Examples

Here are several examples discussed in my article "Adaptive Concatenative Sound Synthesis and Micromontage Composition", in Computer Music Journal. Click on entries in the "Result" column to hear the synthesis. The "Target" column countains that original sound used, and information about where it came from, and its analysis parameters. For instance the Schoenberg target came from his fourth string quartet and was analyzed with a window size of 4096 points, and a hop size of 2048 points. The "Corpus" column contains information about the corpus content. The Primates corpus was analyzed with a window size of 16,384 samples, hop size of 1024 samples, and contains a total of 18,861 feature vectors, or points. The "Synthesis Window" columns contain information about the synthesis. For instance sound example "B2" was synthesized with a window and hop size of 256 samples, and a rectangular window shape. The "Matching Criteria" column contains (if I remembered) the feature vector search parameters. Finally the "Other Options" column contains the synthesis options I specified, or anything else of note (Figure 7).


These examples show how the algorithm perfectly reconstructs the target in the first case (01), and when given the opportunity with relaxed selection thresholds slightly scrambles the target (02).

  Synthesis Window  
Result Target Corpus Size Skip Shape Matching Criteria Other Options
01 Mahler, Ritenuto
(Second Symphony)
(2048,1024)		 Mahler, Ritenuto
(934 points)		 2048 1024 Hann RMS ± 0%
Spectral Rolloff ± 0%		 none
02     2048 1024 Hann RMS ± 10%
Spectral Rolloff ± 10%		 Random Match
 

Reconstructing Mahler's crescendi with many different corpora and selection criteria.

  Synthesis Window  
Result Target Corpus Size Skip Shape Matching Criteria Other Options
A1 Mahler, Ritenuto
(Second Symphony)
(2048,1024)		 Primates
(16384,1024)
(18,861 points)		 16384 1024 Hann RMS ± 5%
Spectral Rolloff ± 10%		 Mixed with Mahler
A2   Primates 16384 4096 Hann RMS ± 5%
Spectral Rolloff ± 10%		 none
A3   Animals
(22050,2048)
(80,003 points)		 2048 1024 Hann RMS ± 5%
Spectral Rolloff ± 5%		  
A4   Tea Cans
(1024,512)
(3,143 points)		 1024 512 Hann (can't remember)  
A5   Imam Chanting Koran
(8192,2049)
(4,682 points)		 4096 2049 Hann Spectral Centroid ± 0.1%
Spectral Rolloff ± 0.1%		 Force RMS
Extend Matches
A6   John Cage
Music for Voice
(2048,1024)
(288,084 points)		 5000 2250 Hann (can't remember) none
A7   Lawrence Welk
3-CD set
(2048,1024)
(381,361 points)		 11000 5000 Tukey, 25%    
A8   Schoenberg
String Quartets 1-4
(2048,1024)
(359,485 points)		 10000 1024 Hann Spectral Centroid ± 0.05%
Spectral Rolloff ± 0.1%		 Force RMS
Extend Matches
 

Speech is a great source of features. Using the flute (B6) preserves much of the sibilance.

  Synthesis Window  
Result Target Corpus Size Skip Shape Matching Criteria Other Options
B1 George W. Bush
(512,256)		 Primates 256 256 Rectangle (can't remember) none
B2   Primates 512 256 Hann    
B3   Primates 2048 256 Hann    
B4   Tea Cans 512 256 Hann    
B5   Mahler, Ritenuto
(2048,1024)		 256 256 Rectangle    
B6   Bach, Partita 1
(2048,512)
(56,657 points)		 2048 256 Hann Spectral Centroid ± 1%
Spectral Centroid ± 1%		 Force Match
Force RMS
 

Longer speech example. Alto saxophone (C3) does a remarkable job at preserving sibilance.

  Synthesis Window  
Result Target Corpus Size Skip Shape Matching Criteria Other Options
C1 "Congregation"
(Ministry, "Psalm 69")
(512,256)		 Tea Cans 512 256 Hann (can't remember) none
C2   Animal Sounds 22050 1024 Hann    
C3   Anthony Braxton
(sax)
(2048,1024)
(102,505 points)		 2048 256 Hann Spectral Centroid ±0.1%
Spectral Rolloff ±0.1%		 Force RMS
Extend Matches
C4   Lawrence Welk 8192 4096 Hann Spectral Centroid ±0.1%
Spectral Rolloff ±0.1%		 Force RMS
Extend Matches
 

The effects of polyphony cause confusion in the descriptors. The theme is somewhat apparent when using pitch (D3). Other settings create interesting results, like the beginning of (D1).

  Synthesis Window  
Result Target Corpus Size Skip Shape Matching Criteria Other Options
D1 Schoenberg, Mvt 1
(String Quartet 4)
(4096, 2048)		 Anthony Braxton 2048 1024 Hann Spectral Centroid ±0.1%
Spectral Rolloff ±0.1%		 Force RMS
Extend Matches
D2     1024 512 Hann Spectral Centroid ±0.5%
Spectral Rolloff ±0.5%		 Force RMS
Extend Matches
D3     4096 2048 Hann Pitch ±1% First 200 frames
Force RMS
Extend Matches
 

These are trials made in composing Saxubus.

