Techniques

 for  Polytemporal  Composition  

 
Christopher  Dobrian  
Department  of  Music,  University  of  California,  Irvine,  USA  
[email protected]  
http://music.arts.uci.edu/dobrian  
 
Proceedings  of  Korean  Electro-­‐Acoustic  Music  Society's  2012  Annual  Conference  (KEAMSAC2012)  
Seoul,  Korea,  26-­‐27  October  2012  
 

Digital   computers   offer   unprecedented   flexibility   and   accuracy   for   the   composition   and   performance   of  polytemporal   music   involving   multiple   time  
streams:  polyrhythms,  metric  modulations,  multiple  tempi,  and  precisely  calculated  independent  continuous  tempo  changes  (accelerandi  and  de-­‐
celerandi).  This  article  addresses  some  important  musical  and  technical  issues  encountered  in  music  with  multiple  simultaneous  tempi,  including  
music   with   independently   fluctuating   tempi.   The   article   includes   a   brief   summary   of   some   important   precedents   in   pre-­‐computerized   music,   a   look  
at   some   implementations   of   polytemporality   in   computer   music,   and   some   suggested   techniques   for   designing   and   manipulating   multiple   time  
streams  in  computer  music  composition.  
 

October  27,  2012  is  the  centenary  of  the  birth  of  Ameri-­‐ amount   of   clock   time,   commonly   beats   per   minute  
can-­‐Mexican   composer   Conlon   Nancarrow.   Nancarrow’s   (BPM)—known   as   the   tempo.   The   onset   time   of   each  
body   of   compositions   for   player   piano   is   the   most   con-­‐ sonic  event  is  the  moment  when  it  is  perceived  to  begin,  
centrated   exploration   of   polytemporal   music   —   music   and  its  duration  is  how  long  it  lasts;  either  may  be  meas-­‐
ured   in   clock   time   and/or   musical   time.   Any   instant   in  
that  proceeds  with  multiple  independent  beat  rates.  His  
time   can   be   ascribed   a   unique   numerical   value   in   either  
painstaking   work   with   mechanized   pianos   was   an   im-­‐ clock   time   or   musical   time   relative   to   an   established   0  
portant  precursor  to  computerized  musical  performance   point;  such  a  representation  of  a  single  moment  is  called  
with  Music  N  scores  and  MIDI  sequences,  and  it  demon-­‐ a  timepoint.  
strated   the   conceptual   and   practical   potential   of   poly-­‐
temporal   music.   Computers   now   offer   unprecedented   Some  History  
flexibility   and   accuracy   for   the   composition   and   perfor-­‐
Perhaps   the   oldest   and   best-­‐known   compositional   tech-­‐
mance   of   music   involving   adventurous   exploration   of  
nique   that   evokes   multiple   time   streams   is   the   classic  
multiple   time   streams—polyrhythms,   metric   modula-­‐ technique  of  hemiola,  originally  used  to  describe  a  situa-­‐
tions,  multiple  tempi,  and  precisely  calculated  independ-­‐ tion   in   which   three   equally-­‐spaced   notes   in   one   voice  
ent   continuous   tempo   changes   (accelerandi   and   decel-­‐ occur   in   the   same   time   as   two   equally   spaced   events   in  
erandi)—yet   the   methodology   of   composing   with   these   another  voice  [Figure  1].  
techniques  has  not  been  very  extensively  documented  or  
theorized.   In   this   article   I   summarize   some   obvious   musi-­‐
cal  implications  of  polytemporality,  and  I  will  show  some  
basic  examples  of  its  implementation  in  computer  music.      
Figure  1.  Classic  hemiola,  implying  two  tempi  with  a  ratio  of  3:2.  

Musical  background   In   actual   usage   the   3:2   ratio   may   be   expressed   not   only  
by  individual  notes  but  by  any  grouping  of  notes  that  —  
by  its  melodic  contour  or  its  harmonic  implications  —  can  
Some  Terminology   be  considered  to  express  a  unified  event  with  a  duration  
In  this  article  I  use  the  word  time  as  a  shorthand  to  refer   3/2  or  2/3  as  long  as  another  event  [Figure  2].  
to   ways   of   measuring   sequences   of   events.   The   most  
objective   type   of   time   measurement,   the   one   most   fre-­‐
quent  in  everyday  discourse  and  scientific  measurement,  
is  the  use  of  a  regular,  reliable  and  agreed-­‐upon  clock;  I  
refer   to   this   as   clock   time.   Music   is   not   commonly   orga-­‐
nized  by  direct  reference  to  clock  time,  but  instead  uses    
a  different  type  of  time  measurement  known  as  musical   Figure  2.  Melody  vs.  harmonic  rhythm  establishes  a  ratio  of  3:2.  
time,   which   relies   on   an   agreed-­‐upon   more-­‐or-­‐less   con-­‐
stant   unit   of   time   known   as   a   beat.   The   relationship   of  
musical   time   to   clock   time   is   based   on   the   rate   of   the   Although  originally  applied  only  to  this  simplest  of  prime  
beat—the   number   of   beats   that   occur   in   a   certain   ratios,  the  term  came  to  be  used  more  generally  to  refer  
 KEAMSAC2012 - Korean Electro-Acoustic Music Society's 2012 Annual Conference - Proceedings

