MEAPsoft Music Showcase

Here are some examples of music made with MEAPsoft, using specific "composer" modules.

If you do something exciting with MEAPsoft, we would love to include it on this page -- please contact us at meapsoft@music.columbia.edu

SimpleSortComposer

• Original: chris_mann.wav
  Processed: chris_mann_glissando.wav
  Start with Chris Mann speech soundfile. Detect events, with sensitivity and density set to high. Extract AvgFrequencySimple. Run SimpleSort, low to high, giving us a speech sound glissando.
  Inspired by Carter Scholz's "Managram".
  [Douglas Repetto]
  
  
• Original: cage.wav
  Processed: cage_crescendo.wav
  Start with the John Cage speech soundfile. Detect events, with sensitivity and density set to high. Extract ChunkPower feature. Run SimpleSort, low to high, and apply fade in/out and crossfade for 10ms. The result is a slow crescendo moving from Cage's gentle speaking voice to the crowd's bursts of laughter.
  Audio source: John Cage Lecture Reading: on Rauschenberg, Duchamp, Johns etc. at L.A. County Museum of Art, 1965, from archive.org.
  [Douglas Repetto]
  
  
• Original: beatit.wav
  Processed: beatit_LB_GF.wav
  Processed: beatit_LF_GB.wav
  Start with Michael Jackson's "Beat It" soundfile. Detect beats. Extract ChunkStartTimes. Run SimpleSort, low to high, with "reverse" checked in the Universal Chunk Operations section. This will give you an output with the global form of the input track running forwards, but the local audio chunks playing backwards. Now do it again, but this time run SimpleSort high to low and uncheck the "reverse" button. This will give you an output with the local chunks playing forward, but the global form of the song going backwards.
  Inspired by Thomas Dimuzio's "Yawriats ot Nevaeh" and "Nevaeh ot Yawriats".
  [Douglas Repetto]
  
  
• Original: Bach - Ach Wie Nichtig, Ach Wie Fluechtig (mp3)
  Processed: Ach wie nichtig-MEAPED.mp3
  
  1. I set it to detect beats, unchecked the half-tempo box, and unchecked "1st event = track start" (to avoid excess silent chunks).
  2. I enabled the Avg Pitch Simple feature extractor, and selected the "Likelihood" Meta feature extractor. I wanted to find which pitches were most likely to occur in the piece.
  3. I selected the Simple Sort composer, and set the individual chunks to be reversed (under Universal Chunk Operations). Then I applied a short fadein/out and crossfade (around 4 on the 0-50 scale) to smooth out the transitions.
  4. I ran the synthesizer!
  [Jeff Snyder]
  
  
ThresholdComposer

• Original: ave_maria.wav
  Processed: ave_maria-one_pitch.wav
  Start with Ave Maria soundfile. Extract events. Extract frequencies. Run ThresholdComposer using a narrow frequency range (115-125Hz) so that we end up with all instances of the few pitches that fall within that range. [Douglas Repetto]
  
  
IntraChunkShuffleComposer

• Original: cheerleader.wav
  Processed: cheerleader_shuffle_1.wav
  Processed: cheerleader_shuffle_0.5.wav
  Start with Dow Jones & the Industrials' "Hold that Cheerleader" sample. Extract beats. Run IntraChunkShuffle with 4 sub chunks. Resulting soundfile maintains the form and forward energy of the original, but with lots of local chaos. Do it all again but extract beats with "cut tempo in half". Still the same form, but less local chaos since the segments are larger. [Douglas Repetto]
  
  
• Original: plum.wav
  Processed: plum_scrambled_wobbly.wav
  Start with the Suzanne Vega "My Favorite Plum" sample. Extract events with sensitivity and density on high. run IntraChunkShuffle composer with 10 sub chunks. Apply fade in/out at 10ms. The result is still clearly the same song, but some strange new qualities. There's a low frequency amplitude modulation (introduced by the tiny chunk size and fade in/out process), and the shuffling of tiny chunks introduces some very strange phrasing in the vocal. [Douglas Repetto]
  
