A Comparative Study of Audio Source Separation Techniques on Indian Classical Music: Performance Evaluation and Analysis

Music, a rich form of auditory expression, encompasses a diverse array of sounds, broadly classified into two main categories: Harmonics and Percussions. Harmonic sounds, distinguished by their distinct pitches, give rise to melodies and chords, while Percussive sounds, produced by instruments like drums and snares, introduce sudden pitch changes. Real-world music compositions often combine these two classes, creating challenges in processing and analysis. This research’s primary goal is to develop and compare effective algorithms for separating audio, into two distinct components, the isolating harmonic component and the percussive component. Exploiting the localization properties of these sound classes, with percussive sounds in time and harmonic sounds in frequency, it aims to achieve precise separation. This separation finds practical use in music remixing, key signature detection, instrument transcription, chord identification, and rhythmic analysis. The paper delves into theoretical foundations, implementation details, and comprehensive spectrogram analyses to demonstrate the approach’s effectiveness.