Proceedings Fonetik 2009 - Institutionen för lingvistik

Proceedings, FONETIK 2009, Dept. of Linguistics, Stockholm Universityinput speech sample with a rule-based outputpitch contour based on 1) the pitch range of theinput speech sample, used for speaker normalisation,2) the annotation, used to identify thetime and pitch gestures to be generated, and 3)the rule file, containing the values of the criticalpitch points. The Praat graphical user interfaceprovides immediate audio-visual feedback ofhow well the rules work, and also allows foreasy additional manipulation of pitch pointswith the Praat built-in Manipulation feature.Modelling compounds with SWINGSWING is now being used in our work with testingand developing the Swedish prosody modelfor compound words. Testing is done by selectingan input sound sample and a rule file of thesame intonational variety. If the model worksadequately, there should be a close match betweenthe F 0 contour of the original version andthe rule-based one generated by the tool. Figure7 shows Praat Manipulation objects for thethree dialect regions Gotland, Dala-Bergslagenand Upper Dalarna Swedish modelled with thecorresponding rules for each dialect region.Figure 7. Simulation of compound words withSWING. Praat Manipulation displays of mobiltelefonenof the three dialect regions Gotland, Dala-Bergslagen and Upper Dalarna Swedish (simulation:circles connected by solid line; original pitch: lightgreyline).The light grey lines show the original pitch ofeach dialect region, while the circles connectedwith the solid lines represent the rule-generatedoutput pitch contours.As can bee seen in Figure 7, the simulated(rule-based) pitch patterns clearly resemble thecorresponding three typical compound wordintonation patterns shown in Figure 4. There isalso a close match between the original pitch ofthe input speech samples and the simulatedpitch contour in all three dialectal regions.Discussion and additional remarksThe present paper partly confirms earlier observationsabout pitch patterns of word accentuationin the regional varieties of Dala andGotland Swedish, and partly adds new specificpieces of information, potentially extending ourknowledge about compound word intonation ofthese varieties.One point of discussion is the internal variationwithin Dala Swedish with a differentiationof the pitch patterns of word accents into UpperDalarna and Dala-Bergslagen intonational subvarieties.This division has earlier been suggestedby Engstrand and Nyström (2002) revisitingMeyer’s pitch curves of the two word accentsin simplex words, with speakers representingUpper Dalarna having a slightly earliertiming of the relevant pitch peak than speakersfrom Dala-Bergslagen. See also Olander’sstudy (2001) of Orsa Swedish as a case in pointconcerning word intonation in a variety of UpperDalarna.Our study of compound word intonationclearly demonstrates the characteristic risingfallingpitch pattern in connection with the primarystress of a compound word in Upper Dalarnaas opposed to the high pitch plateau betweenthe primary and secondary stresses of thecompound in Dala-Bergslagen, even if there isalso some variability among the differentspeakers investigated here. We would even liketo suggest that compound word intonation inDala-Bergslagen and Upper Dalarna Swedish ispotentially distinct. It would also appear to betrue that a compound in Upper Dalarna has gotthe same basic pitch shape as that of SouthSwedish. Generally, word intonation in UpperDalarna Swedish and South Swedish wouldeven seem to be basically the same.Another point of discussion is the suggestedsimilarity of word intonation between Dala andGotland Swedish. Even if the same basic pitchpattern of an accent II compound – the pitchplateau – can be found for speakers representingvarieties of both Dala-Bergslagen and Gotland,there is also an interesting difference to bediscerned and further examined. As has beenshown above, Gotland Swedish speakers tendto display more of an upstepped pitch shape for52