Communication of Emotions in Vocal Expression and Music Performance:Different Channels, Same Code?Patrik N. Juslin and Petri LaukkaUppsala UniversityManyauthorshavespeculatedabout acloserelationshipbetweenvocal expressionofemotionsandmusical expression of emotions, but evidence bearing on this relationship has unfortunately been lacking.This review of 104 studies of vocal expression and 41 studies of music performance reveals similaritiesbetween the 2 channels concerning (a) the accuracy with which discrete emotions were communicatedto listeners and (b) the emotion-specific patterns of acoustic cues used to communicate each emotion. Thepatternsaregenerallyconsistent withK. R. Scherers(1986)theoretical predictions. Theresultscanexplain why music is perceived as expressive of emotion, and they are consistent with an evolutionaryperspective on vocal expression of emotions. Discussion focuses on theoretical accounts and directionsfor future research.Music: Breathing of statues.Perhaps: Stillness of pictures. You speech, where speeches end.You time, vertically poised on the courses of vanishing hearts.Feelings for what? Oh, you transformation of feelings into. . . audible landscape!You stranger: Music.Rainer Maria Rilke, To MusicCommunicationof emotionsiscrucial tosocial relationshipsand survival (Ekman, 1992). Many researchers argue that commu-nication of emotions serves as the foundation of the social order inanimals and humans (see Buck, 1984, pp. 3136). However, suchcommunication is also a significant feature of performing arts suchas theater and music (G. D. Wilson, 1994, chap. 5). A convincingemotional expressionis oftendesired, or evenexpected, fromactors and musicians. The importance of such artistic expressionshouldnot beunderestimatedbecausethereis nowincreasingevidence that how people express their emotions has implicationsfor their physical health (e.g., Booth & Pennebaker, 2000; Buck,1984, p. 229; Drummond & Quah, 2001; Giese-Davis & Spiegel,2003; Siegman, Anderson, & Berger, 1990).Twomodalitiesthat areoftenregardedaseffectivemeansofemotional communication are vocal expression (i.e., the nonverbalaspects of speech; Scherer, 1986) and music (Gabrielsson & Juslin,2003). Both are nonverbal channels that rely on acoustic signalsfor their transmission of messages. Therefore, it is not surprisingthat proposals about a close relationship between vocal expressionandmusichavealonghistory(Helmholtz, 1863/1954, p. 371;Rousseau, 1761/1986; Spencer, 1857). Inaclassicarticle, TheOrigin and Function of Music, Spencer (1857) argued that vocalmusic, and hence instrumental music, is intimately related to vocalexpression of emotions. He ventured to explain the characteristicsof both on physiological grounds, saying they are premised on thegeneral law that feeling is a stimulus to muscular action (p. 400).In other words, he hypothesized that emotions influence physio-logical processes, which in turn influence the acoustic character-istics of both speech and singing. This notion, which we refer to asSpencerslaw, formedthebasisofmostsubsequentattemptstoexplain reported similarities between vocal expression and music(e.g., Fonagy & Magdics, 1963; Scherer, 1995; Sundberg, 1982).Why should anyone care about such cross-modal similarities, iftheyreallyexist?First, theexistenceofacousticsimilaritiesbe-tweenvocal expressionof emotions andmusic couldhelptoexplainwhylistenersperceivemusicasexpressiveof emotion(Kivy,1980,p.59).Inthissense,anattempttoestablishalinkbetween the two domains could be made for the sake of theoreticaleconomy, because principles from one domain (vocal expression)might help to explain another (music). Second, cross-modal sim-ilarities would support the commonalthough controversialhypothesis that speech and music evolved from a common origin(Brown, 2000; Levman, 2000; Scherer, 1995; Storr, 1992, chap. 1;Zucker, 1946).Anumber of researchers have consideredpossible parallelsbetweenvocal expressionandmusic(e.g., Fonagy&Magdics,1963; Scherer, 1995; Sundberg, 1982), but it isfairtosaythatpreviousworkhasbeenprimarilyspeculativeinnature. Infact,only recently have enough data from music studies accumulated tomake possible a systematic comparison of the two domains. Thepurpose of this article is to review studies from both domains todeterminewhetherthetwomodalitiesreallycommunicateemo-tions in similar ways. The remainder of this article is organized asfollows: First, weoutlineatheoretical perspectiveandaset ofpredictions. Second, we review parallels between vocal expressionandmusicperformanceregarding(a) theaccuracywithwhichPatrik N. Juslin and Petri Laukka, Department of Psychology, UppsalaUniversity, Uppsala, Sweden.A brief summary of this review also appears in Juslin and Laukka (inpress). The writing of this article was supported by the Bank of SwedenTercentenary Foundation through Grant 2000-5193:02 to Patrik N. Juslin.WewouldliketothankNancyEisenbergandKlausSchererforusefulcomments on previous versions of this article.Correspondence concerning this article should be addressed to Patrik N.Juslin, Department ofPsychology, UppsalaUniversity, Box1225, SE-751 42 Uppsala, Sweden. E-mail: patrik.juslin@psyk.uu.sePsychological Bulletin Copyright 2003 by the American Psychological Association, Inc.2003, Vol. 129, No. 5, 770814 0033-2909/03/$12.00 DOI: 10.1037/0033-2909.129.5.770770different emotions are communicated to listeners and (b) theacousticmeansusedtocommunicateeachemotion. Finally, weconsider theoretical accounts andproposedirections for futureresearch.An Evolutionary PerspectiveA review needs a perspective. In this overview, the perspectiveisprovidedbyevolutionarypsychology(Buss,1995).Wearguethat this approach offers the best account of the findings that wereview, inparticularifthetheorizingisconstrainedbyfindingsfromneuropsychological andcomparativestudies(Panksepp&Panksepp, 2000). In this section, we outline theory that serves tosupport the following seven guiding premises: (a) emotions mayberegardedas adaptivereactions tocertainprototypical, goal-relevant, andrecurrent lifeproblemsthat arecommontomanylivingorganisms;(b)animportantpartofwhatmakesemotionsadaptive is that theyare communicatednonverballyfromoneorganism to another, thereby transmitting important information;(c) vocal expression is the most phylogenetically continuous of allforms of nonverbal communication; (d) vocal expressions of dis-crete emotions usually occur in similar types of life situations indifferent organisms; (e)thespecificformthat thevocal expres-sionsofemotiontakeindirectlyreflectthesesituationsor,morespecifically, thedistinct physiological patterns that support theemotional behavior called forth by these situations; (f) physiolog-ical reactions influence an organisms voice production in differ-entiated ways; and (g) by imitating the acoustic characteristics ofthesepatternsofvocalexpression, musicperformersareabletocommunicate discrete emotions to listeners.Evolution and EmotionThe point of departure for an evolutionary perspective on emo-tional communication is that all human behavior depends onneurophysiologicalmechanisms.Theonlyknowncausalprocessthat is capable of yielding such mechanisms is evolution by naturalselection. This is a feedback process that chooses among differentmechanisms on the basis of howwell they function; that is,function determines structure (Cosmides & Tooby, 2000, p. 95).Given that the mind acquired its organization through the evolu-tionary process, it may be useful to understand human functioningin terms of its adaptive significance (Cosmides & Tooby, 1994).Thisisparticularlytruefor suchtypesof behavior that canbeobserved in other species as well (Bekoff, 2000; Panksepp, 1998).Several researchers have takenanevolutionaryapproachtoemotions. Before considering this literature, a preliminary defini-tionof emotionsisneeded. Althoughemotionsaredifficult todefine and measure (Plutchik, 1980), most researchers wouldprobably agree that emotions are relatively brief and intense reac-tions to goal-relevant changes in the environment that consist ofmany subcomponents: cognitive appraisal, subjective feeling,physiological arousal, expression, action tendency, and regulation(Scherer, 2000, p. 138). Thus, for example, anevent maybeappraised as harmful, evoking feelings of fear and physiologicalreactions in the body; individuals may express this fear verballyand nonverbally and may act in certain ways (e.g., running away)ratherthanothers. However, researchersdisagreeastowhetheremotions arebest conceptualizedas categories (Ekman, 1992),dimensions (Russell, 1980), prototypes (Shaver, Schwartz, Kirson,& OConnor, 1987), or component processes (Scherer, 2001). Inthis review, wefocus mainlyontheexpressioncomponent ofemotion and adopt a categorical approach.According to the evolutionary approach, the key to understand-ingemotionsistostudywhat functionsemotionsserve(Izard,1993; Keltner & Gross, 1999). Thus, to understand emotions onemust consider howtheyreflect theenvironment inwhichtheydeveloped and to which they were adapted. On the basis of variouskinds of evidence, Oatley and Jenkins (1996, chap. 3) suggestedthat humans environment of evolutionary adaptedness about200,000 years ago was that of seminomadic huntergatherergroups of 10 to 30 people living face-to-face with each other inextended families. Most emotions, they suggested, are presumablyadapted to living this kind of way, which involved cooperating inactivities suchas huntingandrearingchildren. Several of theactivitiesareassociatedwithbasicsurvival problemsthat mostorganisms haveincommonavoidingpredators, findingfood,competing for resources, and caring for offspring. These problems,in turn, required specific types of adaptive reactions. A number ofauthors have suggested that such adaptive reactions were theprototypes of emotions as seen in humans (Plutchik, 1994, chap. 9;Scott, 1980).This view of emotions is closely related to the concept of basicemotions, that is, the notion that there is a small number ofdiscrete, innate, and universal emotion categories from which allother emotions may be derived (e.g., Ekman, 1992; Izard, 1992;Johnson-Laird&Oatley, 1992). Eachbasicemotioncanbede-fined, functionally, in terms of an appraisal of goal-relevant eventsthat have recurred during evolution (see Power & Dalgleish, 1997,pp. 8699). Examples of suchappraisals aregivenbyOatley(1992, p. 55): happiness (subgoals being achieved), anger (activeplanfrustrated), sadness(failureofmajorplanorlossofactivegoal), fear (self-preservation goal threatened or goal conflict), anddisgust(gustatorygoalviolated).Basicemotionscanbeseenasfast andfrugal algorithms(Gigerenzer &Goldstein, 1996) thatdeal with fundamental life issues under conditions of limited time,knowledge, or computational capacities. Having a small number ofcategoriesisanadvantageinthiscontext becauseit avoidstheexcessiveinformationprocessingthat comeswithtoomanyde-grees of freedom (Johnson-Laird & Oatley, 1992).Thenotionofbasicemotionshasbeenthesubjectofcontro-versy(cf. Ekman, 1992; Izard, 1992; Ortony&Turner, 1990;Panksepp, 1992). We propose that evidence of basic emotions