to  any  musical  situation  that  suggests  two  different  met-­‐ 5   share   no   prime   factors,   the   simultaneity   of   triplet  
ric  interpretations  [Figure  3].     eighth   notes,   sixteenth   notes,   and   quintuplet   sixteenth  
notes   yields   11   unique   attack   points   within   each   beat.   By  
writing   syncopated   rhythms   with   those   beat   divisions   in  
dense  polyphony,  he  achieved  clouds  of  notes  that  are  so  
  complicated   they   seem   beatless.   He   thus   used   poly-­‐
Figure  3.  Pitch  contour  vs.  dynamic  accent  implies  a  ratio  of  3:4.   rhythms   not   so   much   to   create   a   sense   of   polytemporali-­‐
ty,  but  rather  in  an  attempt  to  obscure  traditional  beat-­‐
Ratios   with   larger   numbers,   especially   when   occurring   based  musical  time.  
within   a   single   measure   or   a   single   beat   are   more   com-­‐
Truly   polytemporal   instrumental   music   in   which   the  
monly   referred   to   as   cross-­‐rhythms   or   polyrhythms   [Fig-­‐
players  must  maintain  separate  tempi  with  rigorous  pre-­‐
ure  4].  
cision  in  order  to  arrive  in  synchrony  at  specific  moments  
is  difficult  to  perform  because  humans  have  an  imperfect  
sense   of   musical   time.   When   players   are   working   togeth-­‐
er  in  a  common  musical  time  stream,  they  maintain  syn-­‐
chrony   by   constantly   and   flexibly   adjusting   their   tempo  
as   needed   to   maintain   a   group   consensus.   However,   if  
individuals  or  subgroups  of  an  ensemble  are  called  upon  
to   maintain   a   separate   and   perfectly   exact   tempo   while  
  others   do   likewise   in   a   different   time   stream,   it   can   be  
Figure  4.  Polyrhythms  5:4,  3:4,  and  5:8.   difficult   to   adjust   to   small   fluctuations   in   other   time  
streams.   One   solution   is   for   the   musicians   to   listen   to  
Nancarrow   employed   hemiola   extensively   in   his   1940   pre-­‐recorded  click  tracks  to  ensure  accuracy  within  each  
instrumental   Trio   for   clarinet,   bassoon,   and   piano   time   stream,   or   to   have   multiple   conductors   charged  
(Nancarrow,   1991),   and   combined   hemiolas   and   poly-­‐ with  coordinating  the  different  tempi.  A  notable  example  
rhythms   in   his   1945   String   Quartet   No.   1   (Nancarrow,   of   this   approach   of   using   multiple   leaders   is   Karlheinz  
1986).  He  continued  to  use  these  techniques  in  his  stud-­‐ Stockhausen’s   1957   composition   Gruppen   (Stockhausen,  
ies   for   player   piano,   predominantly   focusing   on   prolation   1963)  for  three  orchestras  and  three  conductors.      
canons   in   which   the   voices   proceed   at   entirely   separate  
tempi.   Terry  Riley’s  1964  composition  In  C  (Riley,  1989)  for  any  
number  of  players  (he  suggests  about  35)  establishes  an  
The   prolation   canon,   also   known   as   mensuration   canon,   unflinchingly  regular  base  tempo  with  constant  repeated  
in   which   a   melody   is   performed   in   imitation   but   with   eighth  notes  in  the  highest  register.  Against  that  constant  
augmented  or  diminished  rhythmic  values,  dates  back  at   rhythm,   other   instrumentalists   repeat   musical   cells   of  
least   to   Johannes   Ockeghem’s   Missa   prolationum   in   the   different   lengths,   almost   all   of   which   are   some   whole  
Renaissance   (Plamenac,   1966).   As   a   compositional   tech-­‐ number  multiple  of  the  eighth  note  duration.  Each  play-­‐
nique,  prolation  canon  is  related  to  hemiola,  because  the   er’s   musical   time   is   synchronized   to   the   same   “master  
stretched   or   compressed   versions   of   the   melody   imply   clock”  of  the  repeated  eighth  notes,  but  s/he  is  at  liberty  
different  meters  and/or  rates  of  performance.   to   repeat   each   cell   any   number   of   times.   The   result   is   a  
music  that  is  very  clear  in  its  global  tempo,  yet  consists  of  
Nancarrow   frequently   composed   mensuration   canons   in  
a  constantly  changing,  massively  polyphonic  conglomera-­‐
which   the   rates   of   the   different   voices   are   in   harmonic  
tion  of  hemiolae.  With  a  single  page  of  notation  and  this  
ratios.   Some   of   his   studies   for   player   piano   exhibit   rela-­‐
simple   use   of   indeterminacy—allowing   each   player   the  
tively  simple  ratios  such  as  3:4  in  Study  No.  15,  and  4:5  in  
liberty   to   choose   the   number   of   repetitions   for   each  
Study  No.  14.  As  his  use  of  the  technique  grew  more  ad-­‐
cell—the  composer  allows  for  an  infinite  number  of  pos-­‐
vanced   and   experimental,   he   employed   higher   ratios  
sible   realizations,   each   of   which   is   nevertheless   still   iden-­‐
such   as   14:15:16   in   Study   No.   24,   17:18:19:20   in   Study  
tifiable  as  the  same  composition.  
No.  36,  21:24:25  in  Study  No.  31,  and  even  60:61  in  Study  
No.  48.  His  most  adventurous  piano  piece,  Study  No.  37  is   Steve   Reich,   in   his   1960s   tape   compositions   It’s   Gonna  
a   canon   in   twelve   voices,   each   at   a   different   tempo   Rain   and   Come   Out,   took   musical   advantage   of   the   effect  
(Gann,   1995).   In   such   a   complex   polyphonic   work,   the   of   two   nearly-­‐identical-­‐length   cycles   gradually   going   in-­‐
effect   for   the   listener   is   no   longer   so   much   the   percep-­‐ creasingly   out   of   synchronization.   Initially   the   cycles  
tion   of   polyrhythm   as   of   the   gestural   or   textural   effect   (tape   loops)   are   in   seemingly   perfect   unison,   but   one   of  
produced   by   the   relationships   of   simultaneous   slightly   two  cycles  is  in  fact  ever-­‐so-­‐slightly  longer  than  the  oth-­‐
different  tempi.   er,  such  that  the  temporal  relationship  between  the  two  
Iannis  Xenakis  (1955)  observed  the  potential  for  textural   unison  sounds  gradually  shifts.  The  effect  is  at  first  one  of  
perception   of   polyrhythms   in   dense   polyphony,   and   he   comb   filtering,   as   the   component   frequencies   of   the  
employed   simple   polyrhythms   of   3:4:5   in   his   1956   com-­‐ sounds  themselves  constructively  and  destructively  inter-­‐
position   Pithoprakta   (Xenakis,   1967)   to   obfuscate   the   fere   based   on   the   time   of   the   delay.   As   the   delay   be-­‐
listener’s   sense   of   a   beat.   Because   the   numbers   3,   4,   and   tween   the   two   sounds   increases,   the   effect   becomes   a  

