At the Segment-Prosody Divide The Interplay of Intonation, Sibilant Pitch

20.08.2011 Oliver Niebuhr 1

Spoken Language Analysis Dept. of General & Comparative Linguistics Christian-Albrechts-Universität zu Kiel

At the Segment-Prosody Divide

The Interplay of Intonation, Sibilant Pitch and Sibilant Assimilation

Oliver Niebuhr, Cassandra Lill & Jessica Neuschulz

17th International Congress of Phonetic Science, Hong Kong, China

Oral Session on Sibilant Sounds, August 20th, 11.10-11.30, Room S223



• Presented study is part of a line of research on • “Intonation segments” and “segmental intonations”

• The pitch curves of utterances are not only created by F0. By changing their sound qualities, sound segments can also create different spectral pitches (sibilant pitch, intrinsic pitch of vowels, …)

Are the pitch impressions caused by sound segments adjusted to the intonation context?• What kinds of sounds are concerned?• Which intonation contexts trigger spectral pitch adjustments?• Why does the adjustment occur? (E.g., in order to fill voiceless gaps so that the utterance tune is perceived “subjectively continuous” ? cf. Jones 1909:275)

Widely neglected questions, even though it was already noted by Daniel Jones (1950) that different “voice pitch contexts” create allophonic variation

• Introduction



• So far, studies on German focussed on utterance-final sound segments in L% and H% intonation contexts (Niebuhr 2008, Niebuhr 2009)

• Significant findings:• Different diphthong dynamics in closing diphthongs of German: Shorter onset, longer transition in H% than in L% contexts. //

• // more open, fronted and diphthongized in H% than in L% contexts• The same is true for vocalized <-er> (=[]) endings

• Fricatives have more high-frequency energy after H% than after L% “Fisch” //, fish “Buch” /x/, book

Queston here:• Does this also happen utterance-medially? • Tested with sibilant sequences

• Introduction



• Acoustic analyses based on the ‘KIESEL‘ corpus• Kieler Sammlung Expressiver Lesesprache, Kiel Collection of Expressive Read Speech• www.speechandemotion.de/Ressourcen.htm/

• 2x2 sentence mode and emphasis conditions

• Method

Neutral Statement

SN

Emphatic Statement

SE

Neutral QuestionQN

Emphatic QuestionQE

The same 12 sentences with simple SVO structure O= target word pairs of

Function word + Noun (in singular; with nuclear pitch accent)

“als Sänger” [], as a singer

“aus Schweden” , from Sweden

“bis Sachsen” [], to Saxony

“als Spender” , as a donor

/s/ assimilation condition

/sz/ non-assimilation condition

Created 6x2 different sibilant sequence conditions across word boundaries. E.g.,



• Sentences of 8 female speakers were analyzed• 12 x 8 x 4 = 384 sentences; 48 /s/ and 48 /sz/ tokens in each of the conditions SN, SE, QN, QE

• Crucial point:• the /s/ and /sz/ sequences occurred in very different pitch contexts• high pitch (H*) in statements low pitch (L*) in questions• Under emphasis: • pitch level increases further SN SE• pitch level decreases further QN QE

• Method

question



• Sentences of 8 female speakers were analyzed• 12 x 8 x 4 = 384 sentences; 48 /s/ and 48 /sz/ tokens in each of the conditions SN, SE, QN, QE

• Crucial questions:• (1) Is sibilant pitch adjusted to these different intonation/pitch contexts?• If so, the sibilant pitches of /s/ and /sz/ will decrease in the following order SE > SN > QN >QE • (2) In the cases of /s/: Is regressive /s/-to-[w] (i.e. light-to-dark noise) assimilation involved in this sibilant-pitch adjustment?

• Method

question



• Measurements• Spectral centre-of-gravity (CoG) values calculated in 10 ms intervals across each sibilant sequence.• Based on these values: 1 mean CoG and 1 CoG range (max-min CoG) for each sibilant sequence• Mean CoG = acoustic measure, but closely related to the perceived sibilant pitch impression• CoG range was to estimate the variation of sibilant pitch in each sequence

• Durations of the sequence sequences

• F0 values of the H* peaks and L* valleys

• Statistical tests: Univariate ANOVAs with 3 fixed factors…• (1) question vs statement, (2) /s/ vs /sz/, (3) neutral vs emphatic

• …For mean CoG, duration, and F0 measurements

• Method



• Intonation, Pitch accent• H* F0 peaks higher than L* F0 valleys (p < 0.001)• Emphasis increased H* peaks and lowered L* valleys (p < 0.001)• No significant effect of type of sibilant sequence on F0, but…

• Mean CoGs• clearly higher in H* than in L* contexts (p < 0.001)• higher for entirely alveolar /sz/ than for /s/ sequences (p < 0.001)• significant interactions show…• emphasis increases mean CoGs in statements, but decreases mean CoGs in questions: SE>SN>QN>QE• pitch context effects stronger for /s/ than for /sz/ sequences

• Results

H*

H*

L*

L*



• Mean CoGs• clearly higher in H* than in L* contexts (p < 0.001)• higher for entirely alveolar /sz/ than for /s/ sequences (p < 0.001)• significant interactions show…• emphasis increases mean CoGs in statements, but decreases mean CoGs in questions: SE>SN>QN>QE• pitch context effects stronger for /s/ than for /sz/ sequences

• Results

H*

L*

H*

L*

/s/ assimilation condition SE QEvsSN QNvs

/sz/ non-assimilation conditionSN QNvs



• Sibilant-sequence durations• Sibilant sequences became longer under emphasis (p < 0.001)• Sibilant sequences were shorter in questions than in statements (p < 0.001)• Most importantly, the /s/ sequences were not shorter than the /sz/ sequences.

• CoG ranges• became successively smaller across SE, SN, QN, QE (²=3.8; p < 0.05) /s/ sequences with low mean CoGs became spectrally as stable/homogeneous as /sz/ …• …and remain equally long • must be due to greater /s/-to-[] assimilation of /s/ (/s/ elision < duration)

• Results

H*

L*

H*

L*



• The spectral characteristics of the /s/ and /sz/ sequences varied systematically and in parallel with the F0 contexts provided by the H* and L* pitch accents.

• If the mean CoGs are taken as sibilant-pitch estimations, we may conclude that the pitch impressions caused by the sibilant sequences are adjusted to the F0 (i.e. intonation) context.

• The adjustment of sibilant to intonation pitches was stronger for /s/ than for /sz/.• German // shows lip rounding can be derounded together with shape and place of articulation = inherently greater potential to vary sibilant pitch• Compared with /sz/, the durational and spectral measurements indicate that regressive /s/-to-[] assimilation was used as an additional instrument to vary the sibilant pitch created by /s/ sequences.

• Conclusions



• Spectral pitch of speech sounds are not only adjusted to the intonation in the context of different utterance-final boundary tones (L% vs H%), but also utterance-medially in the context of different H* and L* pitch accents.

• Altogether, this means for research on intonation and sound segments

• Conclusions



Thank you for your attention

At the Segment-Prosody Divide The Interplay of Intonation, Sibilant Pitch

Documents