SYNTHESIS FUNDAMENTALS

SOund svnt hcsi s i s t hc gcner at i on

( r i
a si gnal t hal cr eat es a dcsi r cd acoust i c
sensat i on.
' [ hi s
chapt cr begi ns ui t h t he f Lr ndament al s of si gnal gener at i on and
pr asent s t echni ques of addi t i ve svnt hcsi s. modul at i ( r n, and noi sc gcnr r i r t i L, I | .
Sa\ cr al exampl e conr put er i nst r unt cr r t desi qns ar e gi ven al ong u' i t h cr r mposi t i onal
c\l nrfi l c.i .
3. 1 Computer lnstruments and Unit Generators
l n computer nrusi c. the tcrm i nJtrutnent refers to an al gori thn thi rt rcal i l !' s
( per t or ms)
a mu: i cal r ' vcnt . l t i s cal l cd upon bv a conr put cr pr ogr anl t hut i s
i nl cr pr cl i ng ei t her a scor e st or cd i r r r ner nor y or l he acl r ons of a per f ol mcr on a
t r ansduccr . The i r sl r umcnt al gor i t hr n cal cul at cs t he sampl e vl l ues ol an audi o
si gnal usi ng i nputs. known as paramcters. recci ved from the cal l i ng program. For
exampl c. an i nstrumcnt desi gncd to
pl av a si ngl e. si l npl e tone rni ght be passed
par amet er s cont r ol l i ng t he dur at i on. i r cqucncv. and anpl i t ude of t hc t one. Ot her
p:rrrmcters can bc passed that affe(t
(rther aspects of the sound. When de:i i gni ng an
i nst r unr cnt . t he musi ci an det er mLne: t he number and nat u( c
( ] f
t he par i r mcl cr s l o
bc l r.rssed.
' l ' hese
are bascd on ,r choi cr
(j f
s' hi ch attl i butes of thc s(,und wi l l be
control l ccl cracrral A duri ng the genc' rrti on of thc soun,.l . An i nstrument can al so bc
desi gnccl to accept i .rn audi o si gnal i n di gi tal forrn as an i nput to bc processed bl the
al gor i t hm.
For bot h concept ual cl ar i t v and pr oqr ammi nq conveni encc. a col npl et e sound-
- qencr at i ng
al gor i t hm i s usual h t l i r i dct l i nt o smal l cr . separ at e al gor i t hms cal l ed r r r l r
gctrcrotor,s. Each uni t generator hns i Dpul pi rrameters and al l cast one
(rutpul .
Each
performs a speci l i c functi on of sl !:nal generati on, nrodi l i cati on. or conrbi nati on.
N4ost musi c l anguages express synthcsi s al gori thms i D terms of uni t gcnenrt.rrs-
usi ng t hcm as t he bui l di ng bl ocks wi t h whi ch i nst r ument s ar e made. The i nt er nal
al g<rri thm of each uni t generator hi rs been determi ned and encoded by thc
rrrusi c-systems programmer. Thc musi ci an' s task i s to i nt!' rconnect the i nputs and
i ) ut put s of t he uni t
Lencr at or s
t o achi a\ ! ' t he desi r ed. r ) \ r ' r i t l l s\ nt hcsi s al gor i t hnr .
Thi s i s qui t e si nr i l ar t o t hc wa\ i n $hi ch i nst r umcnl s ar c r eal i zed on anal og
\ vnt hesi zcr s by connect i ng ( pat chi ne) vi r r i ous modul es l ogct her .
- f h! ' unr 1
gcncf t r t or
i s a useful concept bcci ruse i t mi ni mi zcs the i l mount of knowl edge of thc i nrter
u' or ki ngs of cach al gor i t hm r cqui r ed on t hc par t of t h! ' musi ci an. whi l e r et ai ni ng
consi dcr abl e f l exi bi l i t v f <r r t he cons( r uct i <r n of s1' nt hcsi : al gor i t hms. I ndi vi dual uni t
63
)c
r S:
I t e
l l )
ol
l eo
t he
t hc
1()n
Synthesk fundamental8
combi natori al operati on i s addi ti on. represented i n a si gnal fl ow chart by the
svmbol for thc adder shown i n l i gure 3.1. An adder has two or morc i nputs.
dcnoted by arrows, and one output. The pri nci pal use of an adder i s to mi x si gnal s
together. The operati on of subtracti on al so uses an adder. l n thi s case. the sl grr of
the subtrahend i s reversed
j ust
before i t enters the adder. Thi s i s i ndi cated by a
mi nus si gn pl aced near the arrow connecti ng the subtrahend to the adder.
The si gnal fl ow chart shown i n the fi gure i ncl udes a mul ti pl i er and a di vi der. as
\\' el l . Mul ti pl yi ng a si gnal by a constant wi th a val ue greater than one i ncreases the
ampl i tude of the si gnal : thi s process i s cal l ed ampl i l i cati on. The reverse process.
attenuati on. can be obtai ned through mul ti pl yi ng by a constant l ess than one.
Mul ti pl i cati on and di vi si on general l y take the computer substanti al l ) l onger to
perform than addi ti on or subtracti on; therefore. the i nstrument desi gner tri es to
mi ni mi ze the number of these operati ons. The use of di vi si on requi res speci al care.
To avoi d errors, the i nstrument desi gner must make certai n that the di vi sor can
never assume a val ue of zero. Such an error can be seri ous enough to causc the
' \ nt he\ i s
pr ogr am l o cer se oper r l i on.
The parameters passed to an i nstrument are usual l y denoted i n a si gnal fl ow
i hart by dcscri pti ve mnemoni cs. For exampl e. the parameter that control s the
.rmpl i tude of an i nstrument i s often desi gnated AMP.
Every i nstrument must have at l east one output. The fl ow-chart symbol for an
output i s a smal l . empty ci rcl e l ocated at the end the chart. There may be mul ti pl e
(rutputs
correspondi ng to a mul ti -channel audi o system.
-1.3
The Oscillator
The uni t generator fundamental to al most al l computer sound synthesi s i s the
,,sci l l ator. An osci l l ator generates a peri odi c waveform. The control s appl i ed to an
,rsci l l ator determi ne the ampl i tude. frcquency. and type of waveform that i t
troduces. The symbol for an osci l l ator i s shown i n fi gure 3.2. The symbol i nsi de the
,j \ci l l ator (WF i n thi s case) desi gnates the waveform of the osci l l ator. The symbol
;.rn be a mnenoni c of a parti cul ar waveform or a drawi ng of one cycl e of the
,raveform.
The numeri cal val ue that i s fed i nto the l eft i nput sets the peak
.l mpl i tude of the si gnal . The numeri cal val ue appl i ed to the ri ght i nput determi nes
:hc frcqucncy at whi ch the osci l l ator repeats the waveform. The frequency can be
.peci l i ed
i n one of two wa1' s: (1) an actual number of Hertz. or (2) a sampl i ng
65
) n
hc
I
AMP
: I (i L' RE 3. 2 Fl owchart svmbol f or an osci l l at or. The
phase
i nput i s not of t en uscd.
66
Computer Music
i ncr er r r ent .
( Sampl i ng
i ncr cmcnt . a numt r er pr opor t i onal t o t he f r cquencv i s
expl ai ncd bcl ow.
)
Tbe i nput
( ) n
l hc r i ght si de of t hc r r sci l l at or . PHASE. dct cr nr i ncs
l t $hi ch poi nt on thc \\' a\ei orrn the osci l l ator bt' gi ns. PHASL. i s usual l v nol
spc' ci fi ecl unl css requi red l or an expl i ci t purpose. The oul put of the osci l l ator i s a
sequencc of sampl es u' hi ch forms a di gi tal si gnal representi ng the wavcfornr.
C)ne method of i mpl cmcnti ng an Osci l l ator al gori thrn speci fi cs the \\,aveform as
a mrthcmati cal functi on of ti me. Thus. i n order to gctrcrate a si ne wave usi ng thi s
mct hod. t he al gor i t hm woul d have t o cal cul at e t hc \ al uc of t hc mat hemat i cal
functi on. si ne. on cvery sampl e. Thi s mcthod
(di rect eval Lrati on) for nearl y an_"-
f unct i on i s pr ohi bi t i vel v sl ow.
For thc sake of el l i ci ency. most di gi tal osci l l al ors use a stored \\' avefr)rnr: a
$i l c' form that i s eval uated pri or to the generati on of any sound. l ' he computer
ci i l cul :rtcs thc val ue of many poi nts on a cycl e Of the rvaveform. and stores then] i n
conrputcr memory as a bl ock cal l cd a $' avetabl e. Thus. a wavetabl e consi sts of a
l ong sequence of numbers. cach correspondi ng to the val ue ol successi vc poi nts on
the waveform. Oncc the waveform has bccn stored. the osci l l ator can generute
sampl c val ues by si mpl y rctri cri ng val ues from thc !rrvetabl e a much fas(cr
operxti on for the computer than el al uati ng the wavcform di rectl y.
As i rn cxampl e. cQnsi der thc \\i r\cti i bl c i n fi gure
-j .-i .
It contai ns one cycl e of a
si nc r vavc sl or cd i n , 5l l cnt r i es. Each enl r v i s nr l r l , "' d bv an addr ess. denot ed i D t hi s
casc by t hc i nt eger s f r om 0 t hr ough 5l l . The osci l l at or al gor i t hm ml i nt ui t t s i r
numcr i cal val ue. cal l ed t he phase. whi ch i ndi cat es t he addr ess of t he ent r y cur r enl l y
i n usc. At the begi nni ng oI i ts operati on. the osci l l ator i s gi ven an i ni ti al val ue of
phase whi ch denotcs thc fi rst entry i n the wavetabl c to be uscd. On cvcrv sampl c
thc osci l l ator al gori thm obtai ns thc currcnt val ue of the phase (d) and adds to i t an
t
.(l
;rr
AJ
n
a
t
' a
I
'll
t
:\-
t
- \
-
0 0. 0 0 00
0 0r 2f
2 o. 4245
121 0. 9999
124 r . 0000
129 0. 9999
l 1s 0. 69 8 4
324 a. 7a t 1
321
509 0. 016 8
510 0. 0 2 45
511
' , .
I
J
FI C; LI RE l . - l Wa\ et i l bl e coDl r r ni nI one cvcl e of a sr ] c ! r n\ e.
4usic Synthe6is Fu.trdanentals
amount that i s proporti onal to thc frequencv of the osci l l ati on. The new val ue of
phasc deterni ncs the cntry used to cl l cul atc thc nex( outpr( sampl e. Thc amr:runt
added t o t he phasc on cver v sampl e i s cal l ed t he sampl i ng i ncr ement ( Sl ) : t he
di stance i n the *avetahl e betwecn successi ve cntri es sel ectcd h!' the osci l l xtor.
When t he val ue of t he phase excccds t hc nunber of t hc l ast cnt r v i n t he t abl e. i t i s
' ' wr apped
ar ound" t o r poi nl near t he bcgi nni ns of t he t abl e b) ' subt r act i ng t he
total Dun]bcr of tabl c cntri es from l hat phase. l n thi s cxampl e. the nunrber of the
l ast ent r y i n t he t abl e i s 51t . I f <b=512 af t cr addi ng t he sampl i ng i ncr er nent . t hen
l hc osci l l at or al gor i t hm woul d modi l y t he phase so t hat d=d 512
=
0. t her eby
rcturni ng phase to the fl rst l ocati on of the tabl e.
' l ' hercfore.
the osci l l ator al gori thm
can be thought of l s scanni ng thc wavetabl e i n a ci rcul ar l ashi trn.
Thc two vari cti es of di gi tal osci l l ator commonl v encountered i n computer
rtrusi c are the fi xcd sl mpl i ng rrte osci l l ator and thc vari abl e sampl i ng ratc
osci l l ator. l he rerrrai nder of thi s secti ol l descri bes thc operati on of the fi xed
sampl i ne r at e osci l l at or . Sec sect i on l . 10 t or a di scussi on of var i abl e sampl i ng r at e
rrsci l l i ttors.
Usi ng t he wavet abl c i n l i cur e 3. - 1. supposc t hat t he sampl i ng r at c i s. l 0 kHz and
l he osci l l ator i s programmed to scan through the wavetabl e wi th a sampl i ng
i ncr cment of one: t hat i s. one cnt r y l t a t i me. Ther e ar c 512 cnt r i es i n t hc t abl e and
t he t abl e cont ai ns onc ct cl e. so i t woul d t ake 512 sampl es t o pr oduce one cvcl e.
Ther et br e. t he f r cquencv of t he osci l l at or woukl be. 10 kHz15l 2
-
78. 13 Hz.
l f a l one one octave hi ghr"-r i s desi red. the osci l l ator woul d be programmL-d to
tctri eve val ues from cverv other entr) i n the wavetabl e (Sl =2). Because thc
osci l l atr)r woul d go throug.h thc r,uvetabl c tu' i ce as fast. l here are hal f as nrany
sampl cs per cycl e
{256).
anti the frequencv of the osci l l ator $oul d bc.l 0 kHz/256
-
156. 25 Hz.
For a wavcti rbl e wi th N enl ri es. to ebti ri n a frequency/q1. the rcqui red sampl i ng
tncrctnent l 5
For exampl c. i f N=5l 2 and t he sar npl i ng r at c ( . / , ) i s
- 10
kHz. a 2. 5 kHz si gnal woul d
r equi r e a sampl i ne i ncr emcnt r r f
- l l .
That i s, i l t he osci l l at or st ar t s i r r ent r ! zcr ( ) i n
t he wavet abl e. ent r i cs 0. 32. 6. 1. . . . u' i l l be t akcn f r om t he wavet abl e.
Of ctrursc. exccpt for ccrtai n scl ect frequcnci cs. the sumpl i ng i ncremenl wi l l
not be an exact i nt egcr . For i nst ance. wi t h N=- 512. a
. 1, 10- Hz
t onc at a. 10 kHz
sl mpl i ng r at c r ecl ui r es a sampl i ng i l cr ement of 5. 632. Supposc. i n r hi 5 casc. t hat
thc osci l l ator starts at a phase cqual to zcro. On the fi rst sampl e. i t retri eves thc
*l vetbrm val ue from that l ocati on. On the next sampl e. the phase i s
t ) +5. 6- i 2=5. 632. How docs t he osci l l at or t r cat a phasc wi t h a f r act i onal par t . i f t he
cnt r i es i n t hc wa\ ct abl c ar e r nar kcd bv i nt cger s' l Thcr c ar c t hr eL' t echni ques:
I r uncat i on, r oundi ng. and i nt cr pol at i on.
In truncati on. the l racti onal part of the phasc i s i gnored i n dctcrmi ni ng the
rvavetl bl e entry. so that i n thi s casc the val ue i s takcn l rom entrv 5.
-fo
cal cul ate thc
ncxt phase. howevcr. the osci l l l rtor i ncl udes i n i ts addi ti on thc fracti onal part of the
r - ur r enl phasc. Thus. on t he nt ' xt sanpl c. t he phasc i s
- r . 6- 12+5. 632=
I L2t r . l . causi ng
t he sanl pl e t o bc t akcn f r om ent r y I l .
Whcn roundi ng i s used. thc entrv takcn i s the val uc t)f thc phase rounded t() the
nearest i nteger. Thus. for the cxampl c above. the fi rst threc u' avel atrl e ral ues are
67
) l s
I ncs
not
i sa
]t as
l hl s
t i cal
i t nrv
l t : a
' ul Cr
or l n
ot it
5
( ) n
: l at e
l i t cr
ol iI
r l hl s
nsa
r nt l v
r c of
rpl e
i r an
Computer Music
takcn from entri es 0. 6. and 11. rcspecti vel y. Roundi ng vi el ds onl y a sl i ghtl y nrore
accurate wavefbrm than truncati on. and takes more computati on ti mc.
