting television commercials in audio using feature extraction and ML models. Audio content is segmented through windowing, while features are extracted from each element. Various models are evaluated, and optimization is performed through feature selection and grid search. The proposed method achieved an accuracy of 99.79% with the best-performing model while maintaining low computational complexity. This approach provides a faster and more efficient solution than usual DNN-based methods, making it suitable for real-time applications. It also enhances the ability to replace commercials with alternative content, improving airtime optimization and overall viewer experience.