Feature Engineering for Music/Speech Detection in Costa Rica Radio Broadcast

The exponential growth of audio data in radio broadcasts has generated the need for efficient tools for their manipulation and analysis to develop systems such as audio content classification and enhance user experience. In this study, we explore the application of classifiers to discriminate between speech and music in Costa Rican radio broadcasts. The main purpose is first to select the best features for classification algorithms to obtain the best classification performance in terms of computational cost. The study presents a comprehensive comparative analysis of feature-selection methods, introducing a novel proposal based on a voting mechanism that integrates various feature-selection techniques. The research contributes to refining audio content classification systems and analysis for particular accents within broadcast contexts.