uchicago_beijing_creak_kmyu_201906.pdf

20a8ff44959a902d76386e2a75592154?s=47 krisyu
June 14, 2019

 uchicago_beijing_creak_kmyu_201906.pdf

Language Diversity, Contact, and Change. June 14-16, 2019. University of Chicago Center in Beijing. Conference website: https://lucian.uchicago.edu/blogs/lpl/workshop/

Abstract: The phonetic realization of tonal contrasts in many languages involves not just fundamental frequency/pitch, but also phonation. For example, it is well known that Tone 3 in Beijing Mandarin (e.g., the word for horse) is often creaky, and that the falling ”g” tone in Hmong is known to be breathy. In this talk, I illustrate methods for studying the phonetics of phonation contrasts, with an emphasis on creak—from data collection to data analysis. I present empirical data on tonal contrasts involving creak in Cantonese, Hmong, and Beijing and Taiwan Mandarin. All of these languages and language varieties include tones that are both low in fundamental frequency/pitch and often creaky. But it is a mistake to think that this implies that there is a uniform phonetic characterization of the creaky tones across these languages and speakers of these languages. In fact, it is unclear that there is any single phonetic dimension that can effectively characterize the presence of creak in speech. Rather, there are phonetically distinct kinds of creak, although our understanding of these kinds of creak is still limited. Thus, in addition to discussing the phonetic characterization of the contrast between creaky and non-creaky phonation, I also discuss progress towards phonetic characterizations of distinct kinds of creak and what conditions variation in creak across speakers and languages in Cantonese, Hmong, and Beijing and Taiwan Mandarin.

20a8ff44959a902d76386e2a75592154?s=128

krisyu

June 14, 2019
Tweet

Transcript

  1. VARIATION IN THE PHONETICS OF CREAKY VOICE IN CANTONESE, HMONG,

    AND MANDARIN KRISTINE YU (余世宁) ASSOCIATE PROFESSOR, DEPARTMENT OF LINGUISTICS, U. MASSACHUSETTS AMHERST LANGUAGE DIVERSITY, CONTACT, AND CHANGE UCHICAGO BEIJING CENTER JUNE 14, 2019 1
  2. UCHICAGO BEIJING: CREAK MANDARIN TONES ON 5-POINT PITCH SCALE 2

    5 4 3 2 1 Pitch scale 妈 ma (55) ⿇麻 má (35) 骂 mà (51) ⻢马 mǎ (213) High Low
  3. UCHICAGO BEIJING: CREAK MANDARIN TONES: F0 CONTOURS 3 Fundamental frequency

    (f0) is the primary acoustic correlate of pitch Xu (1997) 妈 (55) ⿇麻 (35) ⻢马 (213) 骂 (51) ⻢马 mǎ (213) ⿇麻 má (35) 妈 ma (55) 骂 mà (51)
  4. 4 http://youtu.be/P2pLJfWUjc8?t=3m53s Fundamental frequency is the rate of vocal fold

    vibration
  5. UCHICAGO BEIJING: CREAK VOCAL FOLD MOVEMENT DURING A SINGLE PHONATION

    CYCLE 5 Kreiman & Sidtis (2011): Fig 2.12, p. 32 P h o n a t i o n i s t h e process by which vocal folds modulate o u t g o i n g a i r t o produce sound. Different vocal fold c o n fi g u r a t i o n s i n phonation result in changes in perceived voice quality. Garellek (2019) Front view https://www.researchgate.net/ figure/a-Frontal-view-of-the- human-larynx-b-Picture-of-the- larynx-during-phonation-c- during_fig1_5326043
  6. UCHICAGO BEIJING: CREAK MANDARIN TONES: CREAKY PHONATION (嘎裂化) 6 Xu

    (1997) ⻢马 21(3) ⻢马 mǎ, 21(3) especially associated with creak in phonation. Tone with lowest f0 ⻢马 mǎ (213) ⿇麻 má (25) 妈 ma (55) 骂 mà (51) 嘎裂化 (CREAK)
  7. UCHICAGO BEIJING: CREAK 7 https://chinesepod.com/blog/2016/03/24/speak-mandarin-third-tone-like-a-kardashian/

  8. UCHICAGO BEIJING: CREAK 8 Mandarin: Kuang (2017)+refs within; Cantonese: Yu

    & Lam (2014) + refs within; Hmong: Esposito (2012), Garellek et al. (2013) + refs within; creaky falling/dipping tone typology, “Contour-Unspecified Low Tone”: Zhu (2015, §2), Zhu et al. (2012). Mandarin Cantonese White Hmong Level 55 55, 33, 22 55, 33, 22 Rise 35 25, 23 24 High fall 51 42, 52 (breathy contrast) Low fall 21 (T3) 21 (T4) 21 (m) CREAK IN LOW FALLING TONES ACROSS LANGUAGES creaky Lowest falling tone often produced with creak…
  9. UCHICAGO BEIJING: CREAK DISTINCT TYPES OF CREAK: CANTONESE 21 9

    Time (s) 0.8 0.9 -0.4798 0.3565 0 Narrow-band spectrograms Waveforms Time (s) 1.09 1.19 -0.1679 0.2286 0 Time (s) 0.7288 1.043 0 1000 Frequency (Hz) 0.728834954 Time (s) 1.055 1.276 0 1000 Frequency (Hz) 1.05537273 1.27621711 Some type Another type …and in fact, with variable types of creak… jiu33 lau lau jak̚3
  10. UCHICAGO BEIJING: CREAK 10 …BUT UNIFIED TRANSCRIPTION OF CREAK IN