Of the three techni ques. i nterpol ati on gi ves by far the most accurate approx-
i mati on of the waveform. When a phase fal l s betrvcen two i ntcger val ues. the
waveform val ue i s cal cul ated as a combi nati on of the two entri es betwecn whi ch the
phase fal l s. l f, as above. the phase i s
-5.632.
the osci l l ator al gori thm i nterpol ates the
waveform val uc as a uei ghted average of entri cs 5 and 6. l n thi s case. the phase i s
63.1?i of the di stance between 5 and 6. so the waveform woul d be eval uated as the
sun of 63.2\l c of entD' 6 and 36.l Jcl c
(l f
entry 5. Thi s process can be thought of as
taki ng the \!al eform val uc on a strai ght l i ne thaf connects the val ues of successi ve
wavetabl e entri es. Interpol ati on adds an extra mul ti pl i cati on to the osci l l ator
al gori thn and thus i ncreases the amount ol computati on ti me.
The i naccuraci es i ntroduced i n the waveform by any of the three tcchni ques
di scusscd prcvi ousi y evi dence themsel ves as sorre form of noi se or other unwant!d
si gnal i n the sound. The amount and qual i ty of the noi se crcated depends on the
wavcform, on the tabl e si zc. on the val ue of the sarnpl i ng i ncrement. and on the
techni que used. The l arger the tabl e si ze. the better the si gnal -to-noi se
(SrN)
rati o.
(Sec
secti on I .5.
)
Let k be rcl ated to the tabl e si ze (N
)
by k
=
l og1N. For exampl e. i l
N=-5 l 2
=
2e. then t
=9.
If the entri es i n the tabl e are stored wi th sufl i ci cnt
prcci si on
to prevent si gni l i cant quanti zati on noi se (see secti on 1.5.). the worst S/N rati o that
can occur i s gi ven by the approxi mate expressi ons 6(k-2) dB for truncati on. 6(t-l )
dB f or r oundi ng. and l 2( k- 1) dB f or i nt er pol at i on. r Negl ect i ng f or a moment t he
Si N rati os of the data convcrters. an osci l l ator usi ng a 512-entry tabl e, for exanrpl e,
woul d produce
tones wi th no worse than 42.
,18.
and 96 dB S/N rati os for
truncati on. roundi ng. and i nterpol ati on, respecti vel y. The noi se l evel resul ti ng
from a fracti onal sampl i ng i ncrement vari es di rectl l wi th the ampl i tude of thc
si gnal . Thus. S/N rati o due to thi s effect i s the same on l oud sounds as i t i s on soft
sounds. f)f coursc. the actual SIN rati o of a sound woul d be dctermi ned bv
combi ni ng the noi se due to the data converters and the noi se resul ti ng Irom
f r r ct i onal phase.
As mi ght be expccted. the expressi ons above show that methods requi ri ng
more computati on ti nre or l argcr l abl e si ze perform better. Thc perl ormance of any
method can bc i mprovcd by i ncreasi ng tabl e si ze. and so the di gi tal ' osci l l ator
desi gner i s fac-cd *' i th the common compromi se: computati on speed versus memory
si ze. Many comput er - musi c syst ems make avai l abl e bot h t r uncat i ng ! nd i nt cr pol at -
i ng osci l l ators. so the musi ci an can make the trade-off between si gnal qual i ty and
compututi on specd based on the appl i cati on of a parti cul ar osci l l a(or.
3.4 Definition of the Waveform
Gencrai l -v' , the musi ci an need not di rectl v spcci fy a numeri cal val ue for each
l ocati on i n the wavctabl e. Ordi nari l v l he compul er musi c progrant enabl es a rl ore
si mpl e nr et hod of ent r y: ei t her br ent er i ng i t s r epr er ent at i on ver sus t i me. or bv
speci fvi ng whi ch frequency componenl s i t cDntai ns. The defi ni ti on of thc waveform
versus ti me can be made bv speci fyi ng thc mathemal i cal cqurl i on that rel 2rtes the
ampl i tude of thc desi red wareform to i ts phase. The waveform tersus ti rne can al so
bc del i ned bv a pi cce-wi :e l i nei rr means. Here. the $avcf()rm i s del i ned bl
SynthesiB Fuldamentals
speci fvi ng a number of representati vc poi nts on the waveform. Thesc poi nts. cal l ed
breakpoi nts. are the poi nts where the waveform changcs sl ope. When l i l l i ng the
wavetabl e. the software connects the breakpoi nts wj th strai ght l i nes. In most
progranl s, breakpoi nts are speci fi ed as a pai r ttf numbers: phasc and ampl i tudc at
that phase.
The speci l i cati on of waveforms i n terms of ampl i tude versus l i me can.
however. someti ntes l ead to unexpected resul ts. l f. at the frequency at whi ch i t
repeats. the waveform has any harmoni cs above the Nyqui st frequency. thev wi l l be
fol ded over (al i ased). therebv produci ng unexpected frequenci es i n the sound.
Suppose i n a system wi th a 20-kHz sampl i ng rate. a musi ci an speci fi ed a sawtoorh
wrvefornr (fi eure 3..1a) and used i t i n an osci l l ator programnrcd to produce a l one at
a frequencv of 1760 Hz. The si xth harmoni c of 1760 Hz woul d be 10,560 Hz. whi ch
i s above the Nyqui st frequency of 10 kHz. Therefore. the si xth harnroni c woul d fol d
over l o 20,000 10,-560
:
9.+40 Hz. fhe sevcnth harmoni c. expecred ar 12.320 Hz.
woul d scrund at 7680 Hz. and so on. l i i gure 3.-{b i l l usrrares the i ntended spcctrum oi
the sawtooth wave and Fi gure 3.4c shows how unexpected components appear i n
the spectrum al the output of the D/A convcrter. A sawtooth waveform has a
si sni fi cant amount of energY i n i ts uppcr harmoni cs. and so the resul ti ng spectrum
woul d not sound compl etel l , harnroni c. Thus. to avoi d fol dover when speci fyi ng
waveforms i n terms of ampl i tude versus ti mc. one shoul d del i ne a waveform wi th
l i ttl e si gni fi cant cnergy i n the uppcr harmoni cs.
A :afer u' a1' to sp!ci fy a waveform i s i n terms of i ts spectrum. Here . the
i nstrument desi gner speci fi es the ampLi tude. thc parti al number. and. i f desi red. thc
phase of cach component. The software then cal cul ates and stores a si ngl e c-vcl e of
69
:
j
:
s
S
t r
t s
, d
l c
1e
i f
JN
rat
. r )
fre
l e.
tor
i n i '
t hc
,oft
by
onr
any
Itor
rory
) l at -
and
fs
2
-/1o)
eacn
mt)re
rr b-""
i 0r m
i s the
r r l so
) o ov
f.
z
FI GLTRE . 3. 1
(a)
Sa\ \ ' t oot h rva\ ef ormi
{b)
i t s expect ed spcel rum at a )760 Hz f undamerral
f requency: and (c) i t s act ual Jpect rum af t er conversi on at a l 0 kl l z sarnpl i ng rat e.
70
Computer Music
l hr - ' cor r espondi ngr vavef or nr . Of t er ) . l heampl i t udcsof l hehar r noni csar cdt ' scr i bed
r cl at i ve 1o t he ampl i t ude of t hc f undament al . For i nsl ance. one coul t l spcci f t , a
wavcforrr contai ni ng a uni t-anrpl i tude fundanrental rri th a thi rd-harmoni c anpl i -
tudc l O dB bel orv the fundamental and a scventh h:rrmoni c 22 dB down. When the
wavet br m i s def i ncd i n t er ms of spcct r al cont ent . t hc nr usi ci an easi l ) ' kno$s t he
cxi rct val uc Of the hi ghcst harrnoni c contai ned i n the spectrum. Al i asi ng of certai n
uavefttrms can thus be avoi ded bv l i mi ti ng the' fundrmcntal l requency of the
<r sci l l : r t r l r . For exampl c. on a st - st em wi t h a. 10- kl l z sampl i n{ r at c. t hc f undament al
frcquencv rrt an osci l l ator pr<rcl uci ng l (l harmoni cs shoul d not exceetl 2 kHz.
,A.n osci l l ator wi th a i xed frequencv sampl es the stored u,avefornr
q' i th
a
c()nsl rnt sal npl i ng i ncremcnl . fherefore. a pcri odi c sal eform i s gencratcd. und so
thc spcctrum of the si gnal contai ns nothi ng but exact harmoni cs. Thus. *,hen
descri bi nc a waveform i n terms of spcctral content- usi ng non-i ntcger parti al
numbers u' i l l not resul t i n a si gnal wi th an i nharmoni c spcctrum. Suppose un
i nstrument desi gner. i n hopes trf obtai ni ng an i nharnl oni c spectrum. speci fi ed a
f undanr ent al and a par t i al number of 2. 2. When t hc \ \ avct abl e i s sampl ed bv an
osci l l l t or . t hc r esul t i ng si gnal $oul d be per i odi c. and t her et or e have- a har moni c
spect r um. l nst cad of gener at i ng a c( ) mponcnt nt 2. 2 t i mes t he f undamen( al .
( hc
ener gv cr pcct ed at t hat I r cquencv woul d bc spt cud t br ouehout t he spcct r unt as
har nr r ni cs of t he f undar ner l al . Usual l v t hi s r esul l s i n a spcct r um t hat i s not
band' l i nr i t cd. cr eat i ng t he pol ent i al f or f ol dover .
3.5 Generating Functions of Time
Chapter : denl onstrate(i thl t thc paramcters of musi cal sound are consti l ntl l '
changi ng. l hus. t he i nput s t o an osci l l dt or al most al wa\ s var v wi t h t i me: t hi l t i s. t hc
ampl i l ude and l r equel cr of an osci i l at or ar c cont r ol l c! l bv f unct i ons of t i me. An
osci l l ator can be used to generate these control functi ons. but synthcsi s sl stenrs al so
i nchr(i e envcl opc generators and other functi on
-qencrrtors
that. trecausc thcy arc
tai l orcd for thi s speci ti c purpose. can svnthesi ze control functi ons morc di rectl ;-- .
Fi gure f.5a shou,s one of the si mpl cst computer i nstruments. I' he output ol the
envel (rpe generator (l i gure 3.5b) control s the ampl i tude of thc osci l l ator. so that the
i nst r um! ' nt pr oduccs a f i xed uavcf or m encl osed i n t hc cnvel ope ( hgur e 3. 5c) .
Si r npl c ampl i t ude envckr pcs ( f i gur e 3. 6) have t hr cc sc{r nent s: t he at t ack. $hi ch
descr i bcs ho\ \ ' t hc ampl i t udc r i scs dur i nq t he onsct of t hc
( ) ne:
t he sust ai n. whi ch
dcscri bes thc charactsri sti cs of l he tr:rnc i n i ts sl eari v statr: and the decav. rvhi ch
descri bcs ho$' the tone di cs a\\av. An envel ope g(' n(' ri ttor has at l east four i npul
pari tnel ers: ri se ti mc whi ch i s l he durati on of the attack scgrl cnt. ampl i tude uhi ch
sets the val ue at the peak of the attack. total durati on ol the envel ope. and deca-v
t i me. I n addi t i on, t he shapes of t he at t acks and decay scgment s need t o be
speci fi ecl . Dcpendi ng on the tvpc of envel ope gcnerator. thi s can be done i n one of
t\\,o $' xvs. Some envel ope gencrators determi ne the segmcnt shape by reference to
l functi on stored i n a waveti l bl e. In thi s case. the enti re $' avetabl c i s scanned
cxi l ct l v once i n t he t i nl c of t hc segment . Ot hcr t ypes of envel ope gener at or s ha\ c
pr cdet er mi ned shapes. For exampl e. scver al l anguages i mpl cmcnt a uni t gcner at or
ci r l l cd l i nen" whi ch r cal i zes er r vel opes wi t h st r i ct l v l i ner r at t acks and dccl l s.
Itt
:9C
) ed
r l i
t hc
t hc
ai n
t he
I t al
t a
I So
t el t
1l a I
i t n
(l
a
an
' n
i c
t hc
Las
nol
Synthesis Ftrndamentals
RI SE
TIM E
7l
DTCAY
TI ME
o)
FIGURE
-1..s
(a)
Si mpl e compurer i nsrrument wi l h
outprl t \\' avcfbrm (c).
o)
l l l l l l l l l r
. t I t I
'ultulll4llllliljuw
c)
i t \ . r nr [ j l i t uJe sn\ ( ' l ope
{b) . r nJ r r \
FI ( ; URE 3. 6 Si mpl e r mpl i r ude envel ope.
On many systems. an envel ope
[cnerator can be used as a si gnal processor. A
si gnal i s appl i ed to thc ampl i tudc i nput of the envcl ope gencrator.
.I.hi s
resul ts i n an
output si gnal that i s the i nput si gnal cncased i n an envel oDc. Thc i nstrument of
f i gur e 3. 7 i s i dent i cal i n f unct i on t o r hr t ot t he t r ne i n t i sur e i . 5a. I nst cad of dr i vi ng
t he ampl i t udc i nput of t he osci l l at or r vi t h an envel ope. a const ant ( AMp)
i s appl i er l .
Thi s causes the osci l l ator to produce a waveform wi th a constant ampl i tudc. Fussj ns
xt l !
t he
An
r l so
are
t t,v.
t nc
t he
r i ch
I Ch
l l ch
I put
r l ch
:ci ty
be
eof
ct o
ned
l i t ve
at ol
TIM E
AMP
AMP
72
DECAY TIME
Computer Music
TiM E
T ME
SyDtLc
Jr *l cnr . t
I n l ( r uJ
_i . \ b. t r
l l nc. r r r
\ U
ddc n
e( r n\ ( . 1r
i l mnl ] l L
\ \ \ l an(
' f
he r. ' r
l hJl r . .
mt nl nt u
r hc r . , l r
r n! i I t c
JB. q; . 1
r. 1n! a \ r
. l ! n. 1
I
JI I : _. 1,
I h cr c: t r
' ) ' I
(
i - ' l
-
l h.
: I nt - : - .
4a. . , : l ' ( ' l
' . : ! .
: r
RISE TIME
FI Ci URE
-. 1. 7
Anot her wav of i mpart i ng an envel ope
AA
(t
a si gnrl .
1
. 75
. 50
. 25
0
- 20
40
- 60
1
. 75
. 50
. 25
TIME
TIME
0
-24
- 60
b)
d)
FIfi URE 1.8 Deca! functi ons
j (
a) l i ne^r, ( b ) l i near i n d B, (c)
exponenti al . aod
(d
) exponer-
t i al i n dB.
thi \ si gnal through thc envel ope generaror i mpartt a pattern of ampl i tude vari ati on
onto thc waveti )rm_
The shape ot thc attack and dccay porri ons
of the e n\.el ope has a great effecr on
thc percei \cd
ti mbre of a tone. Fi gure 3.8 depi cts the tr\o sl l apes rnost commonl l ,
encount cr ed I n conput er musi c: l i near ( 3. {Ja)
and exponent i al ( 3. gc) .
Because
\ t
-_---
Sytrthesi6 Fundamentals
l i steners percei ve ampl i tude on a nearl y l ogari thmi c scal e. a more constant change
i n l oudness wi l l be obtai ned wi th an exponenti al shape than a l i ncar one. Fi gures
3.8b and 3.8d shnw how each shape progresses i n terms of dB. A sound wi th a
l i near decay wi l l appear to l i nger on after the begi nni ng of the dccav. and then
suddenl y drop off near the end of the tone. The exponenti al decay refl ects a
constant change i n dB versus ti me and thus u,i l l sound as a smooth di mi nuti on of
ampl i tudc. Natural vi brati ons al rnost al u' avs di e away exponenti al l v.
A truc exponenti al shape can never reach exactl y zero. and so on many
svstems. the musi ci an must speci fy the maxi nrun and mi ni mum val ues of the shape.