2
 KEAMSAC2012 - Korean Electro-Acoustic Music Society's 2012 Annual Conference - Proceedings

timbral   blurring,   then   a   short   but   discrete   “slapback”   separate   harmonically-­‐related   tempi.   The   program   was  
echo,  and  eventually  a  new  composite  rhythm.   inspired  by  the  ideas  presented  by  John  Whitney  (1980)  
relating   musical   and   visual   harmonicity.   As   one   might  
In  the  case  of  these  two  pieces,  the  compositional  tech-­‐ expect   with   so   many   voices,   the   resulting   sounds   are  
nique   was   derived   from   Reich’s   experimentation   with   often   textural   as   much   as   they   are   explicitly   rhythmic,  
tape  recorders,  and  was  the  result  of  the  practical  impos-­‐ but   because   of   the   strictly   harmonic   relationships   be-­‐
sibility   of   perfectly   synchronizing   tape   loops   on   two   dif-­‐ tween   tempi   there   are   frequent   convergence   points   of  
ferent  players.  Gradually  increasing  asynchrony,  as  found   varying  degrees  of  unanimity,  providing  for  sonic  variety  
in  these  compositions,  can  be  caused  by  a  slight  discrep-­‐ and  long-­‐term  rhythmic  formal  structure.  
ancy  in  the  length  of  the  loops,  or  by  a  discrepancy  in  the  
speed  of  playback.  In  either  case,  the  two  sounds  have  a   In   practice,   the   time   streams   in   polytemporal   music   are  
slightly   different   periodicity   of   repetition.   A   discrepancy   often   not   purely   independent,   because   the   different  
in   playback   speed   has   the   side   effect   of   creating   a   dis-­‐ tempi   are   chosen   in   such   a   way   as   to   relate   to   some  
crepancy  in  tuning  as  well  as  in  periodicity.  Careful  analy-­‐ slower   global   tempo   at   a   larger   structural   level,   or   to  
sis   of   Come   Out   (Reich,   1987)   indicates   that   the   loops   create   specific   convergences   between   disparate   time  
were   playing   at   very   slightly   different   speeds.   It   is   quite   streams.   In   the   music   of   Nancarrow,   the   musical   tempi  
easy   to   play   the   same   sound   at   two   subtly   different   rates   are   almost   always   chosen   in   specific   harmonic   ratios,  
in  a  program  such  as  Max  [Figure  5].   even  if  those  combinations  may  be  extremely  complex  or  
may   consist   of   ratio   relationships   that   are   very   obscure.  
These   experiences   led   Reich   to   compose   similar   gradual   Even  his  Study  No.  21—known  as  “Canon  X”  because  one  
tempo   shifts   into   some   of   his   instrumental   composi-­‐ of  its  two  voices  is  constantly  accelerating  while  the  oth-­‐
tions—notably   Piano   Phase   (1967)   and   Drumming   (1971)   er  is  constantly  decelerating,  and  the  pitch  range  of  one  
from   the   time   period   immediately   following   those   tape   voice  descends  while  the  other  ascends—is  carefully  de-­‐
works.  The  compositions  require  the  performers  to,  in  a   signed  so  that  the  voices  have  a  known  point  of  conver-­‐
sense,   emulate   the   behavior   of   the   two   tape   recorders   gence   in   the   middle   and   the   two   melodies   occupy   the  
playing  at  slightly  different  tempi;  one  performer  speeds   desired  amount  of  time  for  the  particular  musical  form.  It  
up   imperceptibly   until   s/he   is   exactly   one   rhythmic   unit   is   relatively   rare—at   least   in   the   examples   reviewed  
(e.g.,  one  sixteenth  note)  ahead  of  the  other  performer.   here—that   the   tempi   in   a   piece   of   music   are   purely   inde-­‐
Huijae   Chung,   in   his   composition   titled   Multi   Tempi   21,   pendent  with  no  shared  reference.  
employed   temporal   canon   at   the   unison   with   tempo   ra-­‐ One   might   argue   that   simultaneous   independent   time  
tio   close   to   1   in   a   highly   polyphonic   texture.   The   music   streams   are   not   inherently   musically   interesting   if   they  
consists   of   an   ascending   eight-­‐note   piano   melody   re-­‐ are  totally  unrelated  (Nemire,  2012).  The  simple  fact  that  
peated   over   and   over,   played   at   21   slightly   different   two  things  coexist  simultaneously  and  appear  to  be  obliv-­‐
tempi  evenly  spaced  between  120  BPM  and  110  BPM.  All   ious   of   each   other   can   be   expressive   in   its   own   right;   tru-­‐
tempi   thus   reside   within   a   1.091:1   ratio.   The   recording   ly  free  counterpoint  might  serve  as  an  expression  of  lib-­‐
was   made   with   digitally   sampled   individual   piano   notes   erty  and  independence,  for  example.  And  such  complete  
triggered  by  MIDI;  the  MIDI  data  was  played  at  a  differ-­‐ independence  can  lead  to  interesting  coincidental  simul-­‐
ent  tempo  for  each  of  twenty-­‐one  different  channels  of  a   taneities   and   juxtapositions.   But   polytemporal   music  
multi-­‐track   audio   recording,   thus   triggering   the   exact   with  no  discernible  points  of  convergence  or  divergence  
same   sounds   at   different   rates   without   causing   any   does   not   necessarily   require   any   particular   composition  
mistuning  of  the  sounds  themselves.   or  control.  The  point  of  using  a  computer  to  realize  poly-­‐
Because  of  the  number  of  voices  and  the  range  of  differ-­‐ temporal  music  is  to  leverage  the  computer’s  powers  of  
ent   tempi,   timbral   blurring   begins   with   the   very   first   calculation   and   its   precision   of   performance.   The   com-­‐
note,   discrete   echoes   can   be   discerned   within   seconds,   puter  can  potentially  aid  the  composer  in  realizing  poly-­‐
and   clear   polytemporality   and   resulting   composite   temporal  music  with  specific  tempo  relationships  and/or  
rhythms   are   heard   soon   thereafter.   Due   to   the   large   specific  desired  points  of  convergence.  
number   of   voices   in   very   high   rational   tempo   relation-­‐
ships   such   as   240:239,   239:238,   etc.,   the   music   passes  
through   a   wide   variety   of   constantly   changing   textures   Computer  applications  
and   rhythms,   and   the   music   never   truly   repeats   within  
the   seven   and   a   half   minutes   that   Chung   lets   the   process   Commercial  software  
run  its  course.  There  is  an  initial  divergence  point  at  the  
Despite   the   applicability   of   computers   for   exploring   and  
beginning,  but  no  true  convergence  point  of  the  different  
cycles.   There   are,   however,   frequently   moments   in   the   realizing  polytemporal  music,  very  few  commercial  com-­‐
composition   when   enough   voices   briefly   achieve   near-­‐ puter   music   applications   provide   for   the   specification   of  
time   in   multiple,   different,   simultaneous   time   streams.  
synchrony  to  give  the  impression  of  concerted  behavior.  
This  is  understandable,  since  the  vast  majority  of  all  mu-­‐
The   Whitney   Music   Box   web   application   (Bumgardner,   sic   takes   place   in—and   is   largely   conceived   as—a   single  
2006)  demonstrates  repeated  tones  played  (and  visually   time   stream.  Because   of   the   scant   user   demand   for   poly-­‐
displayed)  at  48—or  in  some  instances  as  many  as  120— temporal  control,  it  is  presumably  not  commercially  via-­‐

