Description

Input Sound

Baseball practice sounds.

Synthesis Constraints

 
Source 1  

 4 source segments consisting of a spectator laughing sounds.

 
Target 1  

 Hard -1 exclusion constraint most of the time, low weight soft weights for some periods.

 

 

Source 2  

  2 source segments consisting of a spectator calling ball.

 
Target 2  

 Hard -1 exclusion constraint most of the time, low weight soft weights for some periods.

 

 

Source 3  

  Many source segments consisting of a loud ball hitting sounds.

 
Target 3  

 Hard -1 exclusion constraint most of the time, low weight soft weights for some periods.

 

 

Comments

Uses a hard exclusion sounds during most periods and introduces source sounds at given intervals. We only use very light weights so that sources are introduced slightly more frequently than unconstrained synthesis. Notice overlapping source as well as targets.

 

 

 

Self-Similarity Synthesis

 

Segmented Original Sound

 

Shows the segmentation boundaries on original sound

 

 

New Sound

 

(top) Shows the old segmented sound along each constraints source segments. (middle) Shows the synthesized sound. (bottom) Shows Constraint targets along with how the origin of the grains in these targets regions.

(top left) Shows the original probabilities for every grain at each instant of the synthesis over time. (top right) Shows how the original propabilities where rescaled according to the weights at that current in the synthesis. (bottom left) Illustrates the difference between the two initial and new probabilities. (bottom right) Shows the target weights for each constraints. The surface represents the final calculated weights for each grain. The final weights assigned to each grain depends on wether it satisfies or not the current constraint targets. A weight of zero means that it-s original probability is left unchanged. Shows the target weights for each constraints. The surface represents the final calculated weights for each grain. The final weights assigned to each grain depends on wether it satisfies or not the current constraint targets. A weight of zero means that it-s original probability is left unchanged. Shows which grains were excluded from the synthesis from a hard -1 constraint. Shows which grains were hard included in the synthesis from a hard 1 constraint.

 

 

 

 

Natural-Grain Synthesis

 

 

Segmented Original Sound

Shows the segmentation boundaries on original sound

 

New Sound

(top) Shows the old segmented sound along each constraints source segments. (middle) Shows the synthesized sound. (bottom) Shows Constraint targets along with how the origin of the grains in these targets regions.

(top left) Shows the original probabilities for every grain at each instant of the synthesis over time. (top right) Shows how the original propabilities where rescaled according to the weights at that current in the synthesis. (bottom left) Illustrates the difference between the two initial and new probabilities. (bottom right) Shows the target weights for each constraints. The surface represents the final calculated weights for each grain. The final weights assigned to each grain depends on wether it satisfies or not the current constraint targets. A weight of zero means that it-s original probability is left unchanged. Shows the target weights for each constraints. The surface represents the final calculated weights for each grain. The final weights assigned to each grain depends on wether it satisfies or not the current constraint targets. A weight of zero means that it-s original probability is left unchanged. Shows which grains were excluded from the synthesis from a hard -1 constraint. Shows which grains were hard included in the synthesis from a hard 1 constraint.

 

 

 

 

Wavelet Synthesis

 

 

New Sound