The rati o ot the two i s i mportant because i t sets how qui ckl y the anrpl i tude changcs:
that i s. thc rate of change of the se!tment i n dB/second. tf i t i s desi red that the
mi ni munr val ue resul t i n an i naudi bl e si gnal , i t may not be a good strategy to make
the val ue arbi trari l y smal l . Suppose that an exponenti al attack i s to l ast .1 second
and the rati o i s chosen as l :1.000.000 (120 dB). Thi s i s a rate of change of 1200
dB/second. Further assume that the svstem has l 5-bi t D/A converters, a dynami c
range ol about 90 dB. Dependi ng on the ampl i tude of the tone. the envel ope wi l l
have to ri se al l east 30 dB before the converter begi ns to produce a ti me-varyi ng
si gnal . Because the envel ope ri ses at 12() dB/second, there wi l l be at l east a 30
dB-:1200 dBr' second
:
.025 second addi ti onal del a_v i n the onset of l he tone.
Thcrefore. the rati o chosen shoul d bc no greater than the dvnarni c range of thc
system. i n rhe case of l -5 bi ts. 16.-18.1r l .
The durati on of the attack and decay segments al so has a great i nfl uence on
ti mbre. l n acousti c i nstruments, the attack i s normal l y somewhat shorter than the
decay. A very short attack i s characteri sti c of percussi ve sounds. whereas l onger
attacks are found i n acousti c i nstrurnents, such as the organ, whi ch produce sound
by spl i tti ng a stream of ai r across the edge of a surface. Manv acousti c i nstruments
havc l onger al tacks on l ower pi tches. Synthesi zi ng tones wi th shrrrt decays and
rel ati vel y l ong attacks produces an effect si mi l ar to pl ayi ng a tape recordi ng
backwards. Of course, thi s may be desi rabl e under some ci rcumstances.
A reti nemcnt to the si mpl c cnl el ope generator shoqn i n ti gure
-.1.6
i s the
i nserti on of a fourth segment between the attack and sustai n. An envel ope of thi s
type (ti gure 3.9) i s cal l ed ADSR. representi ng i ts scgments-attac k. decav. sustai n.
ATTACK DTCAY
t ' l GURE. l . 9 ADSR envel ope
73
( ) n
oD
nl !
tse
SUSTAI N
7-
Computer Mwic
and r!-l ease. Thc ADSR l hapc i s i tn attcmpt to i nri tate the cnvcl opcs found i n
acousti c i nsl ruments and i \ commonl v' used i n i nexpensi ve el ectroni c kevboard
sl nthesi zers. Hcre thc tone remai ns i n the sustai n state unri l the kcy i s rel eascd.
Envcl opc generatr)r' r on corrl puter musi c s)stenrs r an i n the conrpl exi t) of l he
tvpcs of en\c' l ()pes thcv can rcal i ze. l n most svsi crrl s. thc avai l abl e envel rpe
generators pcrnri t envel opes wi th onl l two or threc breakpoi nts. When the
crrnpl cxi tv of a dcsi red envel ope excecds the capabi l i ti cs oI the avai l abl c envel opc
gcnerrtors. nn osci l l ator can he used. Fi gure
-l .l {l
i l l ustrates the real i zuti on of an
: r mpl i t ude envcl ope i n t hi s \ r av. The wavef or m of t he en vel opc- se ne r al i ng
osci l l ator i s the desi red envel ope shape. and i ts frequcncy i s chosen to be the
i nvcrse of the durati on of the tone. Thi s wav. thc envel ope rvi l l be generatcd once.
It{usi ci ans have al so used thi s conl l guri tti on to real i ze musi ci tl e\cnts l }r.rt are
repcri ti on! of a tonc. hv protrammi nt the !-nvcl ope-generati n!r osci l l ator to
!l o
throuqh several c_!-cl es duri ng the durati on of thc even1.
A scri ous di sadvantage of usi ng an osci l l ator i nstcad ol an envel opc generator
i 5 t hdt t he at t ack and dccal t i mcs ui l l bc al t cr ed uhen t he dur at i on i s changed.
Thi s. unl ess thc shape of the t,avefornr i s compensated. rvi l l cause qui te noti ceatrl e
di fferences i n ti mbrc over a range ol durati ons.
Envel ope gencrators synthesi ze functi ons of ti mc Nhi ch are best sui ted for
cont r ol l i ng t he ampl i t ude of an osci l l at or . I n cont put cr musi c. ot hL' r f uncl i ons ar c
nscded l rr conl rol ol her parl mcters of a sound such as the frcqucncY vari ati on of an
osci l l ator. As a rcsul t. man! svstenl s i mpl en]ent i nterpol ati ng l uncrron general ors
1o provi de greater l l exi bi l i tv i n rcal i zi ns functi ons of ti me. Thesc ars often
rcprcsented on a fl owchrrt bv l rcctangl e wi th a nrncmoni c or pi cture of the
functi on i nsi de. In usi ns thcse. the rnusi ci an speci fi es thc functi ons of ti ne bv l i sti ng
rcprcsentati ve poi nts on the functi on. For each poi nt a pai r of numbers i s gi vcn:
tuncti onal val ue and ti mc cl apsed si ncc thc prcvi ous poi nt. (Some \vsrems usc thc
con\ cnt i on: f unct i onal val uc ancl t i me el apscci si ncc l hc st ar t of l he t unct i on. )
I ) ur i ng synt hesi s t hc f unct i on gener at or cr l cul at es val ues bv i nt er pol at i ng het wcen
lTr.|
, r c I
. : (
. i (
. \ :
. J
' l c
: ]rr
I _
- \ t i
' __- :
I l
: h
AMP
1/ DUR
I ; l ( i Ll RE 3. 10 l hc use of an o: ci l l at or as an envcl oDe sener at or
Synthesis Fundarnentals
t hc br cakpoi nt s. The i nt er pol at i on ct n be ei t her l i nct r or exponent i t l dependi ne
on the parti cul ar l uncti on generator used. For exampl c. onc coul d obtri n a smooth
gl i ssando bv speci fvi ng exponenti al i nterpol ati on for a functi on dri vi ng the
frequencv i nput of an osci l l ator. In thi s case the functi on val ues woul d be gi vcn as
the frequenci es at the end poi nts of l hc gl i ssando.
3.6 lnstrument Definition in MUSIC 4-Type Languages
' I' hcrr-'
are di sti nct approaches taken wi thi n the MUSIC.I fami l y of sound-svnthesi s
l l nguagcs ( scc
scct i on 1. . 1) . Dcspi t e t hc somet i mes gr eat di f f er ences i n appei r nr nce
l nd synt{x of thc l anguugcs. horvevcr, experi ence i n the use of one of them i s
suffi ci cnt preparati on for the usc of another wi th comparati ve ease. Al l the
pr ogr ams shar e cer t ai n common f er [ ur es. For er r nr pl e. r hev al l usc f unct i ons
\torcd i n compul cr mcmory for u,avel brms and other functi ons of ti me. and they
i ncl ude vari ous subrouti nes for generati ng the stored functi ons and wavctabl es.
They al l ) ogi cal l y separ i r t e t he scor c and r he or chest r a. and al l t he pr ogr ams except
NI USI C i ( ) pr ocess t he scor c i n t hr ce scpar at e
' pa\ \ cs. "
Thc Pri nccton-dcscended Ianguages-Musl c
-l Lr. MUSIC
.1BF.
I\l UStC 160.
NI USI C 7. and MUSI C 1l have a st r uct ur e t hat cal l s f er t he musi ci an r o suppl v an
orchestra ol i nstrunents coded i n a speci al l anguage (that vari es wi del v among
them) and to provi de a score. or note l i st. Thc orchestra i nput for thrst' progri i nts i s
i n t he f or m of st at ement s coded i n ei r her a new l anguaqe or . i n t he case of NI USI C
. l BF.
codcd i n FORTRAN whi ch r cpr csent t he conn{ur at i on of uni r g( ncr r t or ' \
ncccssar v t o pr oducc t hc dcsi r cd sounds. Thc scor e i nput i s i n t hc f or nr of
' not cs'
of dat a t o be
"pl avcd
bv t hc or chest r a pr ogr am. For exampl c. i f an i nst r umcnt i s
to i ncl udc thc opti on of pl af i ng tones of di fferent frequeuci es. then the frequcncy
Noul d be suppl i ed as i t
"parameter"
val ue on a note stul ement for the i nstrument.
The fl owchart di agrams i n fi gure 3.l l a. h, and c hel p to i l l ustrate a si rnpl e poi nt:
A basi c i nstrument desi gn i n three di fferent MUSIC 4-type l anguages i s al mo:t the
same. l n each l anguage the output of the envel ope control uni t i s ted to the
ampl i tude i nput ofthe osci l l ator. and the resul t ofthe osci l l ator i s sent Io the output
of the i nstrument. In what fol l ows, we wi l l show the actual text that descri bes the
t hr ee i nst r ument s of t he f i gur e i n MUSI C l l , MUSI C
- 5.
and MUSI C 48F. r espec-
l i vel -v. The exampl es are i ntended onl v to demonstrate the maj or di fferences i n
\!' ntax among the Itnguages.
Codi ng of i nst r ument s i n MUSI C l l l anguage: r esembl es assembl y l anguage
programmi ng. The fi r:t and l ast statements ofthe i nstrument del i ni ti on, i nstr I and
cndi n, respecti vel y. mark the begi nni ng and end of i nstrument number l The
uni t generators are l i nen (l i near envel ope) and osci l (osci l l ator). The fi rst argumcnt
t r f l i nen. p5, i ndi cat es t hat t he ampl i t ude of a not e pl aycd on t he i nst r ument i s
\peci fi ed as the fi fth p-fi el d (p5) of thc note statem!nt. The subsequent argumcnts
of the l i ncn ri se ti me, durati on, and decay ti me are speci l i ed on the note state-
nrents as p6, p3. and p8, respecti vcl y. MUSIC 11 di sti ngui shcs between con-
trol rate and sampl i ns rate operati ons. Thc rcsul t of the l i nen operati on i s pl aced
i nto the storage l ocal i on desi gnated trv the vari abl e kl whi ch i s cal cul atcd at the
I C
I
c
e
) r
l .
l c
r c
i l n
) r 5
cn
t hc
i ru
e n:
t hc
) n. )
ccn
syr
76 Computer Music
A( 4)
cal
0f
ne
ca
o)
hl
FI GURE 3. 11 The si mpl e i nst r unenl of l i gur e 3. 5 pr ogr nmmed i n ( a) MUSI CI 11.
( b)
MUSI C 5. und ( t - ) MUSI C ] BF.
! l i r ' n
r 5 Ff . P3 Fg
l : : :
'
"' ""
EXAMPI-E 3.1 \' l tl sl c l l code for the i nstrument shorvn i n fi sure 3.l l a.
control rate. The posi ti on of k1 as the fi rst argument of the osci l uni t generator
causes the ampl i tude of the osci l to be control l ed bv thc output o{ l i nen. The
frequcncr of the osci l i s taken l rom p.l of the note stal !' menl . The thi rd urgument.
the numher of the wavetabl e used for the rvaveform of the osci l . i ndi cates that
storcd wavetabl e numbcr 2 wi l l be uscd. The out statement' s audi o rate argumcnt.
a2, causes the resul t of the osci l operati on to b!' sent to the output of l he rtrchesl ra.
. r , ,
". ] i , . r . . . r o. ", u. ",
r , "r u,
t sa 82 , Pi . B?. F 2. F2i I
ULT E2, 81:
ENtr ;
EXAMPI-E 3.2 MUSTC 5 code for the i nstrument shown i n fi gure
-1.l l b.
Among Max Mathews' goal s i n the desi gn of MUSIC 5 were for the l anguage
to be as si nrpl e to use as possi bl e, tbr the program to be computati onal l y effi ci ent,
and for i t to run on any general -purpose computer system wi th a FORTRAN com-
pi l er. The Ii rst goal was i mpl emented through user codi ng of al l statements, for both
orchesl ra and score. i n the same format: a mnemoni c fol l owed by arguments. To
enhance executi on effi ci ency, the structure of the executi on stage of MUSIC 5
as
e\
cn
ef
,h
\ p
fu
I t
( l l
E
.lt
g
I I
ir
E
rl
l ,
t
.1
I
Al2)
Synthe3ir f undamentals
causes each uni t generator to contri bute a number of successi ve outputs to a bl ock
ofoutput sampl es. The output bl ock i n MUSIC 5 provi des the means for i ntercon-
necti ng the uni t generators: the output ofevery uni t generator i s stored i n an array,
cal l ed an output bl ock, whi ch can then be appl i ed to the i nput of any subsequent
generator.
-fhe
INS and END statements serv!' essenti al l y the sarl c purpose i n ML)Sl C 5
as i n MUSI C 11 t o del i mi t t he def i ni t i on of t he i nst r ument . I n our NI LI SI C 5
exar)pl e, we cal l for i nstrument number 4 to be defi ned al ti me 0. The MUSIC 5
envel ope generator. ENV. has seven arl l uments. Thc fi rst i s the ampl i tude of the
envel ope-here suppl i ed by parameter 5 of the note statements for thi s i nstrument.
as i n our MUSI C l l exampl e. I n MUSI C 5 a si ngl e wavet abl e i s used t o st or e t he
shapes of t hc r i se. t he sl eady- st at e. and t he decav. ENV s second ar gumenl
speci l i es the waveti tbl e that i s to be referenced for these shapes. In our exampl e.
funcl i on number 1 i s Lrsed. The thi rd argument desi gnates 82 as the output bl ock
for tl .re uni t generator. Arguments 4. 5, and 6 arc sampl i ng i ncrements for the ri se
ti me. steady-state ti me. and decay ti me. respecti vel y. The actual val ues for thesc
wi l l come from p-fi el ds 9. 10, and 11 i n each note statcment. The l i nal argument of
ENV i s a temporarv l ocati on i n memory for stori ns. the current posi ti on i n the
wi tvetabl e. By conventi on, i t i s desi gnated here as the unused parameter Dumber
30.
The fi rst argument of OSC (osci l l ator), B2, i nterconnects the two uni t
generators through the output bl ock by passi ng the output of ENV to the ampl i tude
i nput of OSC. The second argument of OSC. i ts sampl i ng i ncrement, i s obtai ned
from P7 of the note statement. The output ctf the osci l l ator i s sent to output bl ock
82. as i ndi cated tr)' the thi rd argument. The stored wavel abl e for the osci l l ator. as
shown i n argument number 4, i s functi on number F2. The fi nal argument, P29. has a
functi on si mi l ar to the fi nal argument of ENV-storage for a poi nter
to the current
l ocati on i n the wavetabl e. The OUT statement i ndi cates that the contents of bl ock
82 are to be assi gned to output bl ock B I as the output tbr the i nstrument. Chapter
I of Max V. Mathews' The Technol ogv of Computer Masi c' serves as a manual of
t he MUSI C 5 l angLr age.
X=LI NENS( AL r : , . Ai 4 ) )
X=oSCI L( x, AL t 2, . 1
' Ai
13)
aAL! f I CNC
r , ( )
R!TURN
ENC'
EXAI \ , t PLE 1. 3 MUSI C- l BF code f or t he i nst r ument shown i n l i gur e 3. l l c.
Cl ear l y. MUSI C, l BF has a ver y di f f er ent appear ance f r om MUSI C l 1 and MUSI C
-5.