maycome from a range of sources that include findings of (a) distinctbrain substrates associated with discrete emotions (Damasio et al.,2000; Panksepp, 1985, 2000, Table 9.1; Phan, Wager, Taylor, &Liberzon, 2002), (b) distinct patterns of physiological changes(Bloch, Orthous, &Santibanez, 1987; Ekman, Levenson, &Friesen, 1983; Fridlund, Schwartz, &Fowler, 1984; Levenson,1992; Schwartz, Weinberger, &Singer, 1981), (c) primacyofdevelopment of proposed basic emotions (Harris, 1989), (d) cross-cultural accuracy in facial and vocal expression of emotion (Elf-enbein&Ambady, 2002), (e) clustersthat correspondtobasicemotions in similarity ratings of affect terms (Shaver et al., 1987),(f) reduced reaction times in lexical decision tasks when primingwords are taken from the same basic emotion category (Conway &Bekerian, 1987), and (g) phylogenetic continuity of basic emotions(Panksepp, 1998, chap. 13; Plutchik, 1980; Scott, 1980). It is fair771COMMUNICATION OF EMOTIONStoacknowledgethat someofthesesourcesofevidencearenotstrong. In the case of autonomic specificity especially, the jury isstill out (for a positive view, see Levenson, 1992; for a negativeview, see Cacioppo, Berntson, Larsen, Poehlmann, & Ito, 2000).1Arguably, thestrongestevidenceofbasicemotionscomesfromstudies of communication of emotions (Ekman, 1973, 1992).Vocal Communication of EmotionEvolutionaryconsiderationsmaybeespeciallyrelevantinthestudyofcommunicationofemotions, becausemanyresearchersthink that such communication serves important functions. First,expression of emotions allows individuals to communicate impor-tant information to others, which may affect their behaviors. Sec-ond, recognitionof emotionsallowsindividualstomakequickinferencesabout theprobableintentionsandbehaviorofothers(Buck, 1984, chap. 2; Plutchik, 1994; chap. 10). The evolutionaryapproach implies a hierarchy in the ease with which variousemotions are communicated nonverbally. Specifically, perceiversshouldbeattunedtothat informationthat ismost relevant foradaptiveaction(e.g., Gibson, 1979). It hasbeensuggestedthatboth expression and recognition of emotions proceed in terms of asmall number of basic emotion categories that represent the opti-mal compromise between two opposing goals of the perceiver: (a)the desire to have the most informative categorization possible and(b) the desire tohave these categories be as discriminable aspossible (Juslin, 1998; cf. Ross & Spalding, 1994). To be useful asguides to action, emotions are recognized in terms of only a fewcategories related to life problems such as danger (fear), compe-tition (anger), loss (sadness), cooperation (happiness) and caregiv-ing(love).2Byperceivingexpressedemotionsintermsofsuchbasic emotion categories, individuals are able to make usefulinferences in response to urgent events. It is arguable that the sameselective pressures that shaped the development of the basic emo-tionsshouldalsofavorthedevelopmentofskillsforexpressingandrecognizingthesameemotions. Inlinewiththisreasoning,manyresearchershavesuggestedtheexistenceof innateaffectprograms, which organize emotional expressions in terms of basicemotions (Buck, 1984; Clynes, 1977; Ekman, 1992; Izard, 1992;Lazarus, 1991; Tomkins, 1962). Support for thisnotioncomesfrom evidence of categorical perception of basic emotions in facialandvocalexpression(deGelder, Teunisse, &Benson, 1997;deGelder &Vroomen, 1996; Etcoff &Magee, 1992; Laukka, inpress), more or less intact vocal and facial expressions of emotionin children born deaf and blind (Eibl-Eibesfeldt, 1973), and cross-cultural accuracy in facial and vocal expression of emotion (Elf-enbein & Ambady, 2002).Phylogeneticcontinuity. Vocal expressionmaybethemostphylogenetically continuous of all nonverbal channels. In his clas-sicbook, TheExpressionoftheEmotionsinManandAnimals,Darwin (1872/1998) reviewed different modalities of expression,including the voice: With many kinds of animals, man included,the vocal organs are efficient in the highest degree as a means ofexpression(p. 88).3FollowingDarwins theory, anumber ofresearchers of vocal expression have adopted an evolutionaryperspective (H. Papousek, Jurgens, & Papousek, 1992). A primaryassumptionisthat thereisphylogeneticcontinuityofvocal ex-pression. Ploog(1992)describedthemorphological transforma-tion of the larynxfrom a pure respiratory organ (in lungfish) toa respiratory organ with a limited vocal capability (in amphibians,reptiles, andlower mammals) and, finally, tothesophisticatedinstrument that humansusetosingor speakinanemotionallyexpressive manner.Vocal expressionseemsespeciallyimportant insocial mam-mals. Social grouping evolved as a means of cooperative defense,althoughthisimpliesthat somekindof communicationhadtodevelop to allow sharing of tasks, space, and food (Plutchik, 1980).Thus, vocalexpressionprovidedameansofsocialcoordinationand conflict resolution. MacLean (1993) has argued that the limbicsystem of the brain, an essential region for emotions, underwent anenlargement with mammals and that this development was relatedtoincreasedsociality,asevidentinplaybehavior,infantattach-ment, andvocal signaling. Thedegreeof differentiationinthesound-producingapparatus is reflectedintheorganisms vocalbehavior. For example, the primitive condition of the sound-producing apparatus in amphibians (e.g., frogs) permits only a fewinnatecalls, suchas matingcalls, whereas thehighlyevolvedlarynx of nonhuman primates makes possible a rich repertoire ofvocal expressions (Ploog, 1992).Theevolutionof thephonatoryapparatustowarditsforminhumans is paralleled not only by an increase in vocal repertoire butalsobyanincreaseinvoluntarycontrol overvocalization. It ispossible to delineate three levels of development of vocal expres-sionintermsof anatomicandphylogeneticdevelopment (e.g.,Jurgens, 1992, 2002). The lowest level is represented by a com-pletely genetically determined vocal reaction (e.g., pain shrieking).In this case, neither the motor pattern producing the vocal expres-sion nor the eliciting stimulus has to be learned. This is referred toas an innate releasing mechanism. The brain structures responsiblefor the control of such mechanisms seem to be limited mainly tothe brain stem (e.g., the periaqueductal gray).The following level of vocal expression involves voluntarycontrolconcerningtheinitiationandinhibitionoftheinnateex-pressions. For example, rhesus monkeys may be trained in a vocaloperant conditioning task to increase their vocalization rate if each1Itseemstousthatoneargumentisoftenoverlookedindiscussionsregarding physiological specificity, namely that this may be a case in whichpositive results count as stronger evidence than do negative results. It isgenerally agreed that there are several methodological problems involvedin measuring physiological indices (e.g., individual differences, time-dependent nature of measures, difficulties in providing effective stimuli).Given the error variance, or noise, that this produces, it is arguably moreproblematic for the no-specificity hypothesis that a number of studies haveobtainedsimilarandreliableemotion-specificpatternsthanit isforthespecificityhypothesisthatanumberofstudieshavefailedtoyieldsuchpatterns. Failure to obtain patterns may be due to error variance, but howcan the presence of similar patterns in several studies be explained?2Loveisnot includedinmost listsofbasicemotions(e.g., Plutchik,1994, p. 58), althoughsomeauthorsregardit asabasicemotion(e.g.,Clynes, 1977; MacLean, 1993; Panksepp, 2000; Scott, 1980; Shaver et al.,1987), as have philosophers suchas Descartes, Spinoza, andHobbes(Plutchik, 1994, p. 54).3InaccordancewithSpencers law, Darwin(1872/1998) notedthatvocalizations largely reflect physiological changes: Involuntary . . . con-tractions of the muscles of the chest and the glottis . . . may first have givenrise to the emission of vocal sounds. But the voice is now largely used forvarious purposes; one purpose mentioned was intercommunication (p.89).772JUSLIN AND LAUKKAvocalization is rewarded with food (Ploog, 1992). Brain-lesioningstudies of rhesus monkeys have revealed that this voluntary controldepends on structures in the anterior cingulate cortex, and the samebrain region has been implicated in humans (Jurgens &vanCramon, 1982). Neuroanatomical researchhas shownthat theanterior cingulate cortex is directly connected to the periaqueduc-tal region and thus in a position to exercise control over the moreprimitive vocalization center (Jurgens, 1992).The highest level of vocal expression involves voluntary controlover thepreciseacousticpatternsof vocal expression. Thisin-cludes the capability to learn vocal patterns by imitation, as well astheproductionofnewpatternsbyinvention. Theseabilitiesareessential in the uniquely human inventions of language and music.Amongthe primates, onlyhumans have gaineddirect corticalcontrol over the voice, which is a prerequisite for singing. Neuro-anatomical studies have indicated that nonhuman primates lack thedirectconnectionbetweentheprimarymotorcortexandthenu-cleusambiguus(i.e., thesiteofthelaryngealmotoneurons)thathumans have (Jurgens, 1976; Kuypers, 1958).Comparative research. Results fromneurophysiological re-search that indicate that there is phylogenetic continuity of vocalexpression have encouraged some researchers to embark on com-parative studies of vocal expression. Although biologists andethologistshavetendedtoshyawayfromusingwordssuchasemotion in connection with animal behavior (Plutchik, 1994, chap.10; Scherer, 1985), a case could be made that most animal vocal-izations involve motivational states that are closelyrelatedtoemotions (Goodall, 1986; Hauser, 2000; Marler, 1977; Ploog,1986; Richman, 1987; Scherer, 1985; Snowdon, 2003). The statesusually have to be inferred from the specific situations in which thevocalizationsoccurred. Inmost of thecircumstancesinwhichanimal signaling occurs, one detects urgent and demanding func-tionstobeserved, ofteninvolvingemergenciesfor survival orprocreation (Marler, 1977, p. 54). There is little systematic workon vocal expression in animals, but several studies have indicateda close correspondence betweenthe acoustic characteristics ofanimal vocalizations and specific affective situations (for reviews,see Plutchik, 1994, chap. 910; Scherer, 1985; Snowdon, 2003).For instance, Ploog (1981, 1986) discovered a limited number ofvocal expression categories in squirrel monkeys. These categorieswererelatedtoimportant eventsinthemonkeys livesandin-cluded warning calls (alarm peeps), threat calls (groaning), desirefor social contact calls (isolation peeps), and companionship calls(cackling).Givenphylogeneticcontinuityofvocal expressionandcross-speciessimilarityinthekindsof situationsthat generatevocalexpression, it is interesting to ask whether there is any evidence ofcross-species universality of vocal expression. Limited evidence ofthis kind has indeed been found (Scherer, 1985; Snowdon, 2003).For instance, E. S. Morton (1977) noted that birds and mammalsuse harsh, relatively low-frequency sounds when hostile, andhigher-frequency, more pure tonelike sounds whenfrightened,appeasing, or