Acoustic analysis of Indian classical music using signal processing methods

Identification of music components and further acoustic analysis of Indian classical music are challenging issues, due to large variations in these. This papers consists of two major studies, singing-voice analysis and music-source separation. Indian ragas are divided into lyrical composition and alaap regions, i.e., regions of improvisation that indicate versatility of the singer. Acoustic analyses of both male and female singing voice is carried out by examining the production characteristics. The source feature F₀ is obtained using two methods, autocorrelation of linear prediction (LP) residual and zero-frequency filtering (ZFF), for cross-validation. Combined source-filter feature signal energy is examined to identify the regions of alaap and lyrics composition. Spectrogram using short-time Fourier transform are used for validating the results. The music-source separation is carried out using spectral components and exploiting the repetitive nature in the Indian classical music. A measure of separation quality is proposed using a self-similarity matrix technique, which helps in selecting optimal windowing configurations. Performance evaluation results are encouraging. Study helps making the near-ideal choices of the windowing techniques for better separation of lyric (foreground) and the music part (background) in the music-singing voice mixture. Alaap regions are observed to have higher F₀ for female singers, and lyrics composition higher F₀ for male singers.