  
• Original: zappa_solo.wav
  Processed: zappa_solo_ICT.wav
  I've found that the IntraChunkShuffle composer does amazing things to guitar solos. Applied to the solo from Frank Zappa's masterpiece, My Guitar Wants To Kill Your Mama. [Ron Weiss]
  
  
RotComposer

• Original: tubby.wav
  Processed: tubby_rotate.wav
  Start with King Tubby's "Forever Dub" sample. Extract beats. Extract arbitrary feature (we'll ignore it). Run RotComposer with 4 beats/measure, 2 beat rotation, rotate right. The result is still relatively cohesive rhythmically, but has a very strange accent structure because we've rotated the accents out of their normal positions. [Douglas Repetto]
  
  
• We can also have some fun with hip hop. I used RotComposer to swap adjacent chunks (i.e. "a b c d" becomes "b a d c").
  Geto Boys - Damn It Feels Good To Be A Gangsta:
  Original: damn_it_feels_good_to_be_a_gangsta.wav
  Processed: feels_damn_it_to_be_a_good_gangsta.wav
  Outkast - Miss Jackson:
  Original: miss_jackson.wav
  Processed: (by ROT; the beat detector didn't work perfectly on this one): miss_jackson_rot.wav
  (and just for fun I reversed each chunk): miss_jackson_rev.wav
  [Ron Weiss]
  
  
MeapaemComposer

• Original: groove.wav
  Processed: grooveevoorg.wav
  Start with Madonna's "Get into the Groove" sample. Extract events, with sensitivity set at a little less than halfway. Extract arbitrary feature (we'll ignore it). Run MeapaemComposer. Hear the non-groovy goodness as each small segment is played forwards and then backwards, maintaining the song's form but disrupting its rhythmic flow entirely. [Douglas Repetto]
  
  
• Original: Jeff Snyder - Getting Dub (mp3)
  Processed: getting-dub-MEAPED.mp3
  
  1. I set it to detect beats, and unchecked the half-tempo box.
  2. I enabled the Length feature extractor (doesn't matter which one).
  3. I selected the MEAPAEM composer
  4. I ran the synthesizer!
  [Jeff Snyder]
  
  
NearestNeighborComposer

• Original: LIB.wav
  Processed: LIB-order-10.wav
  "Let It Be" beat-tracked with MFCC features, then composed by nearest neighbor walk, then selecting only every 10th segment for the synthesizer (by modifying the EDL outisde of MEAPsoft). [Dan Ellis]
  
  
• Original: oops.wav
  Processed: oops_NN.wav
  Start with Britney Spears's "Oops, I Did It Again" sample. Detect events and set sensitivity and density to high. Extract features: AvgMelSpec, AvgSpecCentroid, AvgSpecFlatness, ChunkLength, ChunkPower, SpectralStability, with all weights set to 1.0. Run NearestNeighborComposer. The result groups similar sounding segments of the song together, so that there are several clumps of drum hits, vocalizations, synth noises, etc. [Douglas Repetto]
  
  
• Original: tiny_cities.wav
  Processed: tiny_cities_nn.wav
  Nearest neighbor Modest Mouse gibberish [Ron Weiss]
  
  
• Original: Chords - from "Three Voices", by Morton Feldman (sung by Joan La Barbara) (mp3)
  Processed: Chords-MEAPED.mp3
  
  1. I set it to detect events at relatively high sensitivity and density
  2. I enabled AvgChroma and AvgPitchSimple, and weighted them both equally, to try to catch similarity in both chords and notes, and to seperate out the breath noises.
  3. I selected the nearest neighbor composer, and applied a very short (about 4 ms) crossfade.
  4. I ran the synthesizer!
  [Jeff Snyder]
  
  
NearestNeighborSwapComposer

• Original: Genie.mp3
  Processed: Genie_NNSwap.mp3
  Start with Christina Aguilera's "Genie In a Bottle". Detect beats. Extract AvgMFCC feature. Run NearestNeighborSwapComposer. The result maintains the form of the song but swaps similar segments for one another. [Douglas Repetto]
  