3
 KEAMSAC2012 - Korean Electro-Acoustic Music Society's 2012 Annual Conference - Proceedings

ble   for   music   software   companies   to   devote   valuable    

programmer   time   to   implementing   multiple   time  
streams,   in   the   form   of   multiple   transport   mechanisms,  
as   a   function   of   a   common   music   application   such   as   a  
DAW  (e.g.,  Pro  Tools).  One  can,  of  course,  perform  digital  
audio   processing   operations   in   nonreal   time   such   as   au-­‐
dio   rate   change,   time   compressions   or   expansion,   trim-­‐
ming,  and  looping,  and  these  operations  have  been  made  
ever   easier   in   many   audio   and   music   programs.   In   that  
sort   of   nonrealtime   processing,   the   computer   is   helping  
with   sound   modifying   operations   but   is   not   particularly  
helping   with   the   calculations   necessary   to   implement  
those  operations  in  a  composition.  
Within  the  paradigm  of  a  single  master  transport  control-­‐
ling  the  passage  of  time,  the  Live  application  by  Ableton  
performs   high-­‐quality   time   compression/expansion   in  
real  time  to  adjust  the  duration  of  live  recorded  or  prere-­‐
corded  sounds  to  a  certain  number  of  measures  in  a  giv-­‐
en   tempo.   Those   time-­‐adjusted   sounds   can   then   be   ex-­‐
ported  to  new  audio  files  for  use  in  polytemporal  compo-­‐
sition.   Live   also   provides   the   capability   to   play   simultane-­‐
ous   loops   of   different   lengths,   thus   achieving   hemiola-­‐
like  effects  with  audio  loops.    
By   and   large,   though,   exploration   of   polytemporality   re-­‐ Figure  5.  Four  loops  with  lengths  in  the  ratios  13:14:15:16.  
mains   predominantly   the   province   of   computer   music  
programming   environments   such   as   Csound,   Max,   Pd,   Flexible,   related   tempi.   The   capability   to   to   implement  
and   SuperCollider.   Even   these   specialized   and   relatively   polytemporality   has   in   fact   always   existed   in   Max.   The  
esoteric   programming   environments   are,   quite   under-­‐ canonical   timing   object   metro   runs   with   as   much   clock  
standably,   mostly   modeled   on   a   single   unifying   sense   of   time  precision  as  is  available  in  the  operating  system,  so  
time  related  to,  or  based  directly  on,  clock  time.   with  multiple  metros  one  can  trigger  events  at  any  com-­‐
bination   of   rates,   and   polyrhythms   can   be   enacted   with  
Recent   development,   spearheaded   by   David   Zicarelli,   in   some   simple   arithmetic   calculations.   The   tempo   object  
the   Max   programming   environment   (Puckette,   1990),   makes   those   tasks   even   simpler   by   allowing   the   quarter  
facilitates   polytemporal   music.   I   will   briefly   present   the   note   tempo   to   be   specified   in   BPM   and   by   allowing   the  
conceptualization  of  multiple  time  streams  that  underlies  
whole  note  to  be  divided  by  any  ratio  of  whole  numbers  
the   implementation   of   tempo-­‐relative   musical   timing   in   up  to  96.  
Max,   and   I   will   provide   specific   examples   in   Max   showing  
simple  applications  of  compositional  techniques  for  hem-­‐ Using  multiple  independent  transport  objects  to  control  
iola,  polyrhythms,  polytemporality,  and  the  use  of  trans-­‐ sets  of  timing  objects,  one  can  run  entire  timed  process-­‐
fer  functions  for  predictable  timepoint  warping  in  multi-­‐ es   at   independent   rates.   The   transports   can   progress  
ple  time  streams.     completely  independently  or  they  can  be  controlled  by  a  
common   global   tempo.   For   example,   with   three   differ-­‐
Simple  implementations  in  Max   ently   named   transport   objects,   one   can   easily   play   the  
same   composition   at   three   different   tempi,   readily   pro-­‐
Hemiola  with  audio  loops.  The  ability  to  build  constantly   ducing   a   tempo   canon.   If   the   three   transports   are   run-­‐
changing  textures  with  loops  of  different  lengths  exists  in   ning  at  related  tempi,  a  common  global  source  can  modi-­‐
many  commercial  programs.  In  Max  one  can  write  a  pro-­‐ fy   all   three   tempi   at   the   same   time,   again   using   simple  
gram   that   permits   the   construction   and   modification   of   arithmetic  to  determine  the  correct  tempo  for  each  one  
such  textures  on  the  fly.  For  example,  one  can  play  mul-­‐ [Figure  6).  
tiple   simultaneous   instances   of   the   same   sound,   yet   loop  
a  different  duration  segment  of  that  sound  in  each  voice   Rate   change   in   audio   playback.   In   the   spirit   of   Reich’s  
of   a   polyphonic   texture.   Segmenting   a   few   seconds   of   Come  Out,  one  can  achieve  interesting  sonic  effects  such  
sound,  which  may  well  contain  its  own  internal  rhythms,   as   phasing   and   chorusing   by   playing   the   same   sound   at  
into   segments   with   durations   in   a   ratio   13:14:15:16   is   a   two   or   more   subtly   different   rates.   The   slight   mistuning  
simple  way  to  make  an  evolving  rhythmic  texture  out  of  a   caused   by   the   rate   changes,   combined   with   the   slight  
single  sound  [Figure  5].   delay   that   results,   causes   interference   between   the   two  
sounds,   resulting   in   filtering   and   blurring   effects   as   well  
as   slapback   echos   as   the   sounds   go   further   out   of   syn-­‐
chonization.  This  is  a  computerized  example  of  temporal  
canon   at   the   unison,   using   a   ratio   of   two   tempi—two  