A Musi c 4BF i nstrument desi gn i s u,ri tten i n FORTRAN and l i nks thc uni t
generators as functi ons and subrouti nes. The i nterconnecti on of uni t generators i s
expresscd al gebrai cal l y, as shown i n the exampl e. The use of an array. A. i s a
mcans of stori ng val ues obtai ned at i ni ti al i zati on ti me oi i r note. In the fl rst l i ne
A( 2) r epr esent s t he ampl i t udc i nput t o t he L( NENS ( l i near envel ope uni t ) . A( . {) i s
the fi rsl l ocati on of a group of array l ocati ons contai ni ng other i ni ti al i zati on-ti me
val ues such as ri se ti nre, durati on. and decay ti me. The i nput of X to the hrst
argument of OSCIL effects the l i nk between the output of LINENS and the
77
78
Computer Music
anr pl i t udc i nput of OSCI L. , As i n t hc pr evi ous exampl cs. t he second ar l ument of
OSCI L i s a f r equcncv i nput . and t he t hi r d ar gument i s t hc number ol t hc st or ed
wavct abl e. The f our t h ur gumcnt . A( l - l ) . i s t hc phi t sc of t he osci l l at or . pcr f or mi ng a
functi on si mi l ar to thc l i nal argunre nt of l v1Ll SI(. 5 s OSC. CALL MONO i nvokes a
subrouti ne that pl aces the resul t of the osci l l ator. X. at the output of the
i nstrument. Chapter l l of Hubert S. Howe' s El ectrnni c Mri sl c Sl rrthesl sr contai ns a
dct ai l ed descr i pt i on of t he MUSI C. l BF pr oer am and i t s use.
1. 7 Addi t i veS; - nl hesi s
- [ hc
si mpl e i nst r ur ncnt t hat was sho$' n i n t i gur cs 3. 5 and . l . l l i s t he f i r st
confi gurati on uscd to svnthcsi zc musi cal sound. It i s bascd <rn a si mpl i l i ed
Hcl mhol tz model of nrusi cal sound. whi ch consi sts of a waveform at a constant
frequencl ' cncl oscd i n an envel ope. Thc spcctral composi ti on of the waveform i s
ti etermi ned from the desi red stead-t--statc spectrum of thc tone. someti mes taken
frLrnt a Fouri er anal _' -si s.
l hc' sound pr oduced br t hi s i nst r ur nent di l t cr s f r om nat ur al sound i n t wo
i mpor t ant r espect s. Fi r st . t he ampl i t udes of al l of t he spect r al componcnt s ar e
var i cd cqual l y bv t he cnvel ope. so t hat t he ampl i t udes of t he component s r cl at i ve
to each othcr do not change duri ng thc course of the tonc. Thus. the sound l acks
i ndependent temporal evol uti on of the harrnoni cs. an i mportant characteri sti c of
natural sound. Second. al l the spectral components arc exact-i nteger harmoni cs of
the fundamental frcquencv. not the sl i ghtl y ni stuned parti al s that often occur i n
a crr r,r s t i ca l l r- genefttted sOunds.
FI Gt JRE
3. 12 Basi c conl i guri l t i on f or addi t i vc synt hesi s. 1. he anl pl i t ude and f requcncv
i Dput s of cach osci l l at or deri ve f rom i ndependenl
f unct i on generat ors. l n many cascs t ne
f unct i on generat ors t akc t he f orm of envcl ope
Jl enc, . t t ors or osci l l at ors.
c
, l
d
l c
i i r c
i l c
cks
; of
sot
r i n
ucncY
e\ t hc
Synthesis Eundamentals
As expl ai ned i n chapter 2. each spcctral component of a sound can be
represented by i ts own i ndependent ampl i tude and frequency functj ons. The
synthcsi s of a tone based on thi s model (ti gure 3.12) requi res a separal e si Dusoi dal
osci l l ator for each parti al . wi th the appropri ate ampl i l ude and frequencl functi ons
appl i ed to i t. The outputs from al l the osci l l al ors are added together to obtai n the
compl el e soul d. Hence. t hc name addi t i ve sr r t hesi s i s used t o desi gnut e t hi s
I cchni que.
Addi t i ve svnt hesi s pr ol i des t hc nl usi ci an wi t h maxi munr f l exi bi l i t y i n t hc t ypes
of sound that can bc svnthesi zed. Gi ven enough osci l l ators. anv sct of i ndependent
spectral components can be s)' nthesi zed. and so vi rtual l y ant sound can be
gencr at ed.
The ampl i tudc and frequenc! functi ons can be obtai ned from thc anal vsi s of
rcal sounds as dcscri bcd i n chapter 2. The name
' Fouri er
rccontp.rsi ti un" i s
sonl el i mes used to descri be thi s svnthesi s from anal ysi s. becausc i t can he thought
of as the reconsti tuti on of the ti me-varvi ng Fouri cr components rtf a sound.
Addi ti ve s-,-nthesi s has proven capl bl e rrl real i zi ng tones that are i ndi sti neui shabl c
l r r ) m r i Jr l t , ' nc. hr . k l l ed mu\ r cr l n\ .
'
Of cour se. t he i nst r umcnt desi gner i s not r est r i ct ed t o usi ng f uncl i ons obt ai ncd
i r on anal vses. When choosi ng f unct i r r r r s. hr . r *cr cr . i t i s hel pl ul l o l r ave a
knor vl cdgc of t he behavi or of f unct i ons t hat descr i be nat ur al sounds. For exampl e.
i n mr nl l coust i c i nst r ument \ t hc hi qher har moni cs at t ack l ast and
( l ccav
t i r r t .
Knorvi ng thi s. a musi ci an mi ght choosc to synthesi ze an unusual sound usi n-g a set
of l uncti ons wi th the opposi tc characteri sti c.
New. natural -soundi ng functi ons crn be generated by i nterpol ati ng between
scts of functi ons that have been deri ved by anal _' -si s." Musi ci ans can usc thi s
techni que to transform gradual l y thc sound of one i nstrument i nto thal of another.
The normal wa-v to i nterpol atc betwcen two functi ons i s to take a wei tl hl ed avcrage
bet u' een compar abl e poi nt s on each f uncl i on. For exampl e. t o geoer at t a f unct i on
of l i mc t hat i s 30- t i of t he wav bel wr en f uncl i on I ( Fl
)
and f unct i on 2
{F2) .
l he nr : w
i uncti on i s l i rrmed as the sum of 7{)l i rrf Fl and
-30?
ol F2. Horvc' rer. appl vi ng thi s
r L' chni que t o t \ o ampl i t ude f unct i r : r ns l hot peok at di f f cr cnt l i mes cr eat es a ne$
l unct i on r vhi ch l acks a si ngl c. shar p maxi mum. I nst ead. t he pcak ot t hc f unct i on
Lrroadens. spread bet\\,een ti mes of the peaks of thc ori gi nal fr.rncti ons. Upon
svnt hesi s. t hi s i nt r oduces an addi t i onal t i mbr al el emcnt t hat i s unr el at ed t o ei t her of
t he or i ci nal sounds. To pr esenc a cl ear . : i ngl e maxi mum and hence. nor e t i mbr al
si mi l ar i t y wi t h t he or i gi nal sounds. t hc t i me at $hi ch t he maxi nr un of t he ncu'
functi on occurs i s i nterpol atcd betreen the ti mcs of thc ori gi nal functi ons. For
exanr pl e. suppose F1 pcaks at . l secot t ds and F2 peaks at . l 5 seconds. Thc f unct i on
thal i s 30?i Of the rval ' between Fl and F2 woul d pcak at . 7 x . 1 + . i x . l 5
=
. 1 1 5
seconds. Havi ng establ i shed the ti me ol the maxi mum. the al tack porti on of the
neu funcl i on i s i nterpol ated bctwccn the attack porti ons of the ori !l i nal l uDcl i ons.
Si mi l ar l r ' . t he deca! por t i on i s i nl er pol at ed usi ng t he decay por t i ons r r f t he
or i gi nul s.
Addi t i ve svnt hesi s pr oduces hi {h- qual i t r sound but r equi r es a compi l r at r vcl v
l i rge amount of data to dcscri hc a sound. becausc each of thc nl i tn\ osci l l ators
rcqui res two functi ons. A further conrpl i cati on ari scs because a gi ven set of
functi ons i s normal l y useful onl y for l i mi ted ranges of pi tch and l oucl ness. If a set i s
determi ned by anal _"-si s for a spcci l i c pi tch. then i t rvi l l producc thc sound qr,ral i t-v of
the ori gi nal source i n onl ! a smal l pi tch i nterval around that poi nt. Anv formants
79
r st
ed
t nt
i s
cn
80 Computer Mustc
present i n the spectrum wi l l move di rectl v wi th fundamental frequency. Thus.
much of the ti mbral si mi l ari ty betwcen toncs of di fferent pi tches wi l l be l ost. l n
addi ti on, the functi ons are hi ghl y sensi ti ve to dynami c l evel . so that a set deter-
mi ned for a ntezzo
forte
wi l l produce
an unreal i sti c pi uni ssi mo. To ful l y real i ze
the benel i ts of addi ti ve synthesi s. i t i s necessar;- to have ei ther a l argc l i brary of
functi on sets or i r compl ex scheme for al teri ng a functi on set on the basi s of pi tch
and ampl i tudc duri ng performance.
An advantagc of addi ti ve svnthesi s i s that i t provi des compl ete. i ndependent
control Ovcr the behavi or of each specl ral conrponent. However. such a l artte
number of control s on the ti mbre can make i t di ffi cul t tbr the musi ci an to achi eve a
parti cul ar sound. A practi cal di sadvantage of addi ti ve synthcsi s i s that i t requi res a
l arge nrrmber of uni t generators. When svnthesi zi ng compl ex sounds. i t i s not
unusual to empl oy ten or more osci l l ators wi th thei r associ ated funcl i on gcnerator\
i n the s),nthesi s of a si ngl c voi ce. Thi s di ffers from the synthesi s techni que\
pr escnt ed i n subsequent chapt er s. whi ch use f e*cr uni t
gener at or s.
3.8 l Vl odul ati on
Modul ati on i s l he al terati on of the ampl i tude, phase, or frequencl ' of an osci l l ator
i n accordance wi th another si gnal . Modul ati on has bccn used for manl
1' el rs
rn
radi o communi cati ons to transni t i nformati on efl i ci entl y. Musi ci ans have expl oi t!d
vari ous modul ati on techni ques i n el ectroni c musi c to create di sti ncti ve sounds
effi ci cntl 1".
The osci l l ator that i s bei ng modul ated i s cal l ed the carri er osci l l aror. Whcn
there i s no modul ati ng si gnal , i t generatcs a conti nuous wavcform cal l ed the carri er
wave. When modul at i on i s appl i ed. t he car der wave i s changed i n some uay. The
changes are i n sympathy wi th the modul ati ng si gnal , so thar the ourput of the
carri cr osci l l ator may be thought of as a combi nati on of the two si gnal s. The nature
of thi s combi nati on depends on the nrodul ati on techni que used and wi l l be
exami ned bel ow.
The spectral components of a modul ated si gnal are cl assi l i ed i nto t\r!o types:
car r i er component s and si debands. f he f r equency of a car r i er comp( r nent i \
deternti ned onl v bv the frequency of the carri er osci l l ator. Thc frequency of a
si deband i s determi ned by both the carri er frequency and the frequencv of the
modul at i on.
3.8A Amplitude Modulation
Thcr e ar e t hr ee mai n t cchni ques of ampl i t ude modul at i on:
' cl assi cal ' or
"st r ai ght '
ampl i tude modul ati on. ri ng modul ati on, and si ngl e-si deband modul ati on. The
l ettcrs AM are nost often used to denol e the fi rst tvpe.
Fi gur e 3. 13 di agr ams an i nst r ument t hat i mpl ement s cl assi cal ampl i t ude
modul ati on
(AM).
The carri er osci l l ator has a constant frequency off, and the
modul ati ng osci l l ator a frccl uency of
J;,,.
For thi s exampl e. the waveform of each
osci l l ator i s a si nusoi d. The ou(put from the modul ati ng osci l l ator i s added to a
val ue that expresses the ampl i tude the carri er osci l l ator woul d have. i f there were
no mr)dul ati on. l ' he ampl i tude of the modul ati ng osci l l ator i s expresscd as a
Synthesis f undamentals
m.AMP
FI GURE 3. 13 Si mpl e i nst r ument
output \\i tvcl orm (b).
b)
i mpl ement s ampl i t ude modul al i on and
8l
) r
l n
:tl
ht '
fhe
ude
t he
ach
: oa
,'ere
ts il
AMP
*AMP
(a) whi ch
l t s
_' l )
cr
ne
ne
re
be
i s:
i s
f a
he
m
_T
l - r
FI(i LJRE 3.1,1 Spectrum of the Ai \,1 si gnal produced h\ the i nstrunrenr of l i ture 1.13.
proporti on of the unnrodul ated ampl i tude of the carri er osci l l ator. Thi s proporti on
i s denoted by the vari abl e nr. u,hi ch i s cal l ed the modul ati on i ndex. When rr-0.
there i s no modul ati on and thc carri er osci l l ator gencrates an unmodul ated
si nusoi d. When m i s l ar ger t han zer o. t hs car r i cr wave r vi l l t akc an cnvel ope wi t h a
si nusoi dal var i at i on ( t i gur e 3. 13b1. Whcn l r : 1. t he ampi i t udc r r f t hc modul at i ng
osci l l ator equal s thc unnrodul ated rnrpl i tude of thc carri er osci l l ator and l 00cri
modul at i on i : sai d t , r t ake pl r r ce.
When both the carri er and modul ati ng wavcforms are si nusoi ds. thc spectrum
of an AM si gnal (l i gure
3.1.1) contai ns energy at thrce frequenci es: the carri er
82
Computer Music
frequcncy (1.) and two si cl ebands (.r, +r,, and
I
/,,,).
l hc arnpl i tude of thc
component at thc carri er frcquency does not varv rvi th the nrodul ati on i ndc' x.
-l ' ht'
amp) i t Lr dc of cach si dcband i s a f act or ol l l 2 l css l har ) t he l nr pl i t ude r r f t he car r i cr .
showi ng that the modul ati or) proccss spl i ts thc cnergy bet\\ecn upper atrd l orvcr
si debands. For exampl e. whcn nr =1. t hc si debands r vi l l have one- hal f t hc ampl i -
tucl e of the carri er. and thercforc be 6 dB bel ow the l cvel of the crrri ur
t ht trcquencv Di the modul ati on tl ctcrnri ncs hrrw a l i sl sncr percei vcs l he Al vl
sound. l f
. f , , ,
i s l cs\ t han about l 0 Hz. l he ear wi l l t r ack t he i ndi vi dual ampl i t ude
vari ati ons. When.f,,, i s grcater than 10 Hz. but smal l cnough that the carri cr and
both :i cl cbands fal l wi thi n th(. same cri ti cal band. thc tonc rvi l l sound wi th a
i oudness pr opor t i onal t o l hc aver age r nr pl i t ude of t hc modul at i ng l r avef or m. A
val ue of
f,,,
whi ch exceeds onc-hal f thc cri ti cal bal d causcs thc si debani l s to bc
per cei ved i ndi vi dual l l . cr eat i ng t he sensat i on of addi t i onal l oudncss.
Musi ci ans have usecl ampl i tude modul ati on to create cl cctroni c
' tremol o" bv
usi ng a smal l nr odul at i on i ndex and sub- audi o modul at i r g f r equeno. Whcn t he
modui at i on i ndex i s l ar ge ( c. 9. . l r =l ) . a mar kedl . v pul sat i ng sound wi l l be
produced.
3. l i B Ri ngl \ l odul at i on
When modul ati on i s appl i cd di rectl y to the ampl i tude i nput of a carri er trsci l l al or.
wi t hout bei ng added t o a val ue r epr cscnt i ng t he r mpl i t udc of an unmodul at cd
carri cr. the pr()cess created i s kno*n as ri ng nr,.rdul ati on. Other namcs l or i t are
bal anced modul at i on and doubl e- si deband
( DSB) modul at i on. Fi gur e 3. 15 i l l us-
tratcs thc si gnal t' l ow chart for an i nstrument $hcrc one osci l l ator ri ng nl odul ates
anot hcr . The ampl i t ude ol t he car r i er osci l l at or i s det er mi ned onl v by t he
modul at i ng si [ na] . so t hal uhcn t her e i s no nodul i t t i on. t h! - r e i s no ! i r r r i er \ \ ' a\ L'
:
. .