    IPA * but see VoQS voice quality symbols extensions to IPA: https://doi.org/10.1017/S0025100317000159, and appendix Zhu (2015)
  11. UCHICAGO BEIJING: CREAK TALK OVERVIEW: LANGUAGE DIVERSITY 11 ▸ High-level,

    introductory overview of diversity in creaky voice within and across languages ▸ Examples from analysis of low falling tones in Cantonese, White Hmong, Beijing Mandarin, (Taiwan Mandarin) ▸ Idea: recognizing types of creaky voice can help us understand how variation in creaky voice is related to linguistic structure
  12. UCHICAGO BEIJING: CREAK PHONATION STUDIES IN LANGUAGES OF CHINA 12

    Kong & Wang (2013) …and many other more recent studies, e.g., see Journal of Chinese Linguistics (2015), Vol. 43 (1B) for collection of articulatory/physiological studies
  13. UCHICAGO BEIJING: CREAK TALK OVERVIEW: QUANTITATIVE METHODS 13 ▸Quantitative methods

    for studying under-studied languages ▸Intro to quantitative methods for acoustic study of phonation (should be possible to apply to already- existing recordings) ▸Methodological points about: (1) challenges of working with highly collinear, high-dimensional data, and (2) the importance of analyzing time-course of measured parameters in addition to parameter means ▸ Exploratory cluster analysis to see what dimensions underlie clusters are identified in “creak” space
  14. CREAKY VOICE AND LINGUISTIC STRUCTURE: 14 ▸ Phonemic contrast (Mpi)

    ▸ Contrast enhancement (low falling tones) ▸ Prosodic phrasing (phrase-final creak) ▸ Use of creaky in tonal perception
  15. UCHICAGO BEIJING: CREAK 15 CREAKY VOICE PHONEMIC CONTRAST EXAMPLE http://phonetics.ucla.edu/vowels/chapter12/mpi.html

    Creaky voice contrast across all pitch contour types
  16. UCHICAGO BEIJING: CREAK 16 Mandarin: Kuang (2017)+refs within; Cantonese: Yu

    & Lam (2014) + refs within; Hmong: Esposito (2012), Garellek et al. (2013) + refs within; creaky falling/dipping tone typology, “Contour-Unspecified Low Tone”: Zhu (2015, §2), Zhu et al. (2012). Mandarin Cantonese White Hmong Level 55 55, 33, 22 55, 33, 22 Rise 35 25, 23 24 High fall 51 42, 52 (breathy contrast) Low fall 21 21 21 CREAK IN LOW FALLING TONES ACROSS LANGUAGES creaky Lowest falling tone often produced with creak…
  17. UCHICAGO BEIJING: CREAK 17 CREAK IN ALL MANDARIN TONES, NOT

    JUST 21 …and likelihood of creak increases with prosodic boundary strength… Kuang (2018), Fig. 2a
  18. UCHICAGO BEIJING: CREAK 18 LISTENERS ARE SENSITIVE TO CREAK TYPES…

    Listeners can discriminate between /t/-glottalization and phrase-final creak in American English (little confusability in button vs. bun, whether phrase-final or not) Garellek (2015), Fig. 1 /button/, non-phrase final /button/, phrase final
  19. UCHICAGO BEIJING: CREAK 19 PITCH-DEPENDENT VOICE QUALITY CONTINUUM Adapted from

    Kuang (2017): Figure 4 Falsetto Tense Modal Vocal fry Pitch scale Creak in production of all Mandarin tones, not just 21. Non-modal phonation can be driven by extreme pitch targets. At low end of pitch scale: vocal fry (or breathy) phonation. (Kuang 2017) (or breathy) High Low
  20. UCHICAGO BEIJING: CREAK 20 Zhu (2012) “MULTI-REGISTER AND FOUR-LEVEL MODEL”

    6–3 5-2 4-1 Pitch scale
  21. UCHICAGO BEIJING: CREAK 21 CREAKY VOICE QUALITY ENHANCES LOW PITCH

    PERCEPT Zhu, Zhang & Yi (2012) 5 4 3 2 1 Pitch scale High Low Idea: Creaky voice quality enhances low pitch, i.e., percept of extra-low “0” on pitch scale ➠ …and see Kuang & Liberman (2016), The effect of vocal fry on pitch perception. ICASSP-2016. languagelog.ldc.upenn.edu/ myl/KuangICASSP2016.pdf
  22. UCHICAGO BEIJING: CREAK 22 IS CREAK USED IN PERCEPTION OF

    21 TONES? ▸ Mandarin: ▸ Yes, (Yang 2015), bias for 21 identification in stimuli with f0 contours resynthesized with PSOLA from naturally produced tones in isolated syllables, also Belotel-Grenié and Grenié (1997) ▸ No (Gårding et al. 1986; Cao 2012, “creak” manually added abrupt f0 drop, “jitter” in f0 contour) ▸ Yes, but only for “extra-low” manipulation and not other kinds of creak properties: Huang (2019, 2019a) with Praat KlattGrid synthesized stimuli ▸ Cantonese: ▸ Yes, creaky 21 tones identified more accurately than non-creaky 21 tones; creak biases tone identification between 21 and 22 towards 21 (Yu and Lam 2014, naturally produced stimuli) ▸ No (Zhang and Kirby in prep, Praat KlattGrid synthesized stimuli, period doubling parameter manipulation) ▸ Hmong: ▸ No, creak doesn’t bias tone identification between 21 and 22 towards 21, f0 contours resynthesized with PSOLA from naturally produced tones in isolated syllable (Garellek et al. 2013) In sum…mixed results
  23. UCHICAGO BEIJING: CREAK 23 WHY INCONSISTENT RESULTS? ▸ Major source

    of variation between experimental tasks: difference in design of experimental stimuli ▸ Naturally produced stimuli (from recordings) ▸ Resynthesized stimuli