approaching in a friendly manner (p. 855; see alsoOhala, 1983). Another general principle, proposedbyJurgens(1979), isthat increasingaversivenessofprimatevocal callsiscorrelatedwithpitch, total pitchrange, andirregularityofpitchcontours. These features have also been associated with negativeemotion in human vocal expression (Davitz, 1964b; Scherer,1986).Physiological differentiation. Inanimal studies, descriptionsofvocal characteristicsandemotional statesarenecessarilyim-precise (Scherer, 1985), making direct comparisons difficult. How-ever, at the least these data suggest that there are some systematicrelationships among acoustic measures and emotions. An impor-tant questionishowsuchrelationshipsandexamplesof cross-species universalitymaybeexplained. AccordingtoSpencerslaw, thereshouldbecommonphysiological principles. Infact,physiologicalvariablesdeterminetoalargeextentthenatureofphonation and resonance in vocal expression (Scherer, 1989), andthere may be some reliable differentiation of physiological patternsfor discrete emotions (Cacioppo et al., 2000, p. 180).It might be assumed that distinct physiological patterns reflectenvironmental demands onbehavior: Behaviors suchas with-drawal, expulsion, fighting, fleeing, and nurturing each makedifferent physiological demands. Amost important functionofemotion is to create the optimal physiological milieu to support theparticular behavior that is called forth (Levenson, 1994, p. 124).This process involves the central, somatic, and autonomic nervoussystems. For example, fear is associated with a motivation to fleeandbringsaboutsympatheticarousalconsistentwiththisactioninvolving increased cardiovascular activation, greater oxygen ex-change, and increased glucose availability (Mayne, 2001). Manyphysiological changes influence aspects of voice production, suchas respiration, vocal fold vibration, and articulation, in well-differentiated ways. For instance, anger yields increased tension inthelaryngeal musculaturecoupledwithincreasedsubglottal airpressure. This changes the production of sound at the glottis andhence changes the timbre of the voice (Johnstone &Scherer,2000). Inother words, dependingonthespecificphysiologicalstate, one may expect to find specific acoustic features in the voice.This general principleunderlies Scherers (1985) componentprocess theory of emotion, which is the most promising attempt toformulateastringent theoryalongthelines of Spencers law.Using this theory, Scherer (1985) made detailed predictions aboutthe patterns of acoustic cues (bits of information) associated withdifferent emotions. Thepredictionswerebasedontheideathatemotions involve sequential cognitive appraisals, or stimulus eval-uation checks (SECs), of stimulus features such as novelty, intrin-sic pleasantness, goal significance, coping potential, and norm orself compatibility (for further elaboration of appraisal dimensions,see Scherer, 2001). The outcome of each SEC is assumed to haveaspecificeffect onthesomaticnervoussystem, whichinturnaffects the musculature associated with voice production. In addi-tion, each SEC outcome is assumed to affect various aspects of theautonomous nervous system (e.g., mucous and saliva production)in ways that strongly influence voice production. Scherer (1985)didnot favorthebasicemotionsapproach, althoughheofferedpredictions for acoustic cues associated with anger, disgust, fear,happiness, and sadnessfive major types of emotional states thatcanbeexpectedtooccur frequentlyinthedailylifeof manyorganisms, both animal and human (p. 227). Later in this review,we provide a comparison of empirical findings from vocal expres-sion and music performance with Scherers (1986) revisedpredictions.Although human vocal expression of emotion is based on phy-logenetically old parts of the brain that are in some respects similarto those of nonhuman primates, what is characteristic of humans isthat they have much greater voluntary control over their vocaliza-773COMMUNICATION OF EMOTIONStion (Jurgens, 2002). Therefore, an important distinction must bemade between so-called push and pull effects in the determinantsof vocal expression (Scherer, 1989). Push effects involve variousphysiological processes, suchasrespirationandmuscletension,that are naturally influenced by emotional response. Pull effects,onthe other hand, involve external conditions, suchas socialnorms, that may lead to strategic posing of emotional expressionfor manipulative purposes (e.g., Krebs & Dawkins, 1984). Vocalexpression of emotions typically involves a combination of pushand pull effects, and it is generally assumed that posed expressiontendstobemodeledonthebasisofnatural expression(Davitz,1964c, p. 16; Owren & Bachorowski, 2001, p. 175; Scherer, 1985,p. 210). However, the precise extent to which posed expression issimilar tonatural expressionisaquestionthat requiresfurtherresearch.Vocal Expression and Music Performance: Are TheyRelated?It is arecurrent notionthat musicis ameans of emotionalexpression(Budd, 1985;S. Davies, 2001;Gabrielsson&Juslin,2003). Indeed, musichasbeendefinedasoneofthefineartswhich is concerned with the combination of sounds with a view tobeauty of form and the expression of emotion (D. Watson, 1991,p. 8). It has been difficult to explain why music is expressive ofemotions, but one possibility is that music is reminiscent of vocalexpression of emotions.Previousperspectives. Thenotionthat thereisacloserela-tionshipbetweenmusicandthehumanvoicehasalonghistory(Helmholtz, 1863/1954; Kivy, 1980; Rousseau, 1761/1986;Scherer, 1995; Spencer, 1857; Sundberg, 1982). Helmholtz (1863/1954)one of the pioneers of music psychologynoted that anendeavor to imitate the involuntary modulations of the voice, andmake its recitationricher andmore expressive, maythereforepossibly have led our ancestors to the discovery of the first meansof musical expression (p. 371). This impression is reinforced bythe voicelike character of most musical instruments: There are inthemusicof theviolin . . . accents socloselyakintothoseofcertain contralto voices that one has the illusion that a singer hastaken her place amid the orchestra (Marcel Proust, as cited in D.Watson, 1991, p. 236). RichardWagner, thefamouscomposer,noted that the oldest, truest, most beautiful organ of music, theorigintowhichaloneour musicowesitsbeing, isthehumanvoice (as citedinD. Watson, 1991, p. 2). Indeed, Stendhalcommented that no musical instrument is satisfactory except in sofar as it approximates to the sound of the human voice (as citedin D. Watson, 1991, p. 309). Many performers of blues music havebeen attracted to the vocal qualities of the slide guitar (Erlewine,Bogdanov, Woodstra, & Koda, 1996). Similarly, people often referto the musical aspects of speech (e.g., Besson & Friederici, 1998;Fonagy&Magdics, 1963), particularlyinthecontextofinfant-directedspeech, wheremothersusechangesinduration, pitch,loudness, and timbre to regulate the infants level of arousal (M.Papousek, 1996).The hypothesis that vocal expression and music share a numberofexpressivefeaturesmightappeartrivialinthelightofallthearguments bydifferent authors. However, these comments areprimarily anecdotal or speculative in nature. Indeed, many authorshave disputed this hypothesis. S. Davies (2001) observed thatit has been suggested that expressive instrumental music recalls thetones and intonations with which emotions are given vocal expression(Kivy, 1989), but this . . . is dubious. It is true that blues guitar andjazzsaxophonesometimesimitatesingingstyles, andthat singingstyles sometimes recall the sobs, wails, whoops, and yells that go withordinaryoccasionsofexpressiveness. Forthegeneral runofcases,though, musicdoesnotsoundverylikethenoisesmadebypeoplegripped by emotion. (p. 31)(See alsoBudd, 1985, p. 148; Levman, 2000, p. 194.) Thus,awaiting relevant data, it has been uncertain whether Spencers lawcan provide an account of musics expressiveness.Boundary conditions of Spencers law. It does actually seemunlikely that Spencers law can explain all of musics expressive-ness. For instance, there are several aspects of musical form (e.g.,harmonicprogression)thathavenocounterpartinvocalexpres-sionbut that nonetheless contribute tomusics expressiveness(e.g., Gabrielsson&Juslin, 2003). Consequently, Spencerslawcannot be the whole story of musics expressiveness. In fact, thereare many sources of emotion in relation to music (e.g., Sloboda &Juslin, 2001)includingmusicalexpectancy(Meyer, 1956), arbi-trary association (J. B. Davies, 1978), and iconic significationthat is, structural similaritybetweenmusical andextramusicalfeatures (Langer, 1951). Only the last of these sources correspondsto Spencers law. Yet, we argue that Spencers law should be partof anysatisfactoryaccount of musics expressiveness. For thehypothesis to have explanatory power, however, it must be con-strained. What is required, wepropose, is specificationof theboundary conditions of the hypothesis.We argue that the hypothesis that there is an iconic similaritybetweenvocalexpressionofemotionandmusicalexpressionofemotion applies only to certain acoustic featuresprimarily thosefeatures of the music that the performer can control (more or lessfreely) during his or her performance such as tempo, loudness, andtimbre. However, the hypothesis does not apply to such features ofa piece of music that are usually indicated in the notation of thepiece (e.g., harmony, tonality, melodic progression), because thesefeatures reflect toalarger extent characteristics of musicas ahuman art form that follows its own intrinsic rules and that variesfromoneculturetoanother(Carterette&Kendall,1999;Juslin,1997c). Neuropsychological research indicates that certain aspectsof music (e.g., timbre) share the same neural resources as speech,whereas others (e.g., tonality) draw on resources that are unique tomusic(Patel &Peretz, 1997; seealsoPeretz, 2002). Thus, weargue that musicians communicate emotions to listeners via theirperformances of music by using emotion-specific patterns ofacousticcuesderivedfromvocalexpressionofemotion(Juslin,1998). The extent to which Spencers law can offer an explanationofmusicsexpressivenessisdirectlyproportionaltotherelativecontribution of performance variables to the listeners perceptionof emotions in music. Because performance variables include suchperceptually salient features as speed and loudness, this contribu-tion is likely to be large.It is well-known that the same sentence may be pronounced ina large number of different ways, and that the way in which it ispronounced may convey the speakers state of emotion. In princi-ple, one can separate the verbal message from its acoustic realiza-tion in speech. Similarly, the same piece of music can be played ina number of different ways, and the way in which it is played mayconvey specific emotions to listeners. In principle, one can sepa-774JUSLIN AND LAUKKArate the structure of the piece, as notated, from its acoustic real-ization in performance. Therefore, to obtain possible similarities,how speakers and musicians express emotions through the ways inwhich they convey verbal and musical contents should