Fl ( i Ll RE 3. l 5 Si nr pl c i nst r unt nt \ \ ' hi ch r mpl cnr cnl : ' r i ng nr odul at i on of one osci l hl or by
anot her .
t c
!l
l c
rd
-ls-
t es
he
l Ve
r hr
Synthesis Fundanental6
83
L, , t l , l t , ' , , ,
SI GNAL I NPUT
l ' "
U
I
.)\
nc
he
) r .
cd
t r c
f"
OUTPUT SPECTRUM
FICURE
-1.l 6
Al terati on of thc specrrunr of a si gnal bi - ri ng modul ari rrn.
I NPUT 1 I NPUT 2
FI GURE
- 1. 17
A nr ul t i pl i er as i r gener al pur pose
r i ng modul ar or .
l nd. i n turn. no output from the i nstrument. Thi s i s a notabl e departure from the
l ol rfi gurati Dn used i n the ANI tcchni que dcscri bcd aborc.
Al t hough r i ng nr odul at i on oper at cs on t he t mpl i t ud! ' of t hc car r i cr . t t l s m( ) st
(rften
used to al tcr the frequcncy of a sound. When both the carri er (f,) and
nrodul ati ng (1,,,)
si gnal s arc si nusoi dal . the spectrunr of the modul rtccl si gnal
contai ns onl y two frcquenci cs:
f<+f,,,
and
l ,-t' ^.
t" othcr uords. thcrc arc
.rdcbands but no carri cr. Bccause nei thcr
f,
nor
f,,,
appears di rcctl v i n the
\[(' ctrum.
the frequencv of the souncl can bc qui tc di ftcrcnt. For exampl e. i f
' . . , - - 1. 10 Hz and
/ , : 16l
Hz. t he r esul t i ng spcct r um conr r i ns ener gr onl y at 179 Hz
.i nd 701 Hz. frequenci es that i tre not harmoni cal l l rel ated to the ori gi nal s. or to
. . r ch ot her . l f t hc ampl i t ude of t he modul at i ng si gnal i n f i gur e 1. 15 i s, , 1. bot h
.r.l cbands
have i rmpl i tudes of ,,1,' 2.
The fr(' quenci es i n a sound that i s appl i ed di rectl ) to the ampl i l ude i nput of an
,\\ci l l ator (fi gure
-j .16)
are changed bv ri ne modul ati on. Supposc a spcech sound
' .r
rth a fundamcntal frequency of 100 Hz ri ng modul ates a si nusoi dal osci l l ator I' ri th
r l rcquencv of I l l .l Hz. The sound that emcrges contai ns rhe sun and di fl ercnce
' ' cl ween
cach harmoni c of thc speech and 112-3 l l z. Thus. thc spectral compr)nent
:h.r{ was thc fundamental of thc speech sound i s <rutour ar both l (12-1 Hz and 1223
I NPUT SPECTRUM
84
Computer Music
t
I NPUT 1
I NPUT 2
OUTPUT
FIGURE 3.l l l Ri ng modul ati on of rwo si gnal s ro produce a densc specrrum.
Hz. the former second harmoni c (ori gi nal l y at 200 Hz) appcars at 92j Hz and l 32j
Hz. and so on. The formerl y harmoni c speech now sounds i nharmonl c and mav nol
be i nt cl l i gi bl e.
When two si gnal s are combi ned by mul ti pl i cati on. the resul t i s ri ng modul a-
ti on. Thus. the mul ti pl i er shown i n f,gure 3. 17 i s a general -purpose ri ng modul ator.
Two si gnal s are often combi ned thi s way for the purpose of frequency al terati on.
Suppose that two si newaves. wi th ampl i tudes A1 and A. and frequenci es
fi
antl
f..
respecti vcl v. are mul ti pl i ed together. The resul ri ng spectrum wi l l conrai n the
fi equenci es of
/r-f,
and
t+/,.
and the ampl i rude of each component wi l l be
AtxA.l 2. Observe that i f ei ther si gnal has an ampl i tude of zero. therc wi l l be no
output from the modul ator. Composers such as Jean-Cl aude Ri sser (see secnon
3.l 2) and James Dashou
T
have used thi s form of ri ng modul ati on for the creati on of
chordal structures.
Thc mul ti pl i cati on of two compl ex sounds produces a spectrum contai ni ng
frequenci es that are the sum and di fference between the frequenci es of each
component i n the fi rst sound and those of each component i n the second. If thcre
are p components i n the fi rst sound and
4
components i n thc second. as many as
2p./ conponents can appear i n the output. Thus. mul ti pl i cati on can be uscd to
create dense spcctra. For exampl e. i f two si gnal s. each wi th 4 components. are
mul t i pl i ed t oget her ( f i gur e 3. 18) . t he r esul t i ng sound wi l l have as manl as 32
components. Therc woul d be fewer components i f the two si snal s were harmoni -
cal l v rcl ated. In thi s casc some of the si bebands url l hJ\e th-e same frequenci es.
reduci ng the ovcral l number of observed spectral componcnts. To avoi d al i asi ng. i t
. . 4
. \ 1
- L_r
Synthe6l5 Furdarnentals
I NPUT
f.
OUTPUT
FIGURE 3.19 A frequency shi fter. i ts i nput. and i ts outpul .
shoul d be noted that the hi ghest frcquency produced by thi s process i s the sum of
thc hi ghest frcquency contai ned i n the hrst sound and the hi ghest i n the second.
- j . 8C
Si ngl e- Si deband Modul at i on
Si ngl e-si deband modul ati on resul ts i n frequency shi fti ng. When a sound i s passed
through a frecl uencv shi fter (fi eure
3.19). a constant
(f,.)
i s added to the frequency
of each cQmponent i n the sound. Shi fti ng i s l l ot the same as traDsposi ti on. because
shi fti ng destroys the rel ati onshi ps between the i nterval s i n the sound. I-i ke ri ng
modul ati on. shi fti ng i s a wav to make a harmoni c spectrum i nharmoni c. but unl i ke
ri ng nrodul ati on, i t does not i ncreasc the number of components i n the sound.
In frequency shi fti ng. the i nput si gnal modul ates ! carri er osci l l ator l ocated
i nsi de the shi fter wi th a modul ati on techni que known as si ngl e-si deband (SSB)
modul ati on. Whereas ri ng modul ati on produces si debands on both si des of i he
carri er, SSB modul ati on, i n theory. produces onl y upper si debands (i .+/;,,) or onl y
Iower si debands (f,-f-), as sel ected by the user. The cffect i s to add or subtract
every component of the i nput si gnal from the frequency of the carri er osci l Lator
( / .
)
SSB modul ators are fai rl y compl i cated techni cal l v. It i s not possi bl e to bui l d or
proSram one that real i zes perfect SSB modul ati on over a wi de range of i nput
frequenci es. The output from actual frequency shi fters contai ns some energy at the
opposi te si deband. good shi fters suppress the opposi te si deband throughout the
audi o range to at l east
:10
dB bel oq the si deband desi red.
-1.8D
Vibrato Simulation by Frequency Modulation
When a modulating signal is applied to the frequency input of a carrier oscillator.
frequency modul ati on occurs. Vi brato, a sl i ght waveri ng of pi tch. can be si mul ated
usi ng the i nstrument i n l i gure 3.20. The carri er osci l l ator generates a tone at the
speci hed ampl i tude and frequencv
(1,).
anrl the vi brato osci l l ator vari es thal
frcquency, at the vi brato rate. by a maxi mum amount equal to the vi brato $' i dth.
85
3l l
n( l I
ul i t -
l or
l () n
i i nd
t he
l o(
I n( )
t i on
I n ol
ni ng
3ach
l l crt
l \ i r :
d t ()
. ar c
rs 3l
r on i '
r c i es.
nu. i t
SI GNAL I NPUT
FREQUENCY
SHIFTER
86
Computer Music
FI(;URE
-3.20
Si mul ati on of si mpl e. peri o<1i c vi brato.
-l hus.
thc i nstantaneous frequencl ' of the carri er osci l l ator changes on ever!
sampl e. varvi ng betu,een.f, pl us the vi brato wi dth and
r.
nti nus the vi brato $' j dth.
Its averagc frcqucrcv i s
f,.
In nearl y every sound the vi brato u' i dth vari cs wi th the averi tge frcqucncv. l t i s
usual l v spcci fi cd as a proporti on of thc fundamental frcquencv of the tone and i s
ordi nari l y no more thi tn a few percent ol l .. In order tbr the frequcncy modul ati on
to bc percei ved as vi brato. the vi brato rate must bc rcstri cted to frequcnci cs bel ou
the audi o range. The vi brato found i n natural sounds can be qui te compl cx. It often
changes duri ng the course of i t tone and frequentl y contai ns a certai n anount ot
randomness. Chapter 6 wi l l di scuss one of the more compl i cated forms of natural
vi brato. that of the si ngi ng voi ce.
When a l arge vi brato wi dth at an audi o ratc i s uscd. the aural effect i s nct l onger
that of a si mpl e vi brato. In thi s casc. frcquenc-v modul ati on i s a powerful synthcsi s
techni que capabl e of produci ng a u,i de vari cty of ti mbres. Chapter
.l
wi l l cover
appl i cati ons of frequency modul ati on s.vnl hesi s i n detai l .
3.9 Noi se Generators
An osci l l ator i s desi gned to produce a peri odi c waveform u' i th wcl l -defi ned spectral
componcnts. Thc spectrum i s a di scrctc spectrum; that i s. thc energy i s found at
speci fi c frecl uenci es. The opposi te of a di scrctc spectrum i s a di stri buted spectrum.
i n u,hi ch cncrgv cxi sts evcrywherc wi thi n a range of frequenci es. Most of thc noi se
VI8 WIDTH
Synthesis Pundamentals
87
0
-20
- +o
- 60
cr )
I t n.
TIME
o) b)
FI(IURE 3.21
(a)
Segmcnt of l he wavel orm of whi te noi sc and
(b)
thc spcctrum of i deal
*hi t e noi se.
sounds found i n nature have di stri butcd spectra. and thus al gori thms dcsi gncd to
qenerate
di stri butcd spcctra are cal l ed noi sc gencrators.
Certai n phcnomena have the characteri sti c that thei r repeated occurrence.
cven under the same sct of condi ti ons. rvi l l not ahvays l ead to the si tme resul t.
l vl crnbers of thi s cl ass are cal l ed random phcnomcna. Even though the exact
outcome carnot be prcdi cted. they exhi bi t a certai n amount of stati sti cal regul ari ty
that can be used to dcscri be them and to prcdi ct thc probabi l i tv of anv gi ven
occurrencc. Thc stati sti cal characteri zati on of a random si gnal i s used to determi ne
i ts frcquencv content. (Addi ti onal
cxpl anati on ol random processcs can bc found i n
chapter l l .
)
In sound svnthesi s. randomness i s used to gcncratc di stri buted spectra. l n the
uaveform i n l i gurc 3.21a. thcre appears to bc a random di stri buti on of the
i mpl i tudc of the sampl e val ucs. Thi s i s a pi cture of a scgmcnt of the waveform of
u' hi te noi se. The ampl i tudc of u' hi te noi se i s charactcri zcd by a range-the i nterval
ui thi n whi ch the maxi mum and mi ni mum sampl c val ues occur. l n the fi gure. the
rangc i s - l to + 1. Noti ce that. unl i ke a peri odi c wavcform, a rcpeati ng pattern of
\ampl es cannot bc i ndcnti fi ed. Thus. si gnal s of thi s type are rettrred to as
aperi oci i c. Whi te noi se has a uni forml y di stri buted spectrum as shown i n fi gurc
-.i .21b.
Between any two frequcnci es a fi xed di stance apart. therc i s a constant
tmount of noi se power. For i nstance. there i s the samc amount of noi se power i n
the band between 1(X) Hz and 200 Hz. as thcrc i s between 7900 Hz and 1t000 Hz.
Whi te noi se makcs thc
' hi ssi ng"
sound often associ ated u' i th noi se generated by
cl cct r oni c neans.
The uni t generator that produces nearl y whi tc noi se i s often cal l ed RAND and
has an ampl i tude i nput. l ts svmbol i s shown i n fi gure 3.22a. The ampl i tudc i nput
l ets the range of the permi ssi bl e output sampl e val ucs and hcncc the ampl i tudc of
thc noi se. l f a val ue AMP i s appl i ed to the i nput. the noi se wi l l range between
AMP and +AMP.
One woul d expecl an al gori thm desi gned to generate whi te noi se si mpl ! to
draw a random number on each sampl e. Thi s makes a good, but not perl ect.
\ hi te-noi se source. The spectral di stri buti on of such a generator i s shown i n fi gure
1.22b. It devi ates sl i ghtl y from a uni form di stri buti on because of a frcquencv bi as
i nherent i n the process of sampl e generati on. Thc actual spectral di stri buti on .9(/)
at frequency
/
i s gi ven by
si n
l r ( 1/ / ) l
r\fl f
.)
i 1 r \
Ll ls
i on
l ou
: t c n
t ol
u ri l l
l ger
Lcsi s
)\' er
I t r al
dat
' um.
LOI SE
Sas
88 Computer Music
0
- 24
- +o
- 60
o)
b)
FIGURE 3.22 Di gi tal whi te noi se
Benerator
(a) and i ts specrrum (b).
At hal f the sampl i ng rate. the power i s down l ess than 4 dB from a uni form
di stri buti on.
Whi te noi se has a vcry l arge bandwi dth. Someti mes i t i s desi rabl e to narrow the
bandwi dth by reduci ng the amount of hi gh-frequency cnergy i n the noi se. A noi se
source wi th most of i ts power at l ow frequenci cs wi l l make a rumbl i ng sound. An
al gori thm that synthesi zes thi s ki nd of spectrum draws random numbers at a rate
l ess than the sampl i ng rate. The uni t generator that accompl i shes thi s i s often cal l ed
RANDH
(fi gure
3.23a) and has two i nputs: arnpl i tude and the frequency (/a) at
whi ch random numbers are drawn. (On some systems. the frequency i s not
speci fi ed di rectl y, but by a number proporti onal to
/R.
i n the same wav that
sampl i ng i ncrement i s proporti onal to the frequency of an osci l l ator.) Choosi ng
random numbers at a rate l ower than
f,
i mpl i es that a random number i s hel d for a
few sampl cs. unti l the next one i s to be drawn. For exampl e. i f/,
=
40 kHz and/' .
=
4 kHz. the al gori thm chooses a random val ue. outputs i t for the next 10 sampl es.
and thcn chooses another.
When noi sc i s generated by thi s process. man_t- of the sampl es are rel ated to
each other because thei r val ue i s the same as the previ ous sampl e. Thi s rel atedness
reduces the noi se power i n the hi gher frequenci es. The l ower the frequency
/R,
the
smal l er the amount of hi gh-frequency energy that wi l l be contai ned i n the noi se.
0
- 20
- 40
- 60
o) b)
FI GURE 3. 23 Di gi t al noi sc gcnerat or (a) i n whi ch noi se sampl cs arc generat ed at a rat e
l owcr t han t he sampl i ng rrt e and i t s spect rLl m (b).
I
AMP
I
@
I
_r l
- r.i
h
:
| :
2
I
2
Synthesk Fundamental3
Thus.
/a
can be thought of as a control on the "bandwi dth" of the noi se. Fi gure
3.23b i l l ustrates the spectrum when
/p=/,/6.
The shape of the spectrum i n the
gcneral case i s gi ven by
s{D
:
:I-l'-(t&ll .
sin
h(ti)l
TlflfR) r(f/f
,)
A vari ati on on thi s techni que. one that provi des noi se spectra wi th even better
attenuati on of the hi gh frequcnci es, i nvol ves i nterpol ati on. As before. random
numbers are drawn at a rate (l ;) that i s l ower than the sampl i ng rate. Instcad of
hol di ng the val ue of the l ast random number unti l the next one i s drawn. the
sampl es i n bctween draws are i nterpol ated l i nearl y between successi ve random
numbers. Thi s can be vi sual i zed as connecti ng successi ve random numbers wi th
strai ght l i nes.' [he uni t generator that performs thi s al gori thm i s often cal l ed
RANDI (fi gure 3.2,1a).