  Synthesis Window  
Result Target Corpus Size Skip Shape Matching Criteria & options Comments
E1 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(4096, 2048)		 4096 2048 Hann Spectral Centroid ±1% Too active
E2 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(4096, 2048)		 4096 2048 Hann Spectral Centroid ±1%, Random Match Too active
E3 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(4096, 2048)		 4096 2048 Hann Spectral Centroid ±1%, Random Match Too active; not much different from E2
E4 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(4096, 2048)		 4096 2048 Hann RMS ±1%, Spectral Centroid ±1% Less fluid, many gaps, crescendi obvious
E5 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(4096, 2048)		 4096 2048 Hann RMS ±20%, Spectral Centroid ±10% Interesting moments at height of crescendi
E6 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(4096, 2048)		 4096 2048 Hann RMS ±20%, Spectral Rolloff ±10% Similar to E5; in general crescendi too obvious
E7 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(4096, 2048)		 16000 2048 Hann Spectral Centroid ±1% Interesting; synthesis windows too long
E8 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(12000, 6000)		 12000 6000 Hann Spectral Centroid ±1%, Random Match Nice feel; crescendi not obvious
E9 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(12000, 6000)		 12000 6000 Hann Spectral Rolloff ±1%, Random Match Different results from using spectral centroid in E8
E10 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(12000, 6000)		 12000 6000 Hann Spectral Rolloff ±1%, Random Match Slightly different from E9
E11 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(12000, 6000)		 12000 6000 Hann Spectral Centroid ±1%, Random Match Slightly different from E7
E12 Mahler, Solti
(512, 256)		 Anthony Braxton
(12000, 6000)		 12000 12000 Tukey 25% Spectral Centroid ±0.5%, Random Match, frames 1-200 Tukey window has nice effect; too few matches
E13 Mahler, Solti
(512, 256)		 Anthony Braxton
(12000, 6000)		 18000 12000 Tukey 25% Spectral Centroid ±0.5%, Random Match, frames 1-662 Longer window, same overlap as E12; nice effects
E14 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(12000, 6000)		 18000 12000 Tukey 25% Spectral Centroid ±0.05%, Random Match More matches with tighter descriptor threshold and longer target unit; results gestting closer to something musical
E15 Mahler, Solti
(4096, 2048)		 Anthony Braxton
(12000, 6000)		 18000 12000 Tukey 25% Spectral Centroid ±0.01%, Random Match closer ...
E16 Mahler, Solti
(2048, 1024)		 Anthony Braxton
(12000, 6000)		 18000 12000 Tukey 25% Spectral Centroid ±0.05%, Random Match This is it!
E17 Mahler, Solti
(2048, 1024)		 Anthony Braxton
(12000, 6000)		 18000 12000 Tukey 25% Spectral Centroid ±0.05%, Random Match, force RMS Too much like the crescendi
 

These are trials made in composing Lix Tetrax.

  Synthesis Window  
Result Target Corpus Size Skip Shape Matching Criteria & options Comments
F1 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 18000 12000 Tukey 25% Spectral Centroid ±0.05%, Random Match Sometimes lyrical, but too disjointed and pulsing
F2 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 18000 6000 Tukey 25% Spectral Centroid ±0.05%, Random Match Interesting beginning, less regular
F3 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 18000 14000 Tukey 25% Spectral Centroid ±0.05%, Random Match, Force RMS Too regular and quiet
F4 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 18000 10000 Tukey 25% Spectral Centroid ±0.01%, Random Match, Extend Matches I like the effect of extending the matches, but too repetitive
F5 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 18000 2048 Hann Spectral Centroid ±0.01%, Extend Matches Swells too much
F6 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 18000 8000 Tukey 25% Spectral Centroid ±0.01% Some nice moments; the pauses are out of place
F7 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 16000 10000 Hann RMS ±0.1% Beginning is very lyrical, high "A" unusual. Ending is fantastic!
F8 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 16000 10000 Tukey 25% RMS ±0.5% Even better, except piercing high "A" needs to go; but this is the "one"
F9 Mahler, Solti
(5000, 2500)		 J. S. Bach Partita
(4096, 2048)		 16000 10000 Tukey 25% RMS ±0.5% Like 8, but more similar to crescendi
F10 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 16000 10000 Tukey 25% Spectral Centroid ±0.05%, Random Match Not as nice as 8 or 9
F11 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 4096 1000 Hann RMS ±0.5%, Force RMS Different direction
F12 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 4096 1000 Tukey 25% RMS ±0.5%, Force RMS Different direction
F13 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 1000 1000 Hann RMS ±0.5%, Force RMS Pointilistic crescendi
F14 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 100 1000 Tukey 25% RMS ±0.5%, Force RMS Hmmm
F15 Mahler, Solti
(5000, 2500) reversed		 J. S. Bach Partita
(4096, 2048)		 4000 100 Tukey 25% RMS ±0.5%, Force RMS Neat texture!
 
G1

An example of "gnome" speech generated using ACSS.

Compositions

Concatenative Variations of a Passage by Mahler
Target 1 for A capella
Target 2 for A capella

Dedication to George Crumb, American Composer