4
 KEAMSAC2012 - Korean Electro-Acoustic Music Society's 2012 Annual Conference - Proceedings

audio  playback  rates—that  is  extremely  close  to  1  [Figure   tempo   (rate   in   units   per   minute)   and   are   often   referred  
7].   to  as  “beats”,  just  as  in  common  practice  Western  music  
notation.   Musical   timings   may   be   expressed   using   deci-­‐
mals   with   a   fractional   part   (e.g.,   1.875   units)   to   specify  
time   with   an   arbitrary   degree   of   precision.   The   musical  
time  unit  is  converted  to  clock  time  with  the  simple  for-­‐
mula  60./M,  where  M  is  the  tempo  specified  in  units  per  
minute.     By   default   the   tempo   is   60   BPM,   causing   the  
units  to  correspond  exactly  with  seconds  in  clock  time.  If  
the   tempo   is   explicitly   specified   other   than   60,   musical  
time   can   proceed   independently   of   clock   time,   with   units  
being  converted  to  their  corresponding  clock  time  value.  
It’s   significant   that   in   Csound   the   tempo   markings   of   a  
section   of   music   are   all   provided   in   a   single   statement  
that   is   pre-­‐processed   before   the   sound   itself   is   comput-­‐
ed.  The  statement  can  specify  as  many  tempo  changes  as  
desired,  each  of  which  is  tagged  with  its  own  start  time.  
These   start   times   must   be   presented   in   non-­‐decreasing  
order.   The   start   time   of   each   tempo   indication   is   speci-­‐
fied  in  musical  time  units,  the  actual  clock  time  of  which  
is   dependent   on   whatever   tempo   indications   occurred  
earlier.   Tempo   can   be   changed   discretely   (i.e.,   instanta-­‐
neously),   or   gradually   by   linear   interpolation   between  
two  tempo  specifications.  If  a  tempo  indication  is  the  last  
one  of  a  given  section,  or  if  it  is  followed  by  subsequent  
indication   of   the   same   tempo,   the   musical   tempo   will  
remain   constant.   If   a   tempo   M1   is   followed   by   a   different  
tempo   M2,   the   tempo   changes   linearly,   beat-­‐by-­‐beat,  
from   M1   to   M2.   (Verco,   1992)   Because   the   start   time   of  
each   tempo   change   is   specified   in   musical   time   rather  
than  clock  time,  each  tempo  indicator  can  be  thought  of  
as  a  point  in  a  tempo  graph,  a  graph  which  is  itself  non-­‐
linear  (with  respect  to  clock  time)  on  its  time  axis  due  to  
the  duration-­‐changing  effect  of  each  tempo.  The  nonlin-­‐
  ear   x   axis   of   such   a   graph   makes   it   fairly   difficult   for   a  
Figure  6.  Tempo  canon  with  the  ratio  3:4:5.   human  composer  to  predict  the  precise  clock  time  dura-­‐
tion   of   a   complete   passage   that   contains   many   tempo  
changes,  accelerandi,  etc.,  but  the  computer  calculates  it  
with   ease.   The   composer   learns   the   clock   time   result   ret-­‐
rospectively  by  consulting  the  duration  of  the  computed  
sound  file.  
Tempo-­‐relative   timing.   In   realtime   programming   envi-­‐
  ronment  such  as  Max,  the  problem  of  accurately  realizing  
Figure  7.  A  sound  looped  in  two  channels  with  a  0.15%  rate  discrepancy.   constantly   changing   tempi,   and   of   predicting   conver-­‐
gences   of   multiple   flexible   time   streams   is   all   the   more  
challenging.   Events   in   Max   are   governed   by   an   always-­‐
Continuously  variable  tempo  relationships  
active   scheduler   based   on   an   inflexible   objective   clock.  
One   asset   of   the   computer   is   its   ability   to   calculate   and   Immediate   or   future   events   may   be   scheduled—either  
enact   discrete   tempo   changes   instantaneously   and   pre-­‐ explicitly  by  the  user  or  automatically  by  a  timing  object  
cisely.   Getting   the   computer   to   play   with   tempo   rubato   such  as  metro—at  any  time.  The  events  are  usually  pro-­‐
(flexible   tempo),   or   to   follow   a   desired   curvature   of   ac-­‐ duced  in  real  time  or  scheduled  on  the  fly,  not  foreseen  
celeration  or  deceleration,  is  a  bit  more  complicated  but   as  they  would  be  in  a  pre-­‐composed  score.  Thus,  the  im-­‐
still  feasible.   plementation   of   tempo-­‐relative   musical   timing   into   the  
existing  Max  timing  objects,  based  on  the  transport  and  
The   Csound   computer   music   programming   language   its  related  objects,  presented  some  special  challenges  to  
(Vercoe,   2008)   provides   a   straightforward   relationship   the  programmers  of  Max  itself.  
between   objective   clock   time   and   the   musical   time   de-­‐
scribed   in   its   score   files.   Starting   times   and   durations   of   The  underlying  Max  scheduler  must  remain  inflexible  and  
events  are  specified  in  musical  time  units  that  are  inher-­‐ reliable  because  its  solidity  is  the  foundation  upon  which  
ently   undefined   but   that   are   related   to   clock   time   by   a   all   timing   in   Max   is   based.   The   Max   scheduler   is   a   time-­‐