Fi gure 3.2.1b i l l ustrates the spectrum of such a noi se generator whenl ,l
:1,i 6.
Obser ve t he di mi ni shed amount of hi gh- f r equency' encr gy. Thc gcncr al shape of
thc spectrum i s gi ven by
St , t
=] @
2.nfi s si n
ft\f
l f
s)l
It i s possi bl e to real i ze noi ses wi th other types of spectral di stri buti on such as l f
noi se
(see
secti on 8.l F), where the spectrum i s di stri buted i n i nverse proporti on to
the frequenc!. Other techni ques. such as the one proposed by Si cgcl , Stei gl i tz. and
Zuckcrman.s are avai l abl e for generati ng random si gnal s wi th speci fi abl c spectral
densi t i cs.
How does the computer. whi ch i s desi gned to store and proccss numbers
accuratel )' and rcproduci bl y, obtai n sccrni ngl l unpredi ctabl c random numbers?
One u,ay i s to si rmpl c an cxtcrnal random phvsi cal process such as thermal noi se.
but thi s requi res addi ti onal hardrvare. A l ess expensi ve and more commonl l used
approach i s to empl oy an al gori thm cal l ed a pseudo-random numhc' r gcncr"roru
$' hi ch produccs a scquence of numbers that sati sfv most of the cri teri a for
randomness. wi th the notabl e excepti on that the sequence rcpeats i tsel f. It i s
possi bl e to make the peri od of the sequence so l ong that for most purposcs i t can be
consi dcred random. A pseudo-random number generator actual l y creatcs a
89
nc
i sc
t l ! '
cd
at
r ol
hi t t
i ng
' r a
?
LCS.
I t o
I CSS
t he
i se.
0
- 24
- 40
- 60
o) b)
FIGURE 3.2.1 l nterpol ati ng noi se generator (a) and i rs specrrum (b)
AMP
f
2
90 Computer Music
di scrcte harmoni c spcctrurn. but the spectrum i s str extraorcl i nari l -\ dcnsc that- i or
most nl usi cal appl i cati orrs. i t i s i ndi sti ncui shabl e from a trul v di stri buted one.
Psstrdo-ri rndrrrn nunl l )cr cencr torr {enerxl l -v use the mos! reccntl -v tarerntcd
r andom number as a basi s f or cal cul at i nq r hc next . As a r esul t . most al gor i t hms gi ve
thc user the opti on r' rf spcci fvi ni t a
"seed
val uc. Whcn a sced i s speci l i ed. thc l i rst
r andom number wi l l bc cal cul at ed on t he hasi s of t he seed val ue.
' f her cf or e.
s(i rrti ng from the samc seed rvi l l uhruv: ri el d thc sanrc sequcncc of randrrm
nurnbers. Thi s mcans that two di ffcrent computer runs i nvol vi ng the generati on ol
oscudo-randont numbers can ftc nrade 1r) hl re cxl rctl v thc samc rcsul l s.
3.10 Osci l l ators wi th Vari abl e Sampl i ng Rates
Thc method tl cscri bed i n secti on 3.3 for i mpl emcnti ng an osci l l ator uscs a l i xcd
sl nrpl i ng ratc-i r :rnpl i ng rai e that remai rs unchangetl regardi ess of the fre-
quencv of osci l l at i on. Anot her t echni que uses i t var i abl e sampl i ng r at e. so t hat t he
sanr pl i ng f r equcncr r ar i cs wi l h t h! - f r er l uenc\ oi Osci l l al i on. l n t hi s met hocl . t he
sampl i ne i ncr emcnt i s set t o a f i xcd i nt eccr val uc ( usual l v one) and t hc sampl i ng
f r cquency- i s scl ect cd t o
gi ve
t hc appr opr i at e f r equcncy of osci l l at i on. To obt ai n t he
fre<l ucncv
l j ,
from a va ri a bl e -sa rnpl i ng-rate
(VSR)
osci l l ator that rcfcrences a
\ \ ' a\ et abl e cor t ai ni ng N ent r i es usi ng a sampl i ng i ncr cnr ent oI onc. t he r anpl i ng
frequencv must be
I n most vSR osci l l i r t or s- t ( ) nes i n t hc hi ghcr oct avcs ar c s\ nt hesi zed bv choosi ng r he
sanpl i ng i ncremcnt equal to a powcr of 2. However. the sarnpl i ng frequenc-v" sti l l
\ i r c\ Jm( r nu nr r l ( \ i n J
i I i \ cn
r cci \ t cr .
When t he sanpl i ng r at e vl r i es. so does t hc Nvqui st f r equencv. i mpl r i ng t hat
t hc l or v pass t i l t er on t hc out put of t hcDr Aconr cr t er shoul dhavcamovahl ccu( of f
f r cqucncv t o el i mi nat c comp( ) nent s above hal f t he sampl i n- q r at c. Wi t hout t he
movabl c f i l t er . l ou' pi t chcd t oncs gcncr at ed bt a VSR osci l l at or hi we an unmi st i r k'
abl c. hi gh- pi t chcd. r r on- har moni c r vhi ne i n t he backgr ound.
' f hi s
can bc avoi decl .
houcver- l rv usi ng a verv l arge \\' i l \cti l bl e.
VSR osci l l at or s ar e usual l v i mpl ement ed i n spcci al i zed. r eal - t i ne har dr var e.
Thcl har e t wt ) ad\ i Lnt l ses: ( I ) t her c i l r c nevcr n) - er r or s due t o a f r i r ct i ol t al
sar npl i ng i ncr emcnt . and ( l ) anv quant i zat i on ' noi se
'
i s har moni cal l y r cl at cd t o t he
si gnal bci nt rcncrrl ed and hcncc. i s noti ccabl r: orrl \ l s a srrri rl l amount
()f
hl rrmoni c
di st or t i on. A di sadr ant age of usi ng VSR s! st ems i s t hat . i n most i nst anccs. t he
r ocuhul ur l ' t i r r sount l s! nt hesi ! usi l ) ! : var i abl c- sar l pl i ng- r l t e' st st ems i s r est r i ct et l t o
t hc capabi l i t i es of f unct i on gcner at or s and osci l l at or s: a var i abl e sampl i ng r at c
rnakcs i t rnorc rl i fl i cul t to i mpl enrent certai n al gori thrrs such l \ th()\e for di gi tal
l i l t er i ng or r ever ber at i on.
3.1I Instrument Designs
Ei l rh of thc chuptcrs of thi s book i n $hi ch strund synthcsi s l cchni ques i rrc di scusscd
i ncl udes a numbcr of i nstrument dcsi gns. Because i t i s anti ci pated that the readers
h
. - l
. :
t
Svnthesis Fnldanentals
wi l l be usi ng a vari et;- of nusi cal programni ng l anguages. wc have used fl ow chafi s
to express the i nstrument desi gns. The i nstrument desi gns are ol tered as a gui dc ttt
whi rt has becn done wi th a parti cul ar sound svnthesi s techni que. The i nstrumcnt
desi gns arc nci ther cxhausti vc nor defl ni ti vc. thc-v- are si mpl y offereL.l hcrc as u
starti ng poi nt for thc reader to devel op a persQnal l ocabul arv of desi gns for
computer synthesl s.
3. 11A Transl at i ng Fl ow Chart Di agrams i nt o Musi c Synt hesi s Codc
Transl ati ng fl ow chart di agrams i nl o wri ttcn i nstrumcnt defl ni ti ons i s a task that
nan.v musi ci ans fi nd i ni ti al l y di ffi cul t. Fol l ou' i ng i s a gcncral gui de to the process. It
i s di vi ded i nto two stages: anal ysi s of the l l ow chart and codi ne of the i nstrumcnt.
There are three steps i nvol ved i n anal l zi ng a fl ow chart. The l l rst step i s to fi nd
thc output or outputs from the i nstrument. Thi s hel ps to show the basi c structure ol
thc i nstrument. Step tu' o consi sts i n desi gnati ng the separate branchcs and
subranches of the i nstrumcnt. By doi ng thi s. thc musi ci an di vi dcs the i nstrument
i nto i ts component parts i n order to know ho$,to di rect thc fl ow of thc si gnal from
one uni t generator te the next. In step threc. thc musi ci an fi nds thc sources for al l
thc i nputs and the desti nati ons of the outputs of al l the branches of the i nstrument
When starti ng to encode the i nstrumcnt desi gn i nto the sound s!nthesi s
l anguage. i t i s essenti al to makc ccrtai n that the use of al l the uni t gencrators i n thc
desi gn i s understood as wel l as the meani ng of thci r i nputs and outputs. Consul t the
manual for the parti cul ar sound svnthcsi s l anguage to be sure. Start the encodi ng
\\' i th the uppermost uni t generator ei ther i n i he fl owchart or i n a branch of the
f l ou, char t . Wr i t c out t hc uni t gener at or name. l abcl i t s out put ( i f appr opr i at c i n t he
l anguagc uscd). and fi l l i n i ts i nputs from the i ni ti al i ral i on val ues. [t i s good practi ce
for most sound svnthcsi s l anguages to l i st thc i ni ti al i zati on val ue' rn i r \cfarate
scct i on at t he head of t he i nst r ument .
Ne\t. fol l ow thc same procedure tbr the subscqucnt uni t {enerators of the
branch or sub-branch unti l the i nstrument i s codcd compl ctel y. Keep i n ni nd that
i nputs of uni t generators to $,hi ch the outputs of other uni t gcncrators are
connected ordi nari l y
gct
thci r val ues at performance ti me. l nputs not fed from
other uni t generators obtai n thci r Yal ues i tt i ni ti al i zati on ti nte.
Aftcr al l thc branchcs ol the i nstrument arc cncodcd. i ntcrconnect tl l cm b)
nl eans appropri atc to the l anguage used. Fi nal l r,. di rcct the resul ts of thc i n-
\trument i nto the output(s) by nteans of an output statement.
Fol l owi ng are some hi nts l br proofreadi ns thc code that descri bes an i nstru-
ment : ( l ) check t hat no uni t gener at or s have been omi t t ed:
( 2)
makc cer t ai n t hat al l
uni t generators i nputs arc of the correct form (for cxampl e, that i tn i nput cxpecti ng
.r frequencl - i n Hz i s not gi vcn frcqucnc,v i n some other notati on): (3) nrake sure
t hat al l uni t
qcner i l t or s
ar e gi ven t he r equi r ed numbcr of i nput s: ( . 1) chcck t o bc
.crtai n thi t al l stored functi ons referred to bv thc uni t gcncrator havc thc ri ght
! ont e nt s.
Common mi st akes i n i nst r umcnt codi ng i ncl ude sendi ng t he out put of onc uni l
! ener at or t o t he wr ong i nput of t hc ncxt . or scndi ng i t t o t he \ vr ong uni t gener i r t or
ent i r cl ! . Be met i cul ous i n checki ng ever v i nput Of ever v uni t gener at or and l n
cl reful l l l abel i ng the branches of thc i nstrumcnt. Anpl e conments shoul d appear
.rt thc head of thc i nstrument to i denti l v i ts functi on and characteri sti cs.
9t
I I
d
SI
nt
. ) l
:ri
hc
ht
ng
hc
n!
t he
, t i l l
hl l
, ol l
t hc
ak-
i cd.
. t r c.
r nul
I hc
r n i c
t hc
t i t rr
r l l t
I l i l I
sscd
r dc r .
92
Computer Music
After encodi ng thc i nstrument desi gn. check thc codc for correct s,vntax by
i nroki ng the orchestra transl ati oi t
ProBri rm.
The transl ator wi l l rl ake a tri al
transl ati (rn of thc code i nto machi ne l anguage and gi ve error messages i f the svntax
i s f aul t l . Ner t . t he musi ci an shoul d t r r out l he i Dst r ul nent on a l ew l vpi cai nol es i n
ordcr to hcar whether the i nsl rument does rvhat i s wantcd. It i s possi bl e. and
i n,.l eed !' ommon. for a desi gn to be s!ntacti cal l ! correct but not to gi ve thc desi red
rcsul ts. The tri al tones wi l l al so be hel pful i n cstabl i shi ng the l i mi ts of the
i nstrumcnt s usetul ncss. Most i nstrumcnts shor great di f{crences i n sound depcnd-
i ng on such factors as note l ength. regi stcr. and ampl i tr.rcl e.
Fi nal l l . t he i nst r umcnt must be t cst ed i n a musi cal cont ext t o f i nd out whet her
i t i s apprtrpri :rte fQr the musi cal ari i cul ati on desi rcd. At thi s poi nt such i ssues rs the
bal ance of thc i nstrumc' n1 wi th copi es of i tscl f i n di fferent regi sters. the bal ance of
t he i nsl r ur nent r i t h ot her i r ' : st r unr ent s i n t he same and ol her r egi st er s. and maski ng
becomc i mp<trtant. l t i s often necessary at thi s poi nt to recast parts of the
i nsl rument to ti t the derrands of the contcxt i n whi ch i t wi l l bc used.
3. l l B I nst rument Desi gn Exampl es
C)ur l i rst i nstrument desi gn uses ri ng modul ati on to pr,:rduce a band of noi se.
Control s on both the ccnter frcquencl - and the wi dth of the band arc provi dcd.
Whi l e more focused noi sc spectra can be synthesi zecl by fi l teri ng whi te noi se
(see
secl i on 5.8). thi s r)rethod i s both cffi ci ent aud useful for tnanr musi cal
Purp()ses.
As shown i n fi gure 3.25. i l noi se generator ri ng modul ates a si nusoi dal
osci l l ator. Thi s process transl ates thc noi sc general or' s l ow-fi equcner ttoi se to l l
hi gher f r equencl r cgi on. ccnt er i ng t he noi se band at t he f r equencv
( FREQ)
of t he
osci l l ator. The anrpl i tude i nput to the Doi ss gencrator (AMP) di rectl y control s thc
arnpl i tudc of the noi se band. The frequcncv at $hi ch the random nQi se i s
gcneratcd,
/^.
deterni nes the band\\' i dth of the noi se (sce secti on 3.9). l l the
Syrthei
hi i nd\ \ : Jl
* r l h . ,
, r r
' n( \ 1. !
. \
nl h. . ! z
t l l \ \ . i : J(]
F : . rt
: r , , . 1L. . , : !
. : 1. i I i i .
: !
Alt?
/
NOI S:
I
RI SE
TIME
)ECAY
TI ME
AMP
/ \
D
RAND]
FRE Q
.\u . :
F I ( ; L R{
l ( r r \
( . 1: . : . .
Rr r r l t t
fR
Aae
FREA
nreans of ri ng modul xtron
Aaa
FI CLi RE l . 15 Gent r al i cr n of noi se hand\ bl
94
Computer Music
A2 AMP. . 67
A5 AMP. 2. 67
A8 AMP. I . J- l
A9 AMP' 1. J1
Dl OUR
02 OUR. I
D. l DUR. . 65
D4 DUR+. 55
D5 DUR. 125
D6 DLl R. Js
O7 DUR. . 25
D8 0UR. . 2
D9 OUR. . 15
D1 0 oUR' . 1
D11 DUR. . O75
Rl FREO. . 56
R2 FREq. 56+ 1
RJ [ REQ. . 92
R4 FREO. . gz+1 7
R5 FRt Q' l . 19
R6 FR!Q. l . 7
R7 FREQI 2.
RB FRt 0. 2. 7a
R9 FREO. I
R10 FRt oj J 76
R11 FRFO. 4 07
:
I
I
FI GURf .
