110 likes | 212 Views
Extracting Individual Tracks from Polyphonic Music. Nick Starr, Computer Systems Lab. Source Separation. • "Cocktail Party" analogy • Magnetoencephalography (MEG) - Brain imaging with magnetic fields • Music. Source Separation (cont). x = Observed s = Source a = Mixing constants
E N D
Extracting Individual Tracks from Polyphonic Music Nick Starr, Computer Systems Lab
Source Separation • "Cocktail Party" analogy • Magnetoencephalography (MEG) - Brain imaging with magnetic fields • Music
Source Separation (cont) x = Observed s = Source a = Mixing constants • x is known • a can be computed • With x and a known, s can be solved for
Methods of Source Separation • Principal Components Analysis • Computational Auditory Scene Analysis • Independent Component Analysis
Fourier Transforms (cont) The Math: The Explanation: • Maps domain from time to frequency • Easily invertible • Many different computational techniques
Independent Component Analysis • Decomposition • Fourier Transform, separate moduli and phases • Matrices, matrices, matrices • Classification • Center of research • Separate important parts from unimportant • Recombination • Bring back the phases • Combine with classified moduli • Inverse Fourier Transform
Results Segmentation fault
Challenges • Understanding the math • C and GSL • Mathematica File I/O