be explored(i.e., Its not what you say, its how you say it).The origins of the relationship. If musical expression of emo-tion should turn out to resemble vocal expression of emotion, howdid musical expression come to resemble vocal expression in thefirst place? The origins of music are, unfortunately, forever lost inthehistoryof our ancestors (but for asurveyof theories, seevarious contributions in Wallin, Merker, & Brown, 2000). How-ever, it is apparent that music accompanies many important humanactivities, and this is especially true of so-called preliterate cultures(e.g., Becker, 2001; Gregory, 1997). It is possible to speculate thatthe origin of music is to be found in various cultural activities ofthe distant past, when the demarcation between vocal expressionandmusicwasnot asclearasit istoday. Vocal expressionofdiscrete emotions suchas happiness, sadness, anger, andloveprobably became gradually meshed with vocal music that accom-panied related cultural activities such as festivities, funerals, wars,andcaregiving. Anumberofauthorshaveproposedthat musicserved to harmonize the emotions of the social group and to createcohesion:Singinganddancingservestodrawgroupstogether,direct the emotions of the people, andprepare themfor jointaction (E. O. Wilson, 1975, p. 564). There is evidence thatlistenerscanaccuratelycategorizesongsof different emotionaltypes (e.g., festive, mourning, war, lullabies) that come fromdifferent cultures (Eggebrecht, 1983) and that there are similaritiesin certain acoustic characteristics used in such songs; for instance,mourning songs typically have slow tempo, low sound level, andsoft timbre, whereasfestivesongshavefast tempo, highsoundlevel, and bright timbre (Eibl-Eibesfeldt, 1989, p. 695). Thus, it isreasonable to hypothesize that music developed from a means ofemotion sharing and communication to an art form in its own right(e.g., Juslin, 2001b; Levman, 2000, p. 203; Storr, 1992, p. 23;Zucker, 1946, p. 85).Theoretical PredictionsIntheforegoing, weoutlinedanevolutionaryperspectiveac-cording to which music performers are able to communicate basicemotions to listeners by using a nonverbal code that derives fromvocal expression of emotion. We hypothesized that vocal expres-sion is an evolved mechanism based on innate, fairly stable, anduniversal affect programs that develop early and are fine-tuned byprenatal experiences(Mastropieri &Turkewitz, 1999; Verny&Kelly, 1981). We made the following five predictions on the basisof this evolutionary approach. First, we predicted that communi-cationof basicemotions wouldbeaccurateinbothvocal andmusical expression. Second, wepredictedthat therewouldbecross-cultural accuracyof communicationof basicemotions inbothchannels, aslongascertainacousticfeaturesareinvolved(speed,loudness,timbre).Third,wepredictedthattheabilitytorecognize basic emotions in vocal and musical expression devel-opsearlyinlife. Fourth, wepredictedthatthesamepatternsofacousticcuesareusedtocommunicatebasicemotionsinbothchannels. Finally, we predicted that the patterns of cues would beconsistent with Scherers (1986) physiologically based predictions.Thesefivepredictionsareaddressedinthefollowingempiricalreview.Definitions and Method of the ReviewBasic Issues and Terminology in NonverbalCommunicationVocal expression and music performance arguably belong to the generalclass of nonverbal communication behavior. Fundamental issues concern-ing nonverbal communication include (a) the content (What is communi-cated?), (b) the accuracy (How well is it communicated?), and (c) the codeusage (How is it communicated?). Before addressing these questions, oneshould first make sure that communication has occurred. Communicationimplies (a) a socially shared code, (b) an encoder who intends to expresssomething particular via that code, and (c) a decoder who responds sys-tematically to that code (e.g., Shannon & Weaver, 1949; Wiener, Devoe,Rubinow, & Geller, 1972). True communication has taken place only if theencoders expressive intention has become mutually known to the encoderand the decoder (e.g., Ekman & Friesen, 1969). We do not exclude thatinformationmaybeunwittinglytransmittedfromonepersontoanother,but this would not count as communication according to the presentdefinition (for a different view, see Buck, 1984, pp. 45).An important aspect of the communicative process is the coding of thenonverbal signs (the manner in which information is transmitted throughthesignal). AccordingtoEkmanandFriesen(1969), thenatureof thecoding can be described by three dimensions: discrete versus continuous,probabilistic versus invariant, and iconic versus arbitrary. Nonverbal sig-nals are typically coded continuously, probabilistically, and iconically. Toillustrate, (a) the loudness of the voice changes continuously (rather thandiscretely);(b)increasesinloudnessfrequently(butnotalways)signifyanger; and (c) the loudness is iconically (rather than arbitrarily) related totheintensityofthefeltanger(e.g.,theloudnessincreaseswhenthefeltintensity of the anger increases; Juslin & Laukka, 2001, Figure 4).The Standard Content ParadigmStudies of vocal expressionandstudies of musicperformancehavetypically been carried out separately from each other. However, one couldarguethatthetwodomainsshareanumberofimportantcharacteristics.First, bothdomainsareconcernedwithachannel that usespatternsofpitch, loudness, and duration to communicate emotions (the content).Second, both domains have investigated the same questions (How accurateis thecommunication?, What is thenatureof thecode?). Third, bothdomains have used similar methods (decoding experiments, acoustic anal-yses). Hence, both domains have confronted many of the same problems(see the Discussion section).In a typical study of communication of emotions in vocal expression ormusicperformance, theencoder(speaker/performerinvocalexpression/music performance, respectively) is presented with material to be spoken/performed.Thematerialusuallyconsistsofbriefsentencesormelodies.Each sentence/melody is to be spoken/performed while expressing differ-ent emotions prechosen by the experimenter. The emotion portrayals arerecorded and used in listening tests to study whether listeners can decodetheexpressedemotions.Eachportrayalisanalyzedtoseewhatacousticcues are usedinthe communicative process. The assumptionis that,because the verbal/musical material remains the same in different portray-als, whatever effects that appear in listeners judgments or acoustic mea-sures should primarily be the result of the encoders expressive intention.This procedure, often referred to as the standard content paradigm (Davitz,1964b), is not without its problems, but wetemporarilypostponeourcritique until the Discussion section.775COMMUNICATION OF EMOTIONSCriteria for Inclusion of StudiesWe used two criteria for inclusion of studies in the present review. First,we includedonlystudies focusingonnonverbal aspects of speechorperformance-relatedaspects of music. This is inaccordance withtheboundaryconditionsofthehypothesisdiscussedabove. Second, wein-cluded only studies that investigated the communication of discrete emo-tions (e.g., sadness). Hence, studies that focused on emotional arousal ingeneral(e.g.,Murray,Baber,&South,1996)oronemotiondimensions(e.g., Laukka, Juslin, &Bresin, 2003)werenot includedinthereview.Similarly, studiesthat usedthestandardparadigmbut that didnot useexplicitly defined emotions (e.g., Cosmides, 1983) or that used onlypositiveversusnegativeaffect (e.g., Fulcher, 1991) werenot included.Suchstudiesdonot allowfortherelevant comparisonswithstudiesofmusic performance, which have almost exclusively studied discreteemotions.Search StrategyEmotioninvocal expressionandmusicperformanceisamultidisci-plinary field of research. The majority of studies have been conducted bypsychologists, butcontributionsalsocomefrom, forinstance, acoustics,speech science, linguistics, medicine, engineering, computer science, andmusicology. Publications are scattered among so many sources that evenmany review articles have not surveyed more than a subset of the literature.To ensure that this review was as complete as possible, we searched forrelevant investigations by using a variety of sources. More specifically, thestudies included in the present review were gathered using the followingInternet-based scientific databases: PsycINFO, MEDLINE, Linguistics andLanguageBehavior, Ingenta, andRILMAbstractsof MusicLiterature.Whenever possible, theyear limitswereset at articlespublishedsince1900. The following words, in various combinations and truncations, wereused in the literature search: emotion, affective, vocal, voice, speech,prosody, paralanguage, music, musicperformance, andexpression. Thegoal wastoincludeall Englishlanguagepublicationsinpeer-reviewedjournals. Wehavealsoincludedadditional studieslocatedviainformalsources, includingstudies reportedinconferenceproceedings, inotherlanguages, and in unpublished doctoral dissertations that we were able tolocate. It shouldbenotedthat themajorityof studiesinbothdomainscorrespond to the selection criteria above. We located 104 studies of vocalexpression and 41 studies of music performance in our literature search,which was completed in June 2002.Emotional States and TerminologyWe review the findings in terms of five general categories of emotion:anger, fear, happiness, sadness, andlovetenderness, primarilybecausethese are the onlyfive emotioncategories for whichthere is enoughevidence in both vocal expression and music performance. They roughlycorrespond to the basic emotions described earlier.4These five categoriesrepresent areasonablepoint ofdeparturebecauseall ofthemcomprisewhat are regarded as typical emotions by lay people (Shaver et al., 1987;Shields, 1984). There is also evidence that these emotions closely corre-spondtothefirst emotiontermschildrenlearntouse(e.g., Camras&Allison, 1985)andthattheyserveasbasic-levelcategoriesincognitiverepresentations of emotions (e.g., Shaver et al., 1987). Their role in musicalexpressionof emotionsishighlightedbyquestionnaireresearch(Lind-strom, Juslin, Bresin, &Williamon, 2003)inwhich135musicstudentswere asked what emotions can be expressed in music. Happiness, sadness,fear, love, and anger were among the 10 most highly rated words of a listof 38 words containing both basic and complex emotions.Animportant questionconcernstheexact wordsusedtodenotetheemotions. A number of different words have been used in the literature, andthere is little agreement so far regarding the organization of the emotionlexicon (Plutchik, 1994, p. 45). Therefore, it is not clear how words suchas happiness and joy should be distinguished. The most prudent approachto take is to treat different but closely related emotion words (e.g., sorrow,grief, sadness) as belonging to the same emotion family (e.g., the sadnessfamily; Ekman, 1992). Table 1 shows how the emotion words used in thepresent studies have been categorized in this review (for some empiricalsupport, see the analyses of emotion words presented by Johnson-Laird &Oatley, 1989; Shaver at al., 1987).Studies of Vocal Expression: OverviewDarwin (1872/1998) discussed both vocal and facial expression ofemotionsinhistreatise. Inrecent years, however, facial expressionhasreceivedfarmoreempiricalresearchthanvocalexpression. Thereareanumber of reasons for this, suchas the problems associatedwiththerecording and analysis of speech sounds (Scherer, 1982). The consequenceis that the code used in facial expression of emotion is better understoodthanthecodeusedinvocal expression. This is unfortunate, however,becauserecentstudiesusingself-reportshaverevealedthat, ifanything,4Most theorists distinguishbetweenpassionatelove(eroticism) andcompanionate love (tenderness; Hatfield&Rapson, 2000, p. 660), ofwhichthelattercorrespondstoourlovetendernesscategory. Somere-searchers suggest that all kinds of love originally derived from this emo-tional state, whichisassociatedwithinfantcaregiver attachment (e.g.,Eibl-Eibesfeldt, 1989, chap. 4; Oatley & Jenkins, 1996; p. 287; Panksepp,1998, chap. 13).Table 1Classification of Emotion Words Used by Different Authors Into Emotion CategoriesEmotion category Emotion words used by authorsAnger Aggressive, aggressiveexcitable, aggressiveness, anger, angerhaterage, angry, arger,argerlich, cold anger, cole`re, collera, destruction, frustration, fury, hate, hot anger,irritated, rage, repressed anger, wutFear Afraid, angst, angstlich, anxiety, anxious, fear, fearful, fear of death, fearpain, fearterrorhorror, frightened, nervousness, panic, paura, peur, protection, scared,schreck, terror, worryHappiness Cheerfulness, elation, enjoyment, freude, freudig, gioia, glad, gladquiet, happiness,happy, happycalm, happyexcited, joie, joy, laughtergleemerriment, serenejoyfulSadness Crying despair, depressedsad, depression, despair, gloomytired, grief, quiet sorrow,sad, saddepressed, sadness, sadnessgriefcrying, sorrow, trauer, traurig,traurigkeit, tristesse, tristezzaLovetenderness Affection, liebe, love, lovecomfort, loving, softtender, tender, tenderness, tenderpassion, tenerezza, zartlichkeit776JUSLIN AND LAUKKAvocal expressions may be even more important predictors of emotions thanfacial expressions in everyday life (Planalp, 1998). Fortunately, the field ofvocal expression of emotions has recently seen renewed interest (Cowie,Douglas-Cowie, &Schroder, 2000; Cowie et al., 2001; Johnstone &Scherer, 2000). Thirty-twostudieswerepublishedinthe1990s, andal-ready19studieshavebeenpublishedbetweenJanuary2000andJune2002.Table 2 provides a summary of 104 studies of vocal expression includedinthis reviewinterms of authors, publicationyear, emotions studied,methodused(e.g., portrayal, manipulatedportrayal, induction, naturalspeech sample, synthesis), language, acoustic cues analyzed (where appli-cable), and verbal material. Thirty-nine studies presented data that permit-tedus toincludetheminameta-analysis of communicationaccuracy(detailedbelow). Themajorityof studies(58%) usedEnglish-speakingencoders, althoughas manyas 18different languages, plus nonsenseutterances, are represented in the studies reviewed. Twelve studies (12%)canbecharacterizedasmoreorlesscross-culturalinthattheyincludedanalyses of encoders or decoders from more than one nation. The verbalmaterial features series of numbers, letters of the alphabet, nonsensesyllables, or regular speech material (e.g., words, sentences, paragraphs).The number of emotions included ranges from 1 to 15 (M 5.89). Ninetystudies(87%)usedemotionportrayalsbyactors, 13studies(13%)usedmanipulationsofportrayals(e.g., filtering, masking, reversal), 7studies(7%) used mood induction procedures, and 12 studies (12%) used naturalspeech samples. The latter comes mainly from studies of fear expressionsinaviationaccidents. Twenty-onestudies (20%) usedsoundsynthesis,orcopysynthesis.5Seventy-sevenstudies(74%)reportedacousticdata,of which6studies usedlisteners ratings of cues rather thanacousticmeasurements.Studies of Music Performance: OverviewStudies of music performance have been conducted for more than 100years (for reviews, see Gabrielsson, 1999; Palmer, 1997). However, thesestudieshavealmost exclusivelyfocusedonstructural aspectsofperfor-mance such as marking of the phrase structure, whereas emotion has beenignored. Those studies that have been concerned with emotion in music, onthe other hand, have almost exclusively focused on expressive aspects ofmusical compositionsuchaspitchormode(e.g., Gabrielsson&Juslin,2003), whereas they have ignored aspects of specific performances. Thatperformance aspects of emotional expression did not gain attention muchearlier is strange consideringthat one of the great pioneers inmusicpsychology, Carl E. Seashore, made detailed proposals about such studiesin the 1920s (Seashore, 1927). Seashore (1947) later suggested that musicresearchers couldusethesameparadigmthat hadbeenusedinvocalexpression(i.e., thestandardcontent paradigm)toinvestigatehowper-formers express emotions. However, Seashores (1947) plea went unheard,and he did not publish any study of that kind himself. After slow initialprogress, therewasanincreaseofstudiesinthe1990s(23studiespub-lished). This seems to continue into the 2000s (10 studies published20002002). The increase is perhaps a result of the increased availabilityof software for digital analysis of acoustic cues, but it may also reflect arenaissancefor researchonmusical emotion(Juslin&Sloboda, 2001).Figure1illustratesthetimelinessofthisreviewintermsofthestudiesavailable for a comparison of the two domains.Table 3 provides a summary of the 41 studies of emotional expression inmusic performance included in the review in terms of authors, publicationyear, emotions studied, methodused(e.g., portrayal, manipulatedpor-trayal, synthesis), instrument used, number and nationality of performersand listeners, acoustic cues analyzed (where applicable), and musicalmaterial. Twelve studies (29%) provided data that permitted us to includethem in a meta-analysis of communication accuracy. These studies coveredawiderangeof musical styles, includingclassical music, folkmusic,Indianragas, jazz, pop, rock, childrenssongs, andfreeimprovisations.Themostcommonmusicalstylewasclassicalmusic(17studies, 41%).Most studiesreliedonthestandardparadigmusedinstudiesof vocalexpression of emotions. The number of emotions studied ranges from 3 to 9(M4.98), andemotionstypicallyincludedhappiness, sadness, anger,fear, and tenderness. Twelve musical instruments were included. The mostfrequently studied instrument was singing voice (19 studies), followed byguitar (7), piano (6), synthesizer (4), violin (3), flute (2), saxophone (2),drums (1), sitar (1), timpani (1), trumpet (1), xylophone (1), and sentograph(1asentographisanelectronicdeviceforrecordingpatternsoffingerpressure over timesee Clynes, 1977). At least 12 different nationalitiesare represented in the studies (Fonagy & Magdics, 1963, did not state thenationalities clearly), with Sweden being most strongly represented (39%),followed by Japan (12%) and the United States (12%). Most of the studiesanalyzed professional musicians (but see Juslin & Laukka, 2000), and theperformances were usually monophonic to facilitate measurement ofacoustic parameters (for anexception, see Dry&Gabrielsson, 1997).(Monophonic melodyis probablyone of the earliest forms of music,Wolfe, 2002.) A few studies (15%) investigated what means listeners usetodecodeemotionsbymeansofsynthesizedperformances. Eighty-fivepercent of the studies reported data on acoustic cues; of these studies, 5used listeners ratings of cues rather than acoustic measurements.ResultsDecoding AccuracyStudies of vocal expression and music performance have con-vergedontheconclusionthat encoderscancommunicatebasicemotions to decoders with above-chance accuracy, at least for thefiveemotioncategoriesconsideredhere. Toexaminethesedatacloser, weconductedameta-analysisofdecodingaccuracy. In-cluded in this analysis were all studies that presented (or allowedcomputation of) forced-choice decoding data relative to someindependent criterion of encoding intention. Thirty-nine studies ofvocal expressionand12studiesofmusicperformancemet thiscriterion, featuring a total of 73 decoding experiments, 60 for vocalexpression, 13 for music performance.One problem in comparing accuracy scores from different stud-ies is that they use different numbers of response alternatives in thedecoding task. Rosenthal and Rubins (1989) effect size index forone-sample, multiplechoice-type data, pi (), allows researcherstotransformaccuracyscoresinvolvinganynumberofresponsealternatives to a standard scale of dichotomous choice, on which.50 is the null value and 1.00 corresponds to 100% correct decod-ing. Ideally, an index of decoding accuracy should also take intoaccount theresponsebiasinthedecodersjudgments(Wagner,1993). However, this requires that results be presented in terms ofa confusion matrix, which very few studies have done. Therefore,we summarize the data simply in terms of Rosenthal and Rubinspi index.Summary statistics. Table 4summarizes the mainfindingsfromthemeta-analysisintermsofsummarystatistics(i.e., un-weighted mean, weighted mean, median, standard deviation,(text continues on page 786)5Copy synthesis refers to copying acoustic features from real emotionportrayalsandusingthemtoresynthesizenewportrayals. Thismethodmakes it possible to manipulate certain cues of an emotion portrayal whileleaving other cues intact (e.g., Juslin & Madison, 1999; Ladd et al., 1985;Schroder, 2001).777COMMUNICATION OF EMOTIONSTable2SummaryofStudiesonVocalExpressionofEmotionIncludedintheReviewStudyEmotionsstudied(intermsusedbyauthors)MethodSpeakers/listenersAcousticcuesanalyzedaVerbalmaterialNLanguage1.Abelin&Allwood(2000)Anger,disgust,dominance,fear,happiness,sadness,surprise,shynessP1/93Swe/Eng(12),Swe/Fin(23),Swe/Spa(23),Swe/Swe(35)SR,pauses,F0,IntBriefsentence2.Albasetal.(1976)Anger,happiness,love,sadnessP,M12/80Eng(6)/Eng(20),Cre(20),Cre(6)/Eng(20),Cre(20)Anytwosentencesthatcometomind3.Al-Watban(1998)Anger,fear,happiness,neutral,sadnessP4/14AraSR,F0,IntBriefsentence4/15Eng4.Anolli&Ciceri(1997)Collera[anger],disprezzo[disgust],gioia[happiness],paura[fear],tenerezza[tenderness],tristezza[sadness]P,M2/100ItaSR,pauses,F0,Int,SpectrBriefsentence5.Apple&Hecht(1982)Anger,happiness,sadness,surpriseP,M43/48EngBriefsentence6.Banse&Scherer(1996)Anxiety,boredom,coldanger,contempt,despair,disgust,elation,happiness,hotanger,interest,panic,pride,sadness,shameP12/12Non/GerSR,F0,Int,SpectrBriefsentence7.Baronietal.(1997)Anger,happiness,sadnessP3/42ItaSR,F0,IntBriefsentence8.Baroni&Finarelli(1994)Aggressive,depressedsad,serenejoyfulP3/0ItaSR,IntBriefsentence9.Bergmannetal.(1988)Argerlich[angry],angstlich[afraid],entgegenkommend[accomodating],freudig[glad],gelangweilt[bored],nachdrucklich[insistent],traurig[sad],verachtlich[scornful],vorwurfsvoll[reproachful]S0/88GerSR,F0,F0contour,Int,jitterBriefsentence10.Bonebright(1996)Anger,fear,happiness,neutral,sadnessP6/0EngSR,F0,IntLongparagraph11.Bonebrightetal.(1996)Anger,fear,happiness,neutral,sadness(sameportrayalsasinStudy10)P6/104EngLongparagraph12.Bonner(1943)FearP3/0EngSR,F0,pausesBriefsentence13.Breitensteinetal.