"Deformable Spectrograms"

Manuel Reyes-Gomez           Nebojsa Jojic                Daniel P.W. Ellis
Columbia University        Microsoft Research        Columbia University

I.- INTRODUCTION

II.- SPECTRAL DEFORMATION MODEL
     VIDEO 1. - Harmonics transformations maps and
                               harmonics tracking application.

III.- MISSING DATA SCENARIO
       VIDEO 2.- Missing Data Application.
       VIDEO 3.- Missing Data Application.
                               Severe Case.

IV.- TWO LAYER SOURCE-FILTER TRANFORMATIONS
       VIDEO 4.- One layer; Missing Data Application;
                               Introducing the need for two layers.
       VIDEO 5.- Harmonics/Formants Separation.
       VIDEO 6.- Two layers Missing Data Application.

V.-  MATCHING-TRACKING MODEL
       VIDEO 7.- Matching-Tracking Model of a single source
                               signal and signal reconstruction from the
                               model.
        VIDEO 10.- Matching-Tracking Model on
                                  Composed Signals.

VI.- GENERAL  APPLICATIONS.
         - Formants and Harmonics Tracking.
       VIDEO 8.- Formants and Harmonics Tracking.
         - Two Speakers Semi-Supervised Source Separation.
       VIDEO 9.- Two Speakers Semi-Supervised
                              Source Separation.
         - Missing data interpolation and Harmonics/Formants Separation.
       VIDEO 2.- Missing Data Application.
       VIDEO 3.- Missing Data Application.
                               Severe Case.
       VIDEO 5.- Harmonics/Formants Separation.
       VIDEO 6.- Two layers Missing Data Application.
 

RESULTS ON SPECIFIC  APPLICATIONS.

VII.- FEATURES FOR SPEECH RECOGNITION

VIII.- DOMINANT SPEAKER UNSUPERVISED SOURCE SEPARATION.
          - Matching-Tracking Model on Composed Signals.
          - Subband Matching-Tracking Model for Composed Signals.
          - Dominant Speaker Segmentation Results.