  
HeadBangComposer

• Original: William Byrd - Galliard (23b) (mp3)
  Processed: Galliard-23b-MEAPED.mp3
  Original: Gregory Isaacs - (from slum in Dub) (mp3)
  Processed: gregory-isaacs-MEAPED.mp3
  Original: Jeff Snyder - If that's the Way you feel (mp3)
  Processed: itsalright-jeffsnyder-MEAPED.mp3
  
  1. I set it to detect events at very low sensitivity (to avoid false triggers) and medium density
  2. I enabled the Length feature extractor
  3. I selected the Headbang composer
  4. I ran the synthesizer!
  [Jeff Snyder]
  
  
• Original: flaming.wav
  Processed: flaming_headbang.wav
  Head banging to The Flaming Lips - Yoshimi Battles The Pink Robots Part 2.
  [Ron Weiss]
  
  
VQComposer

• Original: spoon.mp3
  Processed: spoon_mfcc_vq.mp3
  I ran the beat segmenter on Spoon - All The Pretty Girls Go To The City, extracted AvgMFCC features, and used VQComposer to reconstruct the input with 50 codewords and 1 beat per codeword. So the original song is reconstructed using only 50 representative chunks.
  Processed: spoon_chroma_vq.mp3
  Same as above but with AvgChroma features.
  [Ron Weiss]
  
  
HMMComposer

• Original: LIB.wav
  Processed: LIB_hmm.mp3
  I ran the beat segmenter on The Beatles - Let It Be, extracted AvgChroma features and used HMMComposer (25 states, 8 beats per state) to generate a sequence of segments that mimics (to some extent) the high level dynamics of the input. Note how segments from one part of the song seamlessly transition to segments from a completely different part.
  [Ron Weiss]
  
  
More Advanced Examples

• Epitaph [Larry Polansky, Dartmouth College]
  
  
• Hanging Betsy [Roger Dean, Australia]
  Read the description of Hanging Betsy.
  
  
• Original: davedrums-original.mp3
  Original: harpsichord-junk-original.mp3
  Processed: drums-n-harpsihord-MEAPED.mp3
  I took a recording I had of my friend Dave Skogen (of the Youngblood Brass Band and Cougar) improvising on the drums. The recording wasn't to a click track, and it would be a hassle to try to cut together parts that were the same tempo. I wanted to get only the phrases that he played which would fit together rhythmically, so I used the headbang composer. After that, I had a nice groovy and impossible beat (it jumps rapidly between different playing styles, including waving brushes in the air for a whooshing effect) I then took another bit of source material I had lying around - me testing some just intonation chords on harpsichord, and sorted that file by average mel spectrum. Then I pulled the two tracks into a sequencer (digital performer) and chopped up the sorted harpsichord to match the tempo of the new drumbeat.
  The details:
  1. I loaded in the recording of Dave on drums. I set it to detect events.
  2. I enabled the length feature extractor.
  3. I selected the Headbang composer and applied a short fadein/out, but didn't select crossfade (so that the clicks would be removed but there wouldn't be any tempo change from overlapping of chunks.
  4. I went to the "options" pane and set it to save .feat/.edl files. I wanted to be able to look at the length data later.
  5. I ran the synthesizer.
  6. I loaded in the recording of my random harpsichord chords. I set it to detect events.
  7. I enabled the Average Mel Spectrum feature extractor.
  8. I selected the sort composer and sorted from high to low.
  9. I ran the synthesizer.
  10. I opened up the .edl file from the drums in a text editor. Looking at the segment lengths I found that the most common length was .368, and 60/.368 is about 163.043 BPM. This told me the tempo of my new drum track.
  11. I placed the drumtrack into a sequencer and set the tempo to 163.043. Then, I chopped up the harpsichord chords in their sorted order and placed them at will according to the beat grid associated with that tempo. Voila!
  [Jeff Snyder]