(2001)Angry,frightened,happy,neutral,sadP,S1/65Ger/Ger(35),Ger/Eng(30)SR,F0Briefsentence14.Breznitz(1992)Angry,happy,sad,neutral(posthocclassification)N11/0EngF0Responsesfrominterviews15.Brighettietal.(1980)Anger,contempt,disgust,fear,happiness,sadness,surpriseP6/34ItaSeriesofnumbers16.Burkhardt(2001)Boredom,coldanger,cryingdespair,fear,happiness,hotanger,joy,quietsorrowSP0/7210/30GerGerSR,F0,Spectr,formants,Art.,glottalwaveformBriefsentence17.Burns&Beier(1973)Angry,anxious,happy,indifferent,sad,seductiveP30/21EngBriefsentence18.Cahn(1990)Angry,disgusted,glad,sad,scared,surprisedS0/28EngSR,pauses,F0,Spectr,Art.,glottalwaveformBriefsentence19.Carlsonetal.(1992)Angry,happy,neutral,sadP,S1/18SweSR,F0,F0contourBriefsentence20.Chung(2000)Joy,sadnessN1/30Kor/Kor(10),Kor/Eng(10),Kor/Fre(10)SR,F0,F0contour,Spectr,jitter,shimmerResponsesfromTVinterviews5/22EngS0/2021.Costanzoetal.(1969)Anger,contempt,grief,indifference,loveP23/44Eng(SR,F0,Int)Briefparagraph778JUSLIN AND LAUKKAStudyEmotionsstudied(intermsusedbyauthors)MethodSpeakers/listenersAcousticcuesanalyzedaVerbalmaterialNLanguage22.Cummings&Clements(1995)Angry(10nonemotionterms)P2/0EngGlottalwaveformWord23.Davitz(1964a)Admiration,affection,amusement,anger,boredom,cheerfulness,despair,disgust,dislike,fear,impatience,joy,satisfaction,surpriseP7/20EngBriefparagraph24.Davitz(1964b)Affection,anger,boredom,cheerfulness,impatience,joy,sadness,satisfactionP5/5Eng(SR,F0,F0contour,Int,Art.,rhythm,timbre)Briefparagraph25.Davitz&Davitz(1959)Anger,fear,happiness,jealousy,love,nervousness,pride,sadness,satisfaction,sympathyP8/30EngLettersofthealphabet26.Dusenbury&Knower(1939)Amazementastonishmentsurprise,angerhaterage,determinationstubbornessfirmness,doubthesitationquestioning,fearterrorhorror,laughtergleemerriment,pitysympathyhelpfulness,religiouslovereverenceawe,sadnessgriefcrying,sneeringcontemptscorn,torturegreatpainsufferingP8/457EngLettersofthealphabet27.Eldred&Price(1958)Anger,anxiety,depressionN1/4Eng(SR,pauses,F0,Int)Speechfrompsychotherapysessions28.Fairbanks&Hoaglin(1941)Anger,contempt,fear,grief,indifferenceP6/0EngSR,pausesBriefparagraph29.Fairbanks&Provonost(1938)Anger,contempt,fear,grief,indifference(sameportrayalsasinStudy28)P6/64EngF0Briefparagraph30.Fairbanks&Provonost(1939)Anger,contempt,fear,grief,indifference(sameportrayalsasinStudy28)P6/64EngF0,F0contour,pausesBriefparagraph31.Fensteretal.(1977)Anger,contentment,fear,happiness,love,sadnessP5/30EngBriefsentence32.Fonagy(1978)Anger,coquetry,disdain,fear,joy,longing,repressedanger,reproach,sadness,tendernessP,M1/58FreF0,F0contourBriefsentence33.Fonagy&Magdics(1963)Anger,complaint,coquetry,fear,joy,longing,sarcasm,scorn,surprise,tendernessP/0Hun,Fre,Ger,EngSR,F0,F0contour,Int,timbreBriefsentence34.Frick(1986)Anger(frustration,threat),disgustP1/37EngF0Briefsentence35.Friend&Farrar(1994)Angry,happy,neutralP,M1/99EngF0,Int,SpectrBriefsentence36.Grding&Abramson(1965)Anger,surprise,neutral(2nonemotionterms)P5/5EngF0,F0contourBriefsentence,digitsS0/1137.Gerard&Clement(1998)Happiness,irony,sadness(2nonemotionterms)P3/0FreSR,F0,F0contourBriefsentence1/2038.Gobl&NChasaide(2000)Afraid/unafraid,bored/interested,content/angry,friendly/hostile,relaxed/stressed,sad/happy,timid/confidentS0/8SweInt,jitter,formants,Spectr,glottalwaveformBriefsentence39.Grahametal.(2001)Anger,depression,fear,hate,joy,nervousness,neutral,sadnessP4/177Eng/Eng(85),Eng/Jap(54),Eng/Spa(38)Longparagraph40.Greasleyetal.(2000)Anger,disgust,fear,happiness,sadnessN32/158EngSpeechfromTVandradio(tablecontinues)779COMMUNICATION OF EMOTIONSTable2(continued)StudyEmotionsstudied(intermsusedbyauthors)MethodSpeakers/listenersAcousticcuesanalyzedaVerbalmaterialNLanguage41.Guidetti(1991)Cole`re[anger],joie[joy],neutre[neutral],peur[fear],tristesse[sadness]P4/50Non/FreBriefsentence42.Havrdova&Moravek(1979)Anger,anxiety,joyI6/0CzeSR,F0,Int,SpectrBriefsentence43.Hoffe(1960)Arger[anger],enttauschung[disappointment],erleichterung[relief],freude[joy],schmerz[pain],schreck[fear],trost[comfort],trotz[defiance],wohlbehagen[satisfaction],zweifel[doubt]P4/0GerF0,IntWord44.House(1990)Angry,happy,neutral,sadP,M,S1/11SweF0,F0contour,IntBriefsentence45.Huttar(1968)Afraid/bold,angry/pleased,sad/happy,timid/confident,unsure/sureN1/12EngSR,F0,IntSpeechfromclassroomdiscussions46.Iidaetal.(2000)Anger,joy,sadnessI2/0JapSR,F0,IntParagraphS0/3647.Iriondoetal.(2000)Desire,disgust,fear,fury,joy,sadness,surprise(threelevelsofintensity)P8/1,054SpaSR,pauses,F0,Int,SpectrParagraphS0/48.Joetal.(1999)Afraid,angry,happy,sadP1/KorSR,F0BriefsentenceS0/2049.Johnsonetal.(1986)Anger,fear,joy,sadnessP,S1/21EngBriefsentenceM1/2350.Johnstone&Scherer(1999)Anxious,bored,depressed,happy,irritated,neutral,tenseP8/0FreF0,Int,Spect,jitter,glottalwaveformBriefsentence,phoneme,numbers51.Juslin&Laukka(2001)Anger,disgust,fear,happiness,noexpression,sadness(twolevelsofintensity)P8/45Eng(4),Swe(4)/Swe(45)SR,pauses,F0,F0contour,Int,attack,formants,Spectr,jitter,Art.Briefsentence52.L.Kaiser(1962)Cheerfulness,disgust,enthusiasm,grimness,kindness,sadnessP8/51DutSR,F0,F0contour,Int,formants,SpectrVowels53.Katz(1997)Anger,disgust,happycalm,happyexcited,neutral,sadnessI100/0EngSR,F0,IntBriefsentence54.Kienast&Sendlmeier(2000)Anger,boredom,fear,happiness,sadnessP10/20GerSpectr,formants,Art.Briefsentence55.Kitahara&Tohkura(1992)Anger,joy,neutral,sadnessP1/8JapSR,F0,IntBriefsentenceM1/7S0/2756.Klasmeyer&Sendlmeier(1997)Anger,boredom,disgust,fear,happiness,sadnessP3/20GerF0,Int,Spectr,jitter,glottalwaveformBriefsentence57.Knower(1941)Amazementastonishmentsurprise,angerhaterage,determinationstubbornnessfirmness,doubthesitationquestioning,fearterrorhorror,laughtergleemerriment,pitysympathyhelpfulness,religiouslovereverenceawe,sadnessgriefcrying,sneeringcontemptscorn,torturegreatpainsufferingP,M8/27EngLettersofthealphabet780JUSLIN AND LAUKKATable2(continued)StudyEmotionsstudied(intermsusedbyauthors)MethodSpeakers/listenersAcousticcuesanalyzedaVerbalmaterialNLanguage58.Knower(1945)Amazementastonishmentsurprise,angerhaterage,determinationstubbornnessfirmness,doubthesitationquestioning,fearterrorhorror,laughtergleemerriment,pitysympathyhelpfulness,religiouslovereverenceawe,sadnessgriefcrying,sneeringcontemptscorn,torturegreatpainsufferingP169/15EngLettersofthealphabet59.Kramer(1964)Anger,contempt,grief,indifference,loveP,M10/27Eng(7)/Eng,Jap(3)/EngBriefparagraph60.Kurodaetal.(1976)Fear(emotionalstress)N14/0EngRadiocommunication(flightaccidents)61.Laukkanenetal.(1996)Anger,enthusiasm,neutral,sadness,surpriseP3/25Non/FinF0,Int,glottalwaveform,subglottalpressureBriefsentence62.Laukkanenetal.(1997)Anger,enthusiasm,neutral,sadness,surprise(sameportrayalsasinStudy61)P,S3/10Non/FinGlottalwaveform,formantsBriefsentence63.Leinonenetal.(1997)Admiring,angry,astonished,commanding,frightened,naming,pleading,sad,satisfied,scornfulP12/73FinSR,F0,F0contour,Int,SpectrWord64.Leon(1976)Admiration[admiration],cole`re[anger],joie[joy],ironie[irony],neutre[neutral],peur[fear],surprise[surprise],tristesse[sadness]P1/20FreSR,pauses,F0,F0contour,IntBriefsentence65.Levin&Lord(1975)Destruction(anger),protection(fear)P5/0EngF0,SpectrWord66.Levitt(1964)Anger,contempt,disgust,fear,joy,surpriseP50/8EngBriefparagraph67.Lieberman(1961)Bored,confidential,disbelief/doubt,fear,happiness,objective,pompousP6/20EngJitterBriefsentence68.Lieberman&Michaels(1962)Boredom,confidential,disbelief,fear,happiness,pompous,question,statementP6/20EngF0,F0contour,IntBriefsentenceM3/6069.Markeletal.(1973)Anger,depressionI50/0Eng(SR,F0,Int)ResponsestotheThematicApperceptionTest70.Moriyama&Ozawa(2001)Anger,fear,joy,sorrowP1/8JapSR,F0,IntWord71.Mozziconacci(1998)Anger,boredom,fear,indignation,joy,neutrality,sadnessP3/10DutSR,F0,F0contour,rhythmBriefsentenceS0/5272.Murray&Arnott(1995)Anger,disgust,fear,grief,happiness,sadnessS0/35EngSR,pauses,F0,F0contour,Art.,Spectr,glottalwaveformBriefsentence,paragraph73.Novak&Vokral(1993)Anger,joy,sadness,neutralP1/2CzeF0,SpectrLongparagraph74.Paeschke&Sendlmeier(2000)Anger,boredom,fear,happiness,sadnessP10/20GerF0,F0contourBriefsentence75.Pell(2001)Angry,happy,sadP10/10EngSR,F0,F0contourBriefsentence76.Pfaff(1954)Disgust,doubt,excitement,fear,grief,hate,joy,love,pleading,shameP1/304EngNumbers(tablecontinues)781COMMUNICATION OF EMOTIONSTable2(continued)StudyEmotionsstudied(intermsusedbyauthors)MethodSpeakers/listenersAcousticcuesanalyzedaVerbalmaterialNLanguage77.Pollacketal.(1960)Anger,approval,boredom,confidentiality,disbelief,disgust,fear,happiness,impatience,objective,pedantic,sarcasm,surprise,threat,uncertaintyPM4/184/28EngBriefsentence78.Protopapas&Lieberman(1997)TerrorNS1/00/50EngF0,jitterRadiocommunication(flightaccidents)79.Roessler&Lester(1976)Anger,affect,fear,depressionN1/3EngF0,Int,formantsSpeechfrompsychotherapysessions80.Scherer(1974)Anger,boredom,disgust,elation,fear,happiness,interest,sadness,surpriseS0/10Non/AmSR,F0,F0contour,IntTonesequences81.Schereretal.(2001)Anger,disgust,fear,joy,sadnessP4/428Non/Dut(60),Non/Eng(72),Non/Fre(96),Non/Ger(70),Non/Ita(43),Non/Ind(38),Non/Spa(49)Briefsentence82.Schereretal.(1991)Anger,disgust,fear,joy,sadnessP4/454Non/GerSR,F0,Int,SpectrBriefsentence83.Scherer&Oshinsky(1977)Anger,boredom,disgust,fear,happiness,sadness,surpriseS0/48Non/AmSR,F0,F0contour,Int,attack,SpectrTonesequences84.Schroder(1999)Anger,fear,joy,neutral,sadnessP3/4GerBriefsentenceS0/1385.Schroder(2000)Admiration,boredom,contempt,disgust,elation,hotanger,relief,startle,threat,worryP6/20GerAffectbursts86.Sedlacek&Sychra(1963)Angst[fear],einfacheraussage[statement],feierlichkeit[solemnity],freude[joy],ironie[irony],komik[humor],liebesgefuhle[love],trauer[sadness]I23/70CzeF0,F0contourBriefsentence87.Simonovetal.(1975)Anxiety,delight,fear,joyP57/0RusF0,formantsVowels88.Skinner(1935)Joy,sadnessI19/0EngF0,Int,SpectrWord89.Sobin&Alpert(1999)Anger,fear,joy,sadnessI31/12EngSR,pauses,F0,IntBriefsentence90.Sogon(1975)Anger,contempt,grief,indifference,loveP4/10JapBriefsentence91.Steffen-Batogetal.(1993)Amazement,anger,boredom,joy,indifference,irony,sadnessP4/70Pol(SR,F0,Int,voicequality)Briefsentence92.Stibbard(2001)Anger,disgust,fear,happiness,sadness(samespeechmaterialasinStudy40)N32/EngSR,(F0,F0contour,Int,Art.,voicequality)SpeechfromTV93.Sulc(1977)Fear(emotionalstress)N9/0EngF0Radiocommunication(flightaccidents)94.Tickle(2000)Angry,calm,fearful,happy,sadP6/24Eng(3),Jap(3)/Eng(16),Jap(8)Briefsentence,vowels95.Tischer(1993,1995)Abneigung[aversion],angst[fear],arger[anger],freude[joy],liebe[love],lust[lust],sehnsucht[longing],traurigkeit[sadness],uberraschung[surprise],unsicherheit[insecurity],wut[rage],zartlichkeit[tenderness],zufriedenheit[contentment],zuneigung[sympathy]P4/931GerSR,Pauses,F0,IntBriefsentence,paragraph782JUSLIN AND LAUKKATable2(continued)StudyEmotionsstudied(intermsusedbyauthors)MethodSpeakers/listenersAcousticcuesanalyzedaVerbalmaterialNLanguage96.Trainoretal.(2000)Comfort,fear,love,surpriseP23/6EngSR,F0,F0contour,rhythmBriefsentence97.vanBezooijen(1984)Anger,contempt,disgust,fear,interest,joy,neutral,sadness,shame,surpriseP8/0DutSR,F0,Int,Spectr,jitter,Art.Briefsentence98.vanBezooijenetal.