5
 KEAMSAC2012 - Korean Electro-Acoustic Music Society's 2012 Annual Conference - Proceedings

sorted   list   of   all   the   things   Max   has   to   do.   The   head   of   As  long  as  the  transfer  function  begins  at  point  0,  0  and  
the  list—i.e.,  the  event  scheduled  for  the  most  proximate   ends  at  point  1,  1,  the  two  time  streams  will  be  synchro-­‐
future   moment—is   dealt   with   next.   In   addition   to   the   nized  at  those  points.  Additional  convergence  points  will  
main   scheduler,   there   can   be   tempo-­‐relative   timed   occur   whenever   the   transfer   function   crosses   this   unity  
events   managed   by   the   transport   system.   In   Max   there   is   line.  
one  global  transport  for  managing  tempo-­‐relative  timing,  
In  a  realtime  situation  the  slope  of  the  transfer  function  
but   one   can   also   program   any   number   of   additional   in-­‐
dependent  transports,  each  identified  by  name,  allowing   must  always  be  between  0  and  1;  the  x  and  y  values  both  
must   always   be   increasing,   meaning   that   time   is   pro-­‐
for  a  potential  multitude  of  independent  time  streams.  
gressing   in   both   time   streams.   In   a   nonrealtime   evalua-­‐
Tempo-­‐relative  timepoint  events  are  stored  in  a  separate   tion,  such  as  evaluating  a  precomposed  score  before  at-­‐
time-­‐sorted  list  managed  by  the  relevant  transport.  If  the   tempting   to   perform   it,   one   could   employ   a   model   in  
tempo   of   that   transport   changes,   Max   must   change   the   which   musical   time   (score   time)   could   actually   move  
timing   of   the   events   in   that   list,   taking   into   account   the   backward,  such  that  events  that  occur  later  sequentially  
time  that  has  elapsed  since  the  events  were  first  posted   in   one   time   stream   occur   in   reverse   order   in   the   other  
to   the   list.   These   transport-­‐related   lists   of   events   are   all   time   stream;   however,   in   a   realtime   situation   where   we  
separate   from   the   main   Max   scheduler.   The   main   Max   are   determining   all   time   points   sequentially,   this   is   not  
scheduler  therefore  needs  only  one  scheduled  event  for   practical.   Therefore,   when   implementing   this   idea   in  
each  transport,  namely  a  time  tag  telling  it  when  it  next   Max,  we  will  confine  ourselves  to  transfer  functions  that  
needs  to  consult  the  transport’s  list  (Zicarelli,  2012).   are  always  increasing  on  both  axes.  
Timepoint  stretching.   The   principle   that   underlies   tempo   To   understand   how   this   timepoint   lookup   method   will  
rubato   is   that   the   tempo   of   the   different   voices   of   a   work   with   a   nonlinear   transfer   function,   let’s   consider   a  
composition   might   flex   subtly,   with   the   performer(s)   lookup  table  in  which  the  function  is  made  up  of  straight  
nevertheless   remaining   aware   of   the   other   voices   so   that   line  segments  [Figure  9].  
an   important   global   tempo   is   maintained.   If   one   voice  
“steals  time”  by  speeding  up  or  slowing  down  relative  to  
another   voice,   either   the   other   voices   must   do   likewise  
to  stay  synchronized,  or  that  stolen  time  must  be  “given  
back”   by   flexing   the   tempo   commensurately   in   the   oppo-­‐
site   direction.   The   assumption   of   synchrony   is   that   all  
tempi  are  constantly  in  a  1:1  relationship,  but  if  that  ratio  
changes  slightly  for  a  time,  it  must  be  changed  in  an  op-­‐
posite  manner  for  a  comparable  amount  of  clock  time  in  
order  for  the  two  tempi  to  resynchronize.    
Figure  9.  Breakpoint  line  segments  for  a  timepoint  transfer  function.  
From   this   standpoint,   we   can   compare   all   timepoints   in  
two   different   time   streams   using   a   transfer   function,   also  
referred  to  as  a  lookup  table.  That  is,  for  each  event  that   In   this   example,   the   function   begins   at   point   0,0,   pro-­‐
occurs  in  one  time  stream  (an  incoming  x  value),  we  can   ceeds   linearly   to   0.25,   0.125,   converges   with   unity   at  
use   a   transfer   function   to   look   up   the   corresponding   point   0.5,   0.5,   goes   to   point   0.6,   0.9,   and   resolves   at  
timepoint   (y   value)   in   another   time   stream.   This   lookup   point   1,1.   As   time   stream   x   proceeds   from   time   0   to   time  
table   for   timepoints   has   also   been   called   a   time   map   0.25,  time  stream  y  proceeds  from  0  to  0.125.  The  slope  
(Jaffe,  1985).   of  this  line  segment  is  1:2  so  all  timepoints  in  stream  y  up  
to  that  point  will  have  1/2  the  value  of  the  corresponding  
A  comparison  of  all  timepoints  in  two  perfectly  synchro-­‐ timepoints   in   stream   x.   One   could   also   think   of   this   as  
nized   time   streams   is   a   linear   (unity)   transfer   function   meaning   that   the   musical   tempo   of   stream   y   is   2   times  
(Figure  8].  With  a  linear  transfer  function  all  the  musical   that  of  stream  x,  but  for  this  discussion  we  will  be  focus-­‐
time  points  have  the  same  clock  time.   ing  on  the  timepoints  rather  than  the  tempo,  since  that  is  
what   is   depicted   by   the   function.   As   time   stream   x   pro-­‐
gresses   from   0.25   to   0.5   (a   difference   of   0.25),   time  
stream   y   progresses   from   0.125   to   0.5   (a   difference   of  
0.375).   Between   those   two   points,   the   slope   is   3:2,   so  
during  that  span  of  musical  time  the  timepoints  in  stream  
y   are   spaced   with   1.5   times   the   distance   as   those   of  
stream  x  (i.e.,  musical  time  elapses  2/3  as  fast  in  stream  y  
as   it   does   in   stream   x).   Between   times   0.5   and   0.6   in  
stream   x,   time   progresses   from   0.5   to   0.9   in   stream   y  
  (yielding  a  slope  of  4:1),  and  in  the  remaining  time  from  
Figure  8.  A  linear  function  with  a  slope  of  1:1  causes  no  distortion.   0.6   to   1   in   stream   x,   stream   y   progresses   from   0.9   to   1   (a  
slope   of   1:4).   This   means   that   if   the   musical   time   of  
stream   x   were,   for   example,   constantly   60   BPM   over   a  