-1. : 7
Bel l i nst rurnent based on Ri sset . (Rusel on era l pl e i r Rf t t . l l l nt roduct ory
Cat al rrgr-re of Conrput er S! nt hcsi zed Sounds- Reprl , rerl bi t h pent l l s. t n oJ Jean-( l uut l t
Ri.!Jr,/.
)
so t hat t hc t wo osci l l at or s on t he l ef t si de
( l abel ed
NOI SE end I NHARM) . r vhi ch
contai n the hi gher trequencv conl ponents of the sound. di e arva_r' sooncr than the
osci l l ator on the ri eht (FUND). The l attcr osci l l ak)r sampl es a storcd si ne tone
produci ng a tone at thc fundarncrrtal frequenc)-. Thc INHARM osci l l ator sampl es a
stctred rvaveform (F3)
consi sti ng of parti al s 10. t6. 12. ar)d 23 wi th rel ati ve
ampl i t udes 1. 1. 5. l . und 1. 5. r ( spect i vel \ ' . Whcn t he f r equcncy of t he I NHARM
l ) sci l l at or i s sel t o 1, ' l Ot h t hat of t he FUr r - D osci l l at or . i t s par t i ai s sound at l " 1. 6.
l . l . r nd l . l t i ncs t hc f r et l uency ol t he f undament al . pr oduci ng i n t hi s wav a t r oup
of par t i al s t l l at i s non- har moni c t o t he f undament al .
Ri ssct has empl oved addi ti ve synthesi s i n t number of hi s u' orks to
froducc
befl -l i ke
-sounds.
A desi gn hast' d on one of the bel l soLrnd: of the Compurer Souttd
Carcl og' - i s shown i n fi sure 3.17. The three pri nci pal ftatures that contri bute to thc
fuslc Synthe6is Fsadahentals
bel l - l i ke sound ar c: ( 1) non- har moni c par t i al s. ( ?) dccal - t i mes of t he par t i al s
r oughl y i nver scl v pr opor t i onal t o t hei r f r equenc. and ( 3) [ r eat i ng of pai r s of
conrponents. sl i ghtl y mi stuned. on the l owcsl two parti al s.
Ri ssct poi fl ts ouf that l vhi l c thc parti al s i rre i nharnrrrni c. thcr are not tuned
arbi trari l v. The l i rst fi ve parti al s of bcl l tones approxi mate the fol l owi ng: a
fundanrcnti tl . a nrj nor thi rd. a perfr' cl hfth. a
.' hul l l
tonc' at an ocl al c bel ou thr.'
f undamcnt al . and t he
"nomi nal
i t t an oct a! e above t he f undament al . l ' he r at i os i n
f r cqucncv f or t hi s gr oupi ng of par t i al s ar c 1: 1. 2: 1. 5: 0. 5: 2. l n Ri sset ' s desi gn. he
cxtends the scri es to i ncl ucl c hi gher parti al s. and tunes th(' parti al s to thc fol l orvi ng
r at i os. _ . 56: . 92: 1. 19 : 1. 70. 2: ?. 7 1 : 3: 3. 7 6: 1 07 .
The \r,rveform ol cach componcnt i s a si nusoi d and the envcl opc
(F2) i s an
exponenti al decay from l to 2-r0. The durati on used i n the Sound Catul og i s 20
seconds. When i rnpl cmenti ng the dcsi gn sugeested i n the fi gure. i t i s advi sabl e to
use a mcthod of
' turni ng
off" the osci l l ator pai rs after thei r pl avi ng ti me has
el apsed. i n order to save computal i on ti mc.
Another of Ri sset' s desi gns from the Conrpi rtcr Sourtd Cutal og i s shou,n i n
frgurc .i .2E. l t reprcsents a computer i nstrumenl thal produces an
"endl ess
gl i ssando." or Shepard tonc. Ps)' chol ogi st Roger Shepard di scovered that the
appar ent rcgi ster of toncs
j n
musi cal scal es coul d bc madc anrbi guous bv carcl ul l l
control l i ng the ampl i tude of the parti al s of thc toncs. Shcpard produced scal cs that
$ere percei ved as
' ci rcul ar"
i n pi tch*whi l e appeari Dg to mov!. conti nuousl y i n onc
di recti on al ong the scal e. they actual l v never l eft the rcgi stcr i n whi ch they began.
Ri sset extended thi s pri nci pl c to achi eve the same cffect wi th gl i ssandoi ns tones as
wel l .
The desi gn i s a hi ghl v control l ed
Bl i ssando
confi gurati on i n whi ch l 0 i ntcrpol at-
i ng osci l l at<rrs track thc samc ampl i tude and frcqucncy functi ons. Each si rusoi dal
osci l l ator i s control l ed by two i nterpol ati ng osci l l ators sampl i ng ampl i tude and
frequency functi ons, respecti vel y. The functi on F3 rvhi ch control s the frequency i s
cxponenti al . Thi s produces a consti tnt change of musi cal i nterval per uni t ti me. F3
dccays from l to 2 "' . produci ng a frequency changc ol l 0 octavcs .rrer i ts
durati on. Each pai r of control l i ng osci l l ators has the same i ni ti al phase. However.
thei r phasc i s offset bv
rurrr
of a cycl e from thc phasc ol a nei ghbori ng pai r. Thi s
corresponds to a phase offset of 51.2 whcn usi ng a wavctabl c of 512 l ocati ons.
Because F3 exponenti al l v decays from I ro l -rr' . the phase otfrcl of l l r) cycl e resul ts
i n the l 0 osci l l ators el i ssandtri ne downward i n paral l el octaves. When an osci l l ator
rr' aches the end of F3. i t
"wraps
around" to the tJegi nni ng of the functi on and
conti nues. Ordi nari l v such a l argc di sconti nui t) i n l requcncl '
(a l 0 octave
j unp)
woul d cause a cl i ck and destrov the effect of smooth gl i ssandoi ne. Howcver. duri ng
t he t r ar l si t i oD. t he ampl i t ude f unct i on F2 i s at i t s mi ni mum val ue. pr er t - nt i ng our
hcari ng the cl i ck. On the other hand. when a tone passes through the mi drangc. F2
greatl v emphasi zes i t. The effect oi summi ng thc l 0 si nusoi dal osci l l ators toBether i s
that of a conti nual l y gl i ssandoi ng tone i n whi ch no change of regi ster
(rccurs.
Ri sset has obserred that the computer must have suffi ci ent rvord l ength to
accura(el y represent the phasc i n order to prevcnt noti cel bl e roundoff error. For
the acousti cal i l l usi on to be effecti re a suffi ci ent durati on must tre used. Ri sset
chosc l l 0 scconds for the compl eti on of the cnti rc c\ch of ten gl i ssandos. He uscd
thc desi gn and other cl oscl y rel ated ones i n hi s composi ti on. Mutati ons I
(.see
sect i on 3. 12
) .
A useful cl ass of sounds for certai n ki nds of musi cal textures i s' choral tone.
'
95
,
D1r
I
ch
hc
ne
\ c
M
6.
r p
cc
-tc
)r)
A
M
P

A
M
P
/
A
U
R
l
0
"
'
z

i
r
c
-
:
i
a
{
\
s
1
/
D
U
R
1
/
D
U
R
2
0
4
.
4
J
9
0
0
2
5
6
.
0
3
9
0
0
J
9
0
0
3
9
0
0
.
1
9
0
0
1
9
0
0
1
/
D
U
R
1

/
D
U
R
1
/
D
\
.
)
R
1
/
D
U
R
1
/
D
U
R
1
/
D
U
R
r
5
1
.
6
2
0
4
.
4
2
5
6
.
0
3
A
7

2
F
2
=
E
X
P
(
-
4
.
6
2
8
J
.
(
1

c
o
s
(
2
.
p
h
(
X
-
2
s
s
.
s
)
/
s
r

j
)
)
)
B
to?
Synthesis fitndamentals
whi ch i s anal ogous to thc cffect i n acousti c musi c of morc than one i nstl ument or
voi ce pl a_vi ng a l i ne i n uni son. A spectral anal ysi s of a group of i nstruntcnts pl avi ng
i n uni son r cvcal s a si gni hcant amount of
"spr cadi ng
i n t he component s of t he
spcctrumi that i s. the energ;- of each component wi l l be more u' i del v di stri buted
about i ts average frequency than when a si ngl e i nstrument i s pl ayi ng. Thi s i s the
rcsul t of sl i ght mi stuni ngs of the i nstruncnts and the l ack of correl ati on among
t hei r vi br at os.
Thc cffcct can be approxi mated bv addi ng another cop! of the computer-
i nstrument desi gn at I or I Hz arvav from the ori gi nal and then appl yi ng a smal l
(approxi matel v l 9?
)
amount of random frcquency devi ati on to both i nstruments.
The randomncss i s bcst i mpl emented wi th a noi se generator that has most of i ts
energy bel ow 20 Hz. Al so. because voi ces do not enter and exi t at exactl _v the same
ti Ines. a smal l amount of random devi ati on i n thei r starti ng ti mes and durati ons. as
rvel l as i n the brcakpoi nts of thei r envel opcs i s desi rabl e. A more advanccd method
that uses del ay l i nes i s dcscri bed i n chapter 7.
3.12 Composi ti onal Exampl es
Jcan-Cl aude Ri sset has made somc of thc most el egant appl i cati ons of fundamental
synthesi s techni ques i n the l i teraturc of computer musi c. Si nce hi s earl i est work i n
comput er musi c. Ri ssct has shown an abi l i t y t o f i nd soni cal l y i nt cr c\ l i ng l c\ r ur cs
that can be real i zed uni quel v by di gi tal synthesi s. Ri sset then bui l ds composi ti ons
around these techni ques i n such a way that the composi ti onal structurc and soni c
surfacc are i nseparabl f i ntertwi ned.
A good exampl e of a composi ti on i n whi ch i nstrument desi gn and composi ti on-
al structure serve to support cach other i s Ri sset' s Mutati orts 1." Regardi ng the
desi gn for the composi ti on. Ri ssct has sai d. "The
tttl e Mutati otl s refers to the
gradual transformati on whi ch occurs throughout thc pi ccc, and to thc passagc from
a di sconti nuous pi tch scal e. at thc bcgi nni ng. to the pi tch conti nuum i n the l ast part.
Thcrc i s a transi ti on between the scal e and thc conti nuum. i n parti cul ar through a
proccss of harmoni c devel opment. whi ch causes the successi ve harmoni cs of the
notes of a chord to come out. The hi ghcr the harmoni c ordcr, the l i ner the pi tch
st ep. hence t he scal e f i nal l v di ssol vcs i nt o t he cont i nuum. "' *
Thc opcni ng passage ol Mutati orts I contai ns three el ements that i l l ustrate ver_v'
wel l Ri sset s wa), of i ntegrati ng i nstrument desi gn i nto composi ti on.rj The passage
ot1ers three wavs of arti cul ati ng a group of pi tches (fi gure 3.29). The short notcs
97
. ^l
. 9
t . 1 1. 4 1. 5
E
4
E E. 5 r o. 5 r . 5
. 9 t . 1. 3 1. 5 1. 8
FI GURE 3. 29 Three \ \ , avs of art i cul at i ng pi t ches f rom Ri ssct s Mr{t . ?l i or. r.
1l }ased
on
et unt p[ t i n Rr. rsel J I nt roduct or\ Cat al oque of Comput cr Sy nt hcsi zcd Sounds. Repri nrcd x i t l t
pentlission of Jean-Clautle Risset.\
AMP
98 Computer MLsic
FIGtIRE
-1.-10
Ri ng-nrodul ati on i nstrunrcnt [r()m Ri s\et s,l y' (i dl i o,J.
(Bused on etantpl c tn
Ri.ft?r:r l nl roductor\ Crtaloguc of Conl puler S) nt hesized Sound.. R.pri !?Ll trith perrtistio4
of Jcqrt Cl outl e Ri sset.)
(A)
arti cul ate the toncs of a chord (B). *' hi ch i s prol onged wi th a crescendo-
decrescendo cnvcl ope. The eongl i ke sound
(C)
at
;1.{}"
echoes thc harmonv of
(he
chord whi l e fusi ng the tooes i nto the percepti on of a uni fi ed ti mbre.
' fhe
desi gn of the i nstrument for thi s passage bri ngs out i nterrel ati ons between
pi tch und ti nrbrc. The i Dstrunrent that pl a]"s (,{) and
(B)
i s shown i n fi gure ,j .3(1. l t
uscs ri ng modul ati on of a si nc torc bv a square wave speci fi ed [ry i rmpl i tude-vcrsus-
ti me apprexi mati on. Ri sset chose
l i )
and
/r
to bt: i nharmoni c to cach other. The
i nstrumeDt has as i ts predomi nant frequency tbe l ower si deband of thc rel ati on
/,
-d,
{fundamental
of the square wave - frequcncy of the si ne rl ' ave). The short
tones sound somcwhi rt metal l i r-. duc to the promi nence ol nor-harmoni c parti al s.
The crescerdt>-tl ecrescendo chord produccs a ti nbre change as th! ampl i tude
of thc si debands. produced by the ri ng modul ati on. changes i n response to the
change i n ampl i tudc of th!' si nc tone. The non-hamoni c parti al s enrphasi zed arc
si mi !ar to thc ones on the short toncs. trut because they l ast l ()ngcr. the cftcct i \
et en cl ear cr .
The gongl i ke toDc consi sts of si ne tones at the frequenci cs of the hve preccdi ng
fundanrental s. al l encompassed bv the sane al tack cnvel oFe and wi th di fferent
dccay ratcs for thc ti Ye componcnts. Thc ti mbre i s unmi stakabl -v gongl i kc. and the
pi tch qual i tv unni stakuhl v echoes the same harrnont' as thc pr!cedi ng events.
There i s a great resourceful ncss i n Ri sset s use of the i nstruments to real i ze thi s
passage. A si ngl c i nstrument i s used for
(A) anct (B). wi th the onl v di ffercncc
het wccn t hem bei ng t he shar p at t ack ( . U1") l br t he shor t not es and t hc sl o\ \
cresccndo-decrescencl o envekrpe for (B).
' Ihe
desi gn for (C). i vi th di ffercrt runi n!\
of parti al s, i s uscd for
-tongl i ke
uncl pcrcussi on-l i ke sounds throu!:h()ut l hc
comoosi t i on.
Anot hcr exar npl e of r csour cef ul ness i n t he use of i nst r ument desi | : t ) \ r ( ' nr c. I n
thc l ast
l rart
of Mutati ons /. u' here frequcnc] i s prescnted u\ u l Ll nti nuunl rn .1
vari ct!' of wa,!-s.
' l ' hc
ntr:rsl stri ki ng rcpresentati on i s i n the
' r' ndl c' r
r:l l .' unJr' th.'
Synthesis Funda.Erental6
PHASE
FI(;URE .1..11 Desi en to produce gradual chrnqc of octave. Thc phase of ei rch e()ntrol l i ng
osci l l ator rs otf\et from that of the osci l l {tor to i ts l el t by an rmount equxl to thc $a\' .txbl e
si zc di \i dcd b\ thc nunber of branches. l Ba!.l t)n erantpl ( i n Rrrl <l .r l nrr(rd(cror\
Crl rl oquc oi Ctrnputcr Svnthcsi zcd Sound\. R./)nxr.,./ x.i th perr.| i ssi cttr t)t' Jt,d| (l nui t'
Ri rr.r.)
dcsi gn for rvhi ch was dcscri bed i n sccti on 3. I l . A vari ati on on that dcsi gn i s used i n
the l ast pi rrl of Mutqti (,rts 1. Here. thc gl i ssando i s cl i nri nated so that one percei vcs
onl y t he gr i t dual change of emphasi s i n pi t ch among t he t en oct aves ( f i gur e 3. 31) .
Ri sset' s ,l rrfi armorti que' " for soprano and tape i s al so bascd on an acousti c
sccnari o.
