A framework for music analysis/resynthesis based on matrix factorization

This webpage contains sound examples corresponding to the paper with the above title, published at ICMC 2014.

A corresponding software tool can be downloaded here.

Scramble

• Sound 1: Random scramble with K=20

  Input		(excerpt from "Tristan und Isolde" by R. Wagner)
  Output



• Sound 2: Random scramble with K=20

  Input		(excerpt from "A German Requiem" by J. Brahms)
  Output

Rank

• Sound 3: Direct ranking (increasing brighness against increasing sparsity), K=50

  Input		(excerpt from "Computer World" by Kraftwerk)
  Output



• Sound 4: Inverse ranking (increasing brighness against decreasing sparsity), K=50

  Input
  Output

Constrained scramble

• Sound 5: Scramble constrained to low-brightness components, K=20

  Input
  Output

Cross-synthesis

• Common inputs for all cross-synthesis examples:

  Source
  Target

• Sound 6: Cross-synthesis based on spectral similarity, injective mapping, Ks=30, Kt=30

  Output

• Sound 7: Cross-synthesis based on spectral similarity, merge mapping, Ks=30, Kt=30

  Output

Constrained cross-synthesis

• Sound 9: Cross-synthesis based on spectral similarity, high-brightness target spectra discarded, injective mapping, Ks=30, Kt=30

  Output

• Sound 9b: Cross-synthesis based on spectral similarity, high-brightness target spectra discarded, merge mapping, Ks=30, Kt=30

  Output

Wiener-based resynthesis

• Common input for sounds 10 and 11:

  Input

• Sound 10: Direct ranking, K=50, Griffin and Lim resynthesis

  Output

• Sound 11: Direct ranking, K=50, Wiener resynthesis (=Sound 3)

  Output