6
 KEAMSAC2012 - Korean Electro-Acoustic Music Society's 2012 Annual Conference - Proceedings

time   span   of   60   seconds,   stream   y   would   at   first   have   a   own  musical  time  with  a  metro,  it  can  look  up  the  time  of  
tempo  of  120  BPM  for  7.5  seconds,  then  40  BPM  for  22.5   the   next   scheduled   event   in   the   transfer   function   to   set  
seconds,  then  15  BPM  for  24  seconds,  then  240  BPM  for   the  timepoint  for  that  event  to  occur  in  stream  y  [Figure  
6  seconds.   10].  In  this  example,  a  metro  designating  those  events  in  
stream   x   steps   through   a   table   of   pitches   at   a   constant  
rate  in  its  own  musical  time,  which  happens  to  be  a  con-­‐
stant   rate   in   clock   time   as   well   because   the   tempo   of  
stream   x   is   constant,   60   BPM.   The   timepoint   object   is  
used  to  trigger  each  note  in  stream  y,  and  is  also  used  to  
look  up  the  correct  timepoint  for  the  subsequent  y  event.  
The   time   points   in   stream   y   initially   have   1/2   the   clock  
time  value  of  those  in  stream  x  because  the  slope  of  the  
transfer   function   line   segment   is   1:2.   Each   time   the   slope  
of   the   line   segment   changes,   the   interval   of   the   time  
points   in   stream   y—all   of   which   are   equally   spaced   six-­‐
teenth  notes  in  stream  x—will  change  accordingly.  By  the  
time   the   designated   time   span   of   the   function   has  
elapsed—in   this   example,   60   seconds—the   two   time  
streams  will  reconverge,  landing  in  perfect  synchrony,  in  
this  case  after  240  sixteenth  notes  have  elapsed.  
This   technique   for   warping   timepoints   becomes   more  
interesting  when  the  lookup  function  contains  exponen-­‐
tial  and  logarithmic  curves,  which  result  in  more  gradual  
ritardandi   and   accelerandi   [Figure   11].   A   straight   line  
segment   produces   a   constant   tempo;   the   exact   warping  
effect  of  exponential  and  logarithmic  curves  depends  on  
the  slope  of  the  curve  at  any  given  point,  and  where  it  is  
in  relation  to  the  central  unity  line.  Again,  as  long  as  the  
beginning   point   and   the   ending   point   are   on   the   unity  
line,  e.g.,  0,  0  and  1,  1,    the  two  time  streams  will  be  syn-­‐
chronized  at  those  points.  

 
Figure   11.   Logarithmic  and  exponential  curves  in  the  transfer  function  
result  in  gradual  accelerations  and  decelerations.  