' l
hc composer rvri tes. "ht
IrthurtttortIque. sounds cmcrge from n(]i se. thcn
the voi ce emc' rges from the tapc sounds. fl ouri shcs. and i s eventual l v sent far awa)'
and hur i ed under t hc t ape sount l s.
L'
The comp. . . cr cont i nues. The t i i l c
Ir unno <1ue refers to thc svstemi tti c use of svntheti c tones Inudc up of preci sel v
cont r ol l cd i nhar moni c pr r t i al : . Such t ones i r r c composed l i kc chor ds. anLl l he! ci r n
ei t her f use i nt o pi t ched cl angs r . >r be di f l r act cd i nt o f l ui d t exl ur cs. '
For l he noi sc sounds at thc bcgi nni ng of /rthonnori que. Ri sscl uscs;r desi gn for
creati ng bands uf noi se bv ri ng modul i tl i n{ a si ne wave wi th noi se (fi ture 3.32). Thc
bands of noi se change i n ampl i tudc. bandwi dth. and center frequencl throughout
each not e. ' "
l n the i nstrumcnt desi gn PEAK AMP sets the maxi mum ampl i tude tbr each
note. The cn!cl ope of each note. mul ti pl i etl by the peak ampl i tude throughout the
note. causes thc ampl i tude to change i n the pattcrn of F5 or F6. For Iuhunnotti que.
Ri sset actual l l has each notc pl ayed by a pai r of
"twi n"
i nstrurncnts-one of the
99
1/ DUR
100
Computer
Music
1F8
.l
//i\{
.[\
z_'
\ . ,
\
FI C; URE
31r r - . , - . . *-
"
- - - - - - - - -
t nt t ' , n, n"' l
t :
'
l r l st l umenl
f I . I n
Ri ' \ ( t \
/ ' r '
o *,,,
:' :);,1
T );,*'
: ;ll:
^"x
i!;;::i,
i.",;.,n'
:: ::" fJ,:l
i:,?l:i:
;,,' :
H
Synthesis Fundamentals
desi gn sho*n i n the fi gure usi ng F5 for i ts envel ope. the othcr i dcnti cal to i t cxcept
that i ts envel ope i s control l ed by the shape of F6. The outputs of the twi n
i nstruments are sent to separate channel s, and so the effect i s an exchange of sound
bi rck and forth bct$' ccn the l oudsoeakers i n the course of the note.
The average wi dth of the noi se band i s determi ncd by the \al ue of the constant
BANDWIDTH. The bandwi dth changes. however, i n response both to the shape of
the ampl i tude envel ope and to the output of the RANDH uni t generator i n the ri ght
branch ol the i nstrument. The rate at whi ch the RANDH puts out a new val ue i s
l (X) Hz at tbe begi nni ng of each note and decreases throuBh two i ni er)rcdi ate rates
to 30 Hz for the l i nal si xl l r of the note. Thus, the enl el ope i s fi rst scal ed to the range
bctween 0 and .2 by thc fi rst mul ti pl i cati on of the l eft branch and then to a val ue
bctween t.l b! a neu,random val ue at the rate of ever.v 100 Hz to 30 Hz. The val ue
1.1) i s then acl dcd to make a val ue i n the range .l l to t.l . The fi nal mul ti pl i cati on i n
the l ett branch then resul t\ i n a bandrvi dth that i uctuates at ri tndom wi thi n a:l l %
of the BANDWIDTH constant.
' l ' here
i s an overal l di recti on to the change of center frequency for each note:
The note begi ns around i ts i ni ti al val ue and, after 200/512 of i ts durati on. fal l s
l i nearl y to one quarter of i ts val ue. Thc actual ral uc of the center frequencl '
fl uctui ttes al random wi thi n 120t; ol i ts val ue.
Whi l c several exampl es of non-harmoni c parti al s are evi dent i n the "metal l i c"
qual i ty of many of the sounds i n l nl armoni que. there are al so passages i n whi ch the
harmoni c seri es i tsel f i s promi nent. A tone can be
"thi ckened"
by pl aci ng ncxt to i t
i denti cal tones of sl i ghtl .v di fferent frequenci es. A good exampl e of thi s effect can
be heard i n the secti on of the work whi ch begi ns at T:3 mi nul es. There. the
texture i s domi nated by l ong, dronel i ke tones of vari ous durati ons on A=55. 110.
and 220 Hz. Cl earl v heard above the fundamcntal s are cascades of tones that are
arpeggi ati ng downward through the harmoni c seri es. The cffect i s caused by the
verv sl ow beati ng of components tha( are \,ery cl ose i n frequencv. The spectrunr of
al l the tones i s one ri ch i n hi gher parti al s that are onl y sl i ghtl y l ess emphasi zed than
the fundamental (l i gure 3.33).
A si ngl e dronel i ke tone i s made by pl aci ng ni ne osci l l ators wi th i denti cal
waveform: and envel opes verv cl osc together i n frequency. For exampl e, the ni ne
osci l l ai or s of t he i i r st t one ar e t uned t o 111) . 110. 03. 110. ( ) 6. 110. 09. 110. 12. 109. 91.
109. 9. 1, 109. 91. and l 09. UI J Hz. A hi ghl y compl ex pat t er n of beat i nS i s sel i nt o
moti on b! the smal l di fferences i n i requency between not onl y the ni ne fun-
AMPLI TUDE
r 0l
1 4 5 67
FI GURE 3. -r-l Spect rum of l one\ f rom Ri ssel s l t t hant pni que
HAR MON IC
102
Computer Music
damental s but al so al l the hl rmoni cs ol al ! thc tones. The harmoni cs wi l l beat wi th
each ot hcr at di f f er ent r at cs: whct her t hev l r e i n pht r e at a gi ven moment u' i l l
cause l hem ei ther tO be emphasi zed or attenuated. Thi s effect i s ptl ssi bl c onl v rvhen
usi ng accur at e. st abl e osci l l at or s t hi l t i nt er pol at e bct ween succcssi r e oLr t put
sanrpl es.
The cascadc effec( takes pl uce when the hannoni c! conre i nt<r pha:e succcss-
i vel t' Irom the hi ghest harmoni c on doun to the fundamental . 1-here arc many other
brcath-taki ng eftccts i n Inhurm<ti que. ancl the rcader i s di rected to a text. i n
French. br Deni s l -orrai n for expl anati on and detai l s of dcsi -en of both the
i nst r ument s ancl t he sc<r r e. ' "
3,13 Other Methods of Instrument Definition
Wc hl vc chosen to dcscri be svnthesi s al gori thms i n terms of uni t generators.
Howerer. dependi ng on the software avai l ahl e. the spcci fi cati on of al gori thms to
generute sound ci rn be done at several Icvel s-rangi ng i rom stcp-b).step speci fi ca-
t i on
( ) f
t he al gor i t hm t o t he modi f i cat i on of an cxi st i ng model of an acoust i cal
proccss. For the purposes of ci i scussi on. four l evel s wi l l be dcfi ned: di rect
progranl s. uni t gcnerators. si enal -genc rati n g model s. and extcndi bl e model s. To
permi t a musi ci an to work at a hi gher l evel . a musi c-systems programmer
preprograms the l ower l evel s.
Wi th di rect programs. the musi ci an di rcctl y speci fi es each step of a program to
generate sanpl e val ues. For cxampl e. a program mi ght hc wri ttcn t(l i mpl ement a
speci al ki nd of osci l l ator. At thi s l evcl . the musi ci an nrust bc fami l i ar rvi th some
non- musi cal pr ogr ammi ng l anguaee ( c. g. . I - ORTRAN) . Thi s met hod of i nst r u'
ment speci fi cati on gi vr,' s thc musi ci an the fl exi bi l i ty to generate an) sequence of
sounds. Horvevcr. to program at thi s l evel certai n acousti c. engi neeri ng. and
progrummi ng experti se i s necessary. For i nstance. to use a di gi tal osci l l ator. the
musi ci an woul d need to kno* exactl v how i t works. not
i ust
$' hat i t does. Worki ng
at thi s l evel . the overal l musi cal pl an can bc obscured bv the necess.l ry rtl cnti on tL)
detai l . The appencl i x gi vcs exampl c al gori thms. al readv desi gned. frrr vari ous sound
generati ng and processi ng dcvi ces.
Thr. next l evel of speci l i cati on. b_v mcans of bui l di ng bl ocks cal l c.d uni t
gencrutors. has been di scussed througbout thi s chapter.
Wi th si gnal -generati ne rnodcl s. the thi rd l evcl of speci fi cati on. the musi ci an
chooses frorn a set of avai l abl e svnthesi s techni ques. l n thi s casc. the computer i s
pre-programmed wi th thc appropri atc i ntcrconnecti ons of uni t generators. The
musi ci an sel ccts a techni que and speci fi es thc parameters neccssarv to control i t.
Avai l abl e st t und svnt hcsi s t echni ( l ues of t en i ncl ude addi t i vc. subt r act i ve. and
di st or t i on ( nonl i near ) synt hesi s. Addi t i ve synt hesi s, descr i bed above, i s t he
summati on of several si mpl c toncr to form a compl ex onc. In subtracti ve synthesi s
(chapter
-5).
the al gori thm begi ns *' i th a compl ex ttrne and di nri ni shes the strength
()f
scl ected frequcnci es i n order to rcal i ze the desi rcd spcctrum. Di storti on
s| nthesi s (chaptcr
-1) cncompasscs sevcral tecl l ni ques where a control l ed amount of
di storti on i s appl i eti to a si mpl c tonc to obtai n a morc compl ex onc. A $' i dei v used
mcmbc' r of thi s cl i rss of techni ques i s frcquency modul ati on. *hi ch can be thought
of as the di stcrrti on of thc frequencv of a tone. Anothcr techni que. \l aveshapi ng. i s
thc di sl orti on of thc waveform of a tone.
Svnthesis f undamentals
The l ast l evel of i nstrument speci l i cati on i s the extendi bl e acousti c modcl . Thi s
nl ethod requi res extcnsi ve techni cal rescarch on the part of thr mu\i c-\)\t!' mi
programmers and i s rel ati vel y undevcl oped. The musi ci an i s gi vcn a model of an
acousti c process and al l owed to varv the parameters of the modcl . For cxampl e ,
thc
softwarc mi ght provi de a preprogrammed vi ol i n tone and the l i st of paramctcrs
uscd to gcneratc the tone i n terms of some of the physi cal attri butes of the
model l ed i nstrument. The musi ci an coul d then al tcr thc character of the tone by
changi ng such parameters as the vol ume of the body. the bowi ng poi nt. the
pl acement of the bri dge, and so on. The vi ol i n-svn thesi s al gori thm i s then al tered
accordi ngl y produci ng a change i n tone qual i t.v. Al though a grcat deal of rcsearch
has bcen done to determi ne i tcousti c model s for the vi ol i n.:o i t wi l l requi re a
sophi sti cated program to i mpl ement an extendi bl e synthesi s vcrsi on of one. If such
a program rvere i ntended to assi st vi ol i n makers. i t *oul d demand an cxcepti onal l !
accuri rtc modcl . Howevcr, i f the musi ci an' s purpose were to l i nd sounds that arc
extensi ons of the ori gi nal . far l ess preci si on woul d bc requi rcd. Al tcrati ons to the
model shoul d produce modi fi cati ons i n the sound whi ch have the corrcct qual i tati ve
effect. but the quanti ty of thc changc coul d bc i mprcci se. The pri mary benefi t of an
extendi bl e model woul d be to gi ve the musi ci an a means to predi ct. to some degrcc.
thc cffcct of ti mbral modi fi cati on. For i nstance. an i ncreasc i n body vol ume woul d
bc cxpected to l o$er manv Of the resonances i n the tonc.
The extendi bl e model method of ti mbral spcci l i cati on can be appl i ed to othcr
i nstrumcnts and to speech. In addi ti on. i t has bcen used for dcscri bi ng proccsses
that modi t,v sounds. such as the speci fi cati on of rcvcrbcrati on on the basi s of thc
ph.vsi cul charactcri sti cs of a room. (See chapter 7.) In the comi ng vears. thi s
mcthod coul d bccome a means of provi di ng musi ci ans u' i th a m.rrc i ntui ti ve
approach to computer i nstrument desi gn than wi th the di rect speci fi cati on of the
paramctcrs of a si gnal -processi ng al gori thm.
Notes
l . Moor e, F. R.
"Tabl c
Lr xr kup Noi se f or Si nuso d. r Di Bi t nl Osci l l at or s. " aomput ( y
Musi c
l our nal ,
1t 2) , 1977, 26 29.
2. Vercoe, tsarry L. Rei eren(e Manual i or the rl tt-/S/C // sound
-svntheJi s
l .rnfi u,rge.
Cambr l dge, Mass. , Exper i ment al Musi c St udi o, M. l . T. , 1979.
L M.l thews, Max. Ihe l echno/ogv o/ Computt,r rvtus;c. CJmbri dge, Mnss.: M.l .T.
Pr es5, 1969.
;1.
Howe, Huberl S., )t.
EIecuotl i c,\JuJi c 5\,rthes/i . New York: Norton, 11)7:.
5. R ssct,
Jcan-C.rude.
Comput(' r Studv oi Itump.\ /one\. MLrfr.r-v H l l , \.1.: t3c
Tel ephone Labor.rtori es, I 966.
6. Cr ey,
i ohn.
"An
Expl or at i on of Musi ca Ti nr br e. Doct or al di sser t . r t i on, St Jni or d
Uni vcr si t y, 1975.
/. Dasho\\' ,
Janres.
"Three Methods for the D gi t.rl S,vnthesi s oi Chord.rl StructLrres wi th
Non- H. r r mon c Par t i al s. " l nt er l a( , . , 7
,
1978, 69- 9, 1.
B. Si egel , 1. , Sk' i 8l i t z, K. , and Zucker man, M.
"Thc
Desi 8n oi Mar kov Chai ns i or
Wavei ornr Cenerati on." Proceedi ngs ol the Insti tute oi El ectri cal and El ectron cs Engi neers
TEASCON ) ,
( ) ct ob( , r ,
I 975.
9. Knuth, tl onal cl . Ihe Arl oi Conl pul er Prop,ram:| ,i np, l Vol . 2) Senr/rumeri cal Al gor-
i t / r nr ! . Readl ng, M. t ss. : Addr son- \ t es ey, 1969, 1 160.
10. Tennev,
James.
"Noi sc St udy. " De( ( n Re( or ds l DL 910l ) , 1961.
I03
i
I
j
I
1
I
l
I
s
I
I
f
I
t
104
Computer Music
11. Ri sset , Jean' Cl aude.
] l t t r()cl ual ary Cat al ogue oi Compul er' -s! Dt hei i zed, \ ounds.
Mur-
r ay Hi l l , N. J. : Bel l Tel ephone L. t bor at or i es, 1969.
12. l br d.
I J. Rr sset , Jean Cl aude.
"Mut at r ons
1. " I NA- CRM Recor dr nF
( AM56409) ,
1979.
l . l . Sahr. rder, Baf i y. l nt rodu. l i on t o Ll e. ua A. out t i t Mu' i c. I ngl crvood Cl l l r, N
] . r
Pr ent i ce- H. l l l , 19A2, 197.
15. R sset , l nt odu. t orv Cat al o!
16. Rr sr et , l ean- (
l aude. "l nhar moni que. " t NA- CRM Re( or di ng
( AM56409J,
1979.
17. Schr ader , a( t r c- Acoust t . Musi c.
18. Lorrai n, Dcni s. "l r\ harmoni que, Anal vse de l a Bande de l ' ()euvre do
Jean
Cl . : ucl e
Ri sset . " Rappof t s I R( AM, 26, 1980.
19. l bl d.
20. Hut chi ns, C. M. "t sowed I nst r ument s ancl Musi c Acousl i cs. " I n
Johdn
Sundber g
1ed. ). i ound CeDe, ' al ron , n \ 1' rnds, St ri Dg! , nnd Comput (' rs. St o(khol f i j Royal Swedr5h
Acadenr y oi Musi c, l 9B0