(1983)Anger,contempt,disgust,fear,interest,joy,neutral,sadness,shame,surprise(sameportrayalsasinStudy97)P8/129Dut/Dut(48),Dut/Tai(40),Dut/Chi(41)Briefsentence99.Wallbott&Scherer(1986)Anger,joy,sadness,surpriseP6/11GerSR,F0,IntBriefsentenceM6/10100.Whiteside(1999a)Coldanger,elation,happiness,hotanger,interest,neutral,sadnessP2/0EngSR,formantsBriefsentence101.Whiteside(1999b)Coldanger,elation,happiness,hotanger,interest,neutral,sadness(sameportrayalsasinStudy100)P2/0EngF0,Int,jitter,shimmerBriefsentence102.Williams&Stevens(1969)FearN3/0EngF0,jitterRadiocommunication(flightaccidents)103.Williams&Stevens(1972)Anger,fear,neutral,sorrowPN4/01/0EngEngSR,pauses,F0,F0contour,Spectr,jitter,formants,Art.Briefsentence104.Zuckermanetal.(1975)Anger,disgust,fear,happiness,sadness,surpriseP40/61EngBriefsentenceNote.Adashindicatesthatnoinformationwasprovided.Typesofmethodincludedportrayal(P),manipulatedportrayal(M),synthesis(S),naturalspeechsample(N),andinductionofemotion(I).SweSwedish;EngEnglish;FinFinnish;SpaSpanish;SRspeechrate;F0fundamentalfrequency;Intvoiceintensity;CreCree-speakingCanadianIndians;AraArabic;ItaItalian;Spectrcuesrelatedtospectralenergydistribution(e.g.,high-frequencyenergy);NonNonsenseutterances;GerGerman;Art.precisionofarticulation;KorKorean;FreFrench;HunHungarian;AmAmerican;JapJapanese;CzeCzechoslovakian;DutDutch;IndIndonesian;RusRussian;PolPolish;TaiTaiwanese;ChiChinese.aAcousticcueslistedwithinparentheseswereobtainedbymeansoflistenerratingsratherthanacousticmeasurements.783COMMUNICATION OF EMOTIONSTable2(continued)Table3SummaryofStudiesonMusicalExpressionofEmotionIncludedintheReviewStudyEmotionsstudied(intermsusedbyauthors)MethodNInstrument(n)Nat.AcousticcuesanalyzedaMusicalmaterialp/lp/l1.Arcosetal.(1999)Aggressive,calm,joyful,restless,sad,tenderP,S1/SaxophoneSpaTempo,timing,Int,Art.,attack,vibratoJazzballads2.Baars&Gabrielsson(1997)Angry,fearful,happy,noexpression,sad,solemn,tenderP1/9SingingSweTempo,timing,Int,Art.Folkmusic3.Balkwill&Thompson(1999)Anger,joy,peace,sadnessP2/30Flute(1),sitar(1)Ind/Can(Tempo,pitch,melodicandrhythmiccomplexity)Indianragas4.Baronietal.(1997)Aggressive,saddepressed,serenejoyfulP3/42SingingItaTempo,timing,Int,pitchOpera5.Baroni&Finarelli(1994)Aggressive,saddepressed,serenejoyfulP3/0SingingItaTiming,IntOpera6.Behrens&Green(1993)Angry,sad,scaredP8/58Singing(2),trumpet(2),violin(2),timpani(2)AmImprovisations7.Bresin&Friberg(2000)Anger,fear,happiness,noexpression,sadness,solemnity,tendernessS0/20PianoSweTempo,timing,Int,Art.Childrenssong,classicalmusic(romantic)8.Bunt&Pavlicevic(2001)Angry,fearful,happy,sad,tenderP24/24VarEng(Tempo,timing,Int,timbre,Art.,pitch)Improvisations9.Canazza&Orio(1999)Activeanimated,aggressiveexcitable,calmoppressive,gladquiet,gloomytired,powerfulrestlessP2/17Saxophone(1),piano(1)ItaTempo,Int,Art.Jazz10.Dry&Gabrielsson(1997)Angry,fearful,happy,neutral,sad,solemn,tenderP4/20RockcomboSwe(Tempo,timing,timbre,attack)Rockmusic11.Ebie(1999)Anger,fear,happiness,sadnessP56/3SingingAmNewlycomposedmusic(classical)12.Fonagy&Magdics(1963)Anger,complaint,coquetry,fear,joy,longing,sarcasm,scorn,surpriseP/Var(Tempo,Int,timbre,Art.,pitch)Folkmusic,classicalmusic,opera13.Gabrielsson&Juslin(1996)Angry,fearful,happy,noexpression,sad,solemn,tenderP6/37ElectricguitarSweTempo,timing,Int,Spectr,Art.,pitch,attack,vibratoFolkmusic,classicalmusic,popularmusic3/56Flute(1),violin(1),singing(1)14.Gabrielsson&Lindstrom(1995)Angry,happy,indifferent,softtender,solemnP4/110Synthesizer,sentographSweTempo,timing,Int,Art.Folkmusic,popularmusic15.Jansensetal.(1997)Anger,fear,joy,neutral,sadnessP14/25SingingDutTempo,Int,F0,Spectr,vibratoClassicalmusic(lied)16.Juslin(1993)Anger,happiness,solemnity,tenderness,noexpressionP3/13ElectricguitarSweTempo,timing,Int,timbre,Art.,attackFolkmusic17.Juslin(1997a)Anger,fear,happiness,sadness,tendernessP,S1/15Electricguitar(1),synthesizerSweFolkmusic18.Juslin(1997b)Anger,fear,happiness,noexpression,sadnessP3/24ElectricguitarSweTempo,timing,Int,Art.,attackJazz19.Juslin(1997c)Anger,fear,happiness,sadness,tendernessP,M,S1/12Electricguitar(1),synthesizerSweTempo,timing,Int,Spectr,Art.,attack,vibratoFolkmusicS0/42Synthesizer20.Juslin(2000)Anger,fear,happiness,sadnessP3/30ElectricguitarSweTempo,Int,Spectr,Art.Jazz,folkmusic784JUSLIN AND LAUKKAStudyEmotionsstudied(intermsusedbyauthors)MethodNInstrument(n)Nat.AcousticcuesanalyzedaMusicalmaterialp/lp/l21.Juslinetal.(2002)SadnessS0/12PianoSweTempo,Int,Art.Ballad22.Juslin&Laukka(2000)Anger,fear,happiness,sadnessP8/50ElectricguitarSweTempo,timing,Int,Spectr,Art.Jazz,folkmusic23.Juslin&Madison(1999)Anger,fear,happiness,sadnessP,M3/20PianoSweTempo,timing,Int,Art.Jazz,folkmusic24.Konishietal.(2000)Anger,fear,happiness,sadnessP10/25SingingJapVibratoClassicalmusic25.Kotlyar&Morozov(1976)Anger,fear,joy,neutral,sorrowP11/10SingingRusTempo,timing,Art.,Int,attackClassicalmusicM0/1126.Langeheineckeetal.(1999)Anger,fear,joy,sadnessP11/0SingingGerTempo,timing,Int,Spectr,vibratoClassicalmusic27.Laukka&Gabrielsson(2000)Angry,fearful,happy,noexpression,sad,solemn,tenderP2/13DrumsSweTempo,timing,IntRhythmpatterns(jazz,rock)28.Madison(2000b)Anger,fear,happiness,sadnessP,M3/10PianoSweTiming,Int,Art.(patternsofchanges)Popularmusic29.Mergletal.(1998)Freude[joy],trauer[sadness],wut[rage]P20/74XylophoneGer(Tempo,Int,Art.,attack)Improvisations30.Metfessel(1932)Anger,fear,grief,loveP1/0SingingAmVibratoPopularmusic31.Morozov(1996)Fear,hotanger,joy,neutral,sorrowP,M/SingingRusSpectr,formantsClassicalmusic,pop,hardrock32.Ohgushi&Hattori(1996a)Anger,fear,joy,neutral,sorrowP3/0SingingJapTempo,F0,Int,vibratoClassicalmusic33.Ohgushi&Hattori(1996b)Anger,fear,joy,neutral,sorrowP3/10SingingJapClassicalmusic34.Oura&Nakanishi(2000)Anger,happiness,sadnessP1/30PianoJapTempo,IntClassicalmusic35.Rapoport(1996)Calm,excited,expressive,intermediate,neutral-soft,short,transitionalmultistage,virtuosobP/0SingingVarF0,intonation,vibrato,formantsClassicalmusic,opera36.Salgado(2000)Angry,fearful,happy,neutral,sadP3/6SingingPorInt,SpectrClassicalmusic(lied)37.Scherer&Oshinsky(1977)Anger,boredom,disgust,fear,happiness,sadness,surpriseS0/48SynthesizerAmTempo,F0,F0contour,Int,attack,SpectrClassicalmusic38.Senju&Ohgushi(1987)Sad(9nonemotionterms)P1/16ViolinJapClassicalmusic39.Sherman(1928)Angerhate,fearpain,sorrow,surpriseP,M1/30SingingAm40.Siegwart&Scherer(1995)Fearofdeath,madness,sadness,tenderpassionP5/11SingingInt,SpectrOpera41.Sundbergetal.(1995)Angry,happy,hateful,loving,sad,scared,secureP1/5SingingSweTempo,timing,Int,Spectr,pitch,attack,vibratoClassicalmusic(lied)Note.Adashindicatesthatnoinformationwasprovided.Typesofmethodincludedportrayal(P),synthesis(S),andmanipulatedportrayal(M).pperformers;llisteners;Nat.nationality;SpaSpanish;Intintensity;Art.articulation;SweSwedish;IndIndian;CanCanadian;ItaItalian;AmAmerican;Varvarious;EngEnglish;Spectrcuesrelatedtospectralenergydistribution(e.g.,high-frequencyenergy);DutDutch;F0fundamentalfrequency;JapJapanese;RusRussian;GerGerman;PorPortuguese.aAcousticcueslistedwithinparentheseswereobtainedbymeansoflistenerratingsratherthanacousticmeasurements.bThesetermsdenoteparticularmodesofsinging,whichinturnareusedtoexpressdifferentemotions.785COMMUNICATION OF EMOTIONSTable3(continued)range) of pi values, as well as confidence intervals.6Also indicatedis the number of encoders (speakers or performers) and studies onwhichtheestimatesarebased. Theestimatesforwithin-culturalvocal expression are generally based on more data than are thosefor cross-cultural vocal expressionandmusicperformance. Asseen in Table 4, overall decoding accuracy is high for all three setsofdata(.84.90). Indeed, theconfidenceintervalssuggestthat decoding accuracy is typically significantly higher than whatwould be expected by chance alone ( .50) for all three types ofstimuli. The lowest estimate of overall accuracy in any of the 73decoding experiments was .69 (Fenster, Blake, &Goldstein,1977). Overall decoding accuracy across within-cultural vocalexpression and music performance was .89, which is equivalent toarawaccuracyscoreof .70inaforced-choicetaskwithfiveresponsealternatives(theaveragenumberofalternativesacrossbothchannels; see, e.g., Table1ofRosenthal &Rubin, 1989).However, overall accuracy was significantly higher, t(58) 3.14,p .01, for within-cultural vocal expression ( .90) than forcross-cultural vocal expression (.84). The differences inoverall accuracy between music performance (.88) andwithin-cultural vocal expression and the differences between mu-sic performance and cross-cultural vocal expression were not sig-nificant. The results indicate that musical expression of emotionswasaboutasaccurateasvocalexpressionofemotionsandthatvocal expressionof emotions was cross-culturallyaccurate, al-though cross-cultural accuracy was 7% lower than within-culturalaccuracy in the present results. Note also that decoding accuracyfor vocal expressionwas well abovechancefor bothemotionportrayals and natural expressions.The patterns of accuracy estimates for individual emotions aresimilar across the three sets of data. Specifically, anger ( .88,M .91) and sadness ( .91, M .92) portrayals were bestdecoded, followedbyfear (.82, M.86) andhappinessportrayals(.74, M.82). Worst decodedthroughout wastenderness ( .71, M .78), although it must be noted that theestimates for this emotion were based on fewer data points. Furtheranalysis confirmed that, across channels, anger and sadness weresignificantlybetter communicated(t tests, p.001) thanfear,happiness, and tenderness (remaining differences were not signif-icant). This pattern of results is consistent with previous reviews ofvocal expressionfeaturingfewer studies(Johnstone&Scherer,2000) but differs fromthe pattern found in studies of facialexpressionof emotion, inwhichhappiness was usuallybetterdecoded than other emotions (Elfenbein & Ambady, 2002).The standard deviation of decoding accuracy across studies wasgenerallysmall, withthelargest beingfor tendernessinmusicperformance.