In  this  example,  the  time  streams  start  in  synchrony;  the  

first  segment  of  the  function  is  a  straight  line  with  a  slope  
of   1:1.     The   second   segment   is   an   exponential   ritard  
(lengthening   of   the   timepoints),   the   third   segment   is   an  
acceleration  (shortening  of  the  timepoints)  but  still  does  
  not   achieve   the   original   tempo,   and   the   final   segment  
Figure  10.  An  implementation  of  a  timepoint  transfer  function  in  Max.   crosses   the   unity   line   at   a   high   rate   of   speed   before   a  
final  ritard  to  resynchronize  with  the  x  value.  This  type  of  
How  can  one  implement  this  in  Max  using  tempo-­‐relative   gradual   timepoint   warping   is   arguably   more   expressive  
timing   without   manually   doing   all   the   foregoing   calcula-­‐ than   the   discrete   shifts   and   constant   tempi   caused   by  
tions?  At  each  event  that  occurs  in  stream  x,  we  can  use   straight   line   segments,   and   the   accelerations   and   decel-­‐
its  timepoint  to  look  up  the  corresponding  y  value  for  the   erations   permit   us   to   perceive   the   sort   of   phasing   ef-­‐
next   scheduled   timepoint   to   determine   when   that   fect—the   constantly   changing   relationship   between   the  
timepoint   should   occur   in   stream   y.   For   example,   if   two   or   more   time   streams—that   is   reminiscent   of   the  
stream  x  is  scheduling  events  every  sixteenth  note  in  its  

7
 KEAMSAC2012 - Korean Electro-Acoustic Music Society's 2012 Annual Conference - Proceedings

acceleration  studies  of  Nancarrow  and  the  early  phasing   Reich,  Steve  (1967).  Piano  Phase.  London:  Universal  Editions.  
works  by  Reich.  
Reich,  Steve  (1971).  Drumming.  New  York:  Boosey  &  Hawkes.  

Reich,   Steve   (1987).   “Come   Out”   (1966).   Early   Works.   New  

Conclusion   York:  Elektra/Nonesuch.  
Time   is   our   way   of   measuring   sequences   of   events,   and   Riley,   Terry   (1989).   In   C   (1964).   New   York:   Associated   Music  
sound   is   music’s   way   of   articulating   time.   Every   bit   as   Publishers,   Inc.   (originally   published   by   Celestial   Har-­‐
much  as  sound,  time  is  the  material  from  which  music  is   monies).  
made.   Musicians   have   been   equally   fascinated   by   both  
sound  and  time.  Because  composers  make  music  in  non-­‐ Stockhausen,   Karlheinz   (1963).   Gruppen.   London:   Universal  
real   time—out   of   time,   as   it   were—they   have   always   had   Editions.  
a  particular  interest  in  the  intricate  manipulation  of  time.  
Vercoe,   Barry,   et   al   (1991-­‐2008).   Csound   (music   programming  
Conlon   Nancarrow   was   an   innovator   in   the   composition   language).  http://sourceforge.net/projects/csound/  
of  time,  and  a  pioneer  in  the  use  of  machinery  to  support  
his   musical   innovation.   There   is   still   much   interesting   Vercoe,   Barry,   et   al   (1992).   Csound   Manual.   Cambridge,   MA:  
work   to   be   accomplished   in   the   theory,   taxonomy,   and   Massachusetts  Institute  of  Technology.  
experimentation  of  time  manipulation  as  a  compositional   Whitney,  John  (1980).  Digital  Harmony:  On  the  Complementari-­‐
technique.   ty  of  Music  and  Visual  Art.  New  York:  McGraw-­‐Hill  Inc.  
Computers  provide  us  with  unprecedented  resources  for  
Xenakis,   Iannis   (1955).   “La   crise   de   la   musique   sérielle”.  
experimentation  with  musical  time,  not  just  for  increased  
Gravesaner  Blätter,  1,  2-­‐4.  
precision—although   they   certainly   do   facilitate   that—but  
also   for   experimenting   with   the   very   plasticity   of   time   Xenakis,   Iannis   (1967).   Pithoprakta   (1955-­‐56).   London:   Boosey  
and  the  role  of  polytemporality  in  music.   &  Hawkes.  
It  is  my  hope  that  this  essay  on  polytemporality,  present-­‐ Zicarelli,   David   (2012).   Personal   correspondence   with   the   au-­‐
ed   in   the   context   of   a   computer   music   conference   on   the   thor,  September  5,  2012.  Unpublished.    
centenary  of  Nancarrow’s  birth,  might  point  to  the  rele-­‐
vance   of   his   ideas   in   the   field   of   computer   music,   and    
might  inspire  practitioners  of  computer  music  to  explore  
challenging  new  temporal  aspects  of  music.  

References  
Bumgardner,  Jim  (2006).  Whitney  Music  Box  (web  application).  
http://whitneymusicbox.org/  

Chung,  Huijae  (2007).  Multi  Tempi  21.  Premiered  at  Nong  Pro-­‐
ject  2007,  Seoul,  Korea.  http://huijaemusic.com/  

Gann,   Kyle   (1995).   The   Music   of   Conlon   Nancarrow.   Cambridge,  

UK:  Cambridge  University  Press.  
Jaffe,   David   (1985).   “Ensemble   Timing   in   Computer   Music”.  
Computer  Music  Journal,  9:4,  38-­‐48.  

Nancarrow,   Conlon   (1986).   String   Quartet   (1945).   Baltimore:  

Sonic  Arts  Editions,  Smith  Publications.  
Nancarrow,   Conlon   (1991).   Trio   (1940).   Baltimore:   Sonic   Arts  
Editions,  Smith  Publications.  

Nemire,   Julie   A   (2012).   “Convergence   Points   in   Nancarrow’s  

Tempo  Canons”.  Online  Symposium:  Conlon  Nancarrow,  
Life  and  Music.  http://conlonnancarrow.org/symposium/  

Plamenac,   Dragan,   ed.   (1966).   Johannes   Ockeghem:   Complete  

Works,  vol.  2.  New  York:  American  Musicological  Socie-­‐
ty.  

Puckette,   Miller,   et   al   (1990-­‐2011).   Max   (computer   music   pro-­‐

gramming   environment).   San   Francisco:   Cycling   ’74  
(originally  published  by  Paris:  IRCAM).