(Thisisalsoindicatedbythesmallconfidencein-tervalsfor all emotionsexcept tendernessinthecaseof musicperformance.) Thisfindingissurprising; onewouldexpect theaccuracy to vary considerably depending on the emotions studied,theencoders, theverbal or musical material, thedecoders, theprocedure, andsoon. Yet, thepresent results suggest that theestimatesofdecodingaccuracyarefairlyrobust withrespect tothesefactors. Considerationofthedifferent measuresofcentraltendency (unweighted mean, weighted mean, and median) showsthat they differed little and that all indices gave the same patternsoffindings. Thissuggeststhatthedatawererelativelyhomoge-nous. This impression is confirmed by plotting the distribution ofdata on decoding accuracy for vocal expression and music perfor-mance (see Figure 2). Only eight (11%) of the experiments yieldedaccuracyestimatesbelow.80.Theseincludethreecross-culturalvocalexpressionexperiments(twothatinvolvednaturalexpres-sion), four vocal expression experiments using emotion portrayals,andonemusicperformanceexperiment usingdrumplayingasstimuli.Possible moderators. Although the decoding data appear to berelativelyhomogenous, weinvestigatedpossiblemoderators ofdecoding accuracy that could explain the variability. Among themoderatorsweretheyearofthestudy, numberofemotionsen-coded (this coincided with the number of response alternatives inthepresent dataset), numberofencoders, numberofdecoders,recordingmethod(dummycoded, 0portrayal, 1naturalsample), response format (0 forced choice, 1 rating scales),laboratory(dummycodedseparatelyfor Knower, Scherer, andJuslin labs; see Table 5), and channel (dummy coded separately forcross-culturalvocalexpression,within-culturalvocalexpression,and music performance). Table 5 presents the correlations amongthe investigated moderators as well as their correlations withoverall decoding accuracy. Note that overall accuracy was nega-tively correlated with year of the study, use of natural expressions(recording method), and cross-cultural vocal expression, whereasitwaspositivelycorrelatedwithnumberofemotions.Thelatterfindingis surprisinggiventhat onewouldexpect accuracytodecreaseasthenumber of responsealternativesincreases(e.g.,Rosenthal, 1982). One possible explanation is that certain earlierstudies(e.g., thosebyKnowerslaboratory) reportedveryhighaccuracy estimates (for Knowers studies, mean .97) althoughthey used a large number of emotions (see Table 5). Subsequentstudiesfeaturingmanyemotions(e.g., Banse&Scherer, 1996)have reported lower accuracy estimates. The slightly lower overallaccuracyfor music performance thanfor within-cultural vocalexpression could be related to the fact that more studies of musicperformancethanstudiesofvocalexpressionusedratingscales,whichtypicallyyieldloweraccuracy.Ingeneral,itissurprising6The mean was weighted with regard to the number of encodersincluded.Figure 1. Number of studies of communication of emotions published forvocal expression and music performance, respectively, between 1930 and2000.786JUSLIN AND LAUKKAthat recent studies tended to include fewer encoders, decoders, andemotions.A simultaneous multiple regression analysis (Cohen & Cohen,1983)withoveralldecodingaccuracyasthedependentvariableand six moderators (year of study, number of emotions, recordingmethod, responseformat, Knowerlaboratory, andcross-culturalvocal expression) as independent variables yielded a multiplecorrelation of .58 (adjusted R2 .27, F[6, 64] 5.42, p .001;N71with2outliers, standardresidual 2 , removed).Cross-cultural vocal expression yielded a significant beta weight( .38, p .05), but Knower laboratory ( .20), responseformat (.19), recordingmethod(.18), number ofemotions ( .17), and year of the study ( .07) did not. Theseresults indicate that only about 30% of the variability in decodingdata can be explained by the investigated moderators.7Individual differences. The present results indicate that com-municationof emotions invocal expressionandmusicperfor-mance was relatively accurate. The accuracy (mean across datasets .87) was well beyondthefrequentlyusedcriterionforcorrect responseinpsychophysical research(proportioncorrect[Pc] .75), which is midway between the levels of pure guessing(Pc .50) and perfect detection (Pc 1.00; Gordon, 1989, p. 26).However, studies in both domains have yielded evidence of con-siderableindividual differencesinbothencodinganddecodingaccuracy (see Banse & Scherer, 1996; Gabrielsson & Juslin, 1996;Juslin, 1997b; Juslin & Laukka, 2001; Scherer, Banse, Wallbott, &Goldbeck, 1991; Wallbott & Scherer, 1986; for a review of genderdifferences, seeHall, Carter, &Horgan, 2001). Particularly, en-coders differ widely in their ability to portray specific emotions.This problem has probably contributed to the noted inconsistencyof data concerning code usage in earlier research (Scherer, 1986).Because many researchers have not taken this problem seriously,several studieshaveinvestigatedonlyonespeakerorperformer(see Tables 2 and 3).Individual differences indecodingaccuracyhave alsobeenreported, thoughtheytendtobelesspronouncedthanthoseinencoding. Moreover, even when decoders make incorrect re-sponses their errors are not entirely random. Thus, error distribu-tionsareinformativeabout thesubjectivesimilarityof variousemotional expressions (Davitz, 1964a; van Bezooijen, 1984). It isof interest that the errors made in emotion decoding are similar forvocal expressionandmusicperformance. For instance, sadnessandtendernessarecommonlyconfused, whereashappinessandsadness are seldom confused (Baars & Gabrielsson, 1997; Davitz,1964a; Davitz & Davitz, 1959; Dawes & Kramer, 1966; Fonagy,1978; Juslin, 1997c). Similarerrorpatternsinthetwodomains7It may be argued that in many studies of vocal expression, estimatesare likely to be biased because of preselection of effective portrayals beforeinclusion in decoding experiments. However, whether preselection ofportrayals is a moderator of overall accuracy was not examined becauseonly a minority of studies stated clearly the extent of preselection carriedout. However, it should be noted that decoding accuracy of a comparablelevelhasbeenfoundinstudiesthatdidnotusepreselectionofemotionportrayals (Juslin & Laukka, 2001).Table 4Summary of Results From Meta-Analysis of Decoding Accuracy for Discrete Emotions in Terms of Rosenthal and Rubins (1989) PiCategoryEmotionOverall Anger Fear Happiness Sadness TendernessWithin-cultural vocal expressionMean (unweighted) .93 .88 .87 .93 .82 .9095% confidence interval .021 .037 .040 .020 .083 .023Mean (weighted) .91 .88 .83 .93 .83 .90Median .95 .90 .92 .94 .85 .92SD .059 .095 .111 .056 .079 .072Range .771.00 .651.00 .511.00 .801.00 .69.89 .691.00No. of studies 32 26 30 31 6 38No. of speakers 278 273 253 225 49 473Cross-cultural vocal expressionMean (unweighted) .91 .82 .74 .91 .71 .8495% confidence interval .017 .062 .040 .018 .024Mean (weighted) .90 .82 .74 .91 .71 .85Median .90 .88 .73 .91 .84SD .031 .113 .077 .036 .047Range .86.96 .55.93 .61.90 .82.97 .74.90No. of studies 6 5 6 7 1 7No. of speakers 69 66 68 71 3 71Music performanceMean (unweighted) .89 .87 .86 .93 .81 .8895% confidence interval .067 .099 .068 .043 .294 .043Mean (weighted) .86 .82 .85 .93 .86 .88Median .89 .88 .87 .95 .83 .88SD .094 .118 .094 .061 .185 .071Range .741.00 .691.00 .681.00 .791.00 .561.00 .75.98No. of studies 10 8 10 10 4 12No. of performers 70 47 70 70 9 79787COMMUNICATION OF EMOTIONSprovide a first indication that there could be similarities betweenthe two channels in terms of acoustic cues.Developmental trends. The development of the ability to de-code emotions from auditory stimuli has not been well researched.Recentevidence,however,indicatesthatchildrenasyoungas4years old are able to decode basic emotions from vocal expressionwithbetter thanchanceaccuracy(Baltaxe, 1991; Friend, 2000;J. B. Morton & Trehub, 2001), at least when the verbal content ismadeunintelligiblebyusingutterancesinaforeignlanguageorfilteringout theverbal information(Friend, 2000).8Theabilityseemstoimprovewithage, however, at least until school age(Dimitrovsky, 1964; Fenster et al., 1977; McCluskey&Albas,1981; McCluskey, Albas, Niemi, Cuevas, &Ferrer, 1975) andperhaps even until early adulthood (Brosgole & Weisman, 1995;McCluskey & Albas, 1981).Similarly, studies of music suggest that children as young as 3or 4 years old are able to decode basic emotions from music withbetter than chance accuracy (Cunningham & Sterling, 1988; Dol-gin & Adelson, 1990; Kastner & Crowder, 1990). Although few ofthese studies have distinguished between features of performance(e.g., tempo, timbre) andfeaturesof composition(e.g., mode),DallaBella, Peretz, Rousseau, andGosselin(2001) foundthat5-year-oldswereabletousetempo(i.e., performance) but notmode(i.e., composition)todecodeemotionsinmusical pieces.Again, decoding accuracy seems to improve with age (Adachi &Trehub, 2000; Brosgole & Weisman, 1995; Cunningham & Ster-ling, 1988; Terwogt & van Grinsven, 1988, 1991; but for excep-tions, see Giomo, 1993; Kratus, 1993). It is interesting to note thatthedevelopmental curveover thelifespanappearssimilar forvocal expression and music but differs from that of facial expres-sion. Inacross-sectional study, BrosgoleandWeisman(1995)foundthattheabilitytodecodeemotionsfromvocalexpressionandmusicimprovedduringchildhoodandremainedasymptoticthrough age 43. Then, it began to decline from middle age onward(see alsoMcCluskey&Albas, 1981). It is hardtodeterminewhether emotion decoding occurs in children younger than 2 yearsold, as they are unable to talk about their experiences. However,there is preliminaryevidence that infants are at least able todiscriminate between some emotions in vocal and musical expres-sions (see Gentile, 1998; Mastropieri & Turkewitz, 1999; Nawrot,2003; Singh, Morgan, & Best, 2002; Soken & Pick, 1999; Svejda,1982).Code UsageMost early studies of vocal expression and music performancewere mainly concerned with demonstrating that communication ofemotions is possibleat all. However, if onewants toexplorecommunication as a process, one cannot ignore its mechanisms, inparticular thecodethat carriestheemotional meaning. Alargenumberofstudieshaveattemptedtodescribethecuesusedbyspeakers and musicians to communicate specific emotions to lis-teners. Most studies to date have measured only a small number ofcues, but some recent studies have been more inclusive (seeTables2and3). Beforetakingacloserlookat thepatternsofacoustic cues used to express discrete emotions in vocal expressionandmusicperformance, respectively, weneedtoconsider thevarious cues that were used in each modality. Table 6 shows howeach acoustic cue was defined and measured. The measurementswereusuallycarriedout usingadvancedcomputersoftwarefordigital analysis of speech signals. The cues extracted involve thebasic dimensions of frequency, intensity, and duration, plus vari-8This is because the verbal content may interfere with the decoding ofthe nonverbal content in small children (Friend, 2000).Figure 2. The distributions of point estimates of overall decoding accuracy in terms of Rosenthal and Rubins(1989) pi for vocal expression and music performance, respectively.788JUSLIN AND LAUKKAous combinations of these dimensions (see Table 6). For a moreextensivediscussionof theprinciplesunderlyingproductionofspeech and music and associated measurements, see Borden, Har-ris, and Raphael (1994) and Sundberg (1991), respectively. In thefollowing, we divide data into three sets: (a) cues that are commontovocal expressionandmusic performance, (b) cues that arespecific to vocal expression, and (c) cues that are specific to musicperformance. Thecommoncuesareofmainimportancetothisreview, althoughchannel-specific cues maysuggest additionalaspects of potential overlap that can be explored in future research.Comparisonsof commoncues. Table7presentspatternsofacoustic cues used to express different emotions as reported in 77studies of vocal expression and 35 studies of music performance.Veryfewstudies have reporteddata insuchdetail topermitinclusion in a meta-analysis. Furthermore, it is usually difficult tocompare quantitative data across different studies because studiesuse different baselines (Juslin & Laukka, 2001, p. 406).9The mostprudent approachwastosummarizefindingsintermsofbroadcategories (e.g., high, medium, low), mainlyaccordingtotheinterpretation of the authors of each study but (whenever possible)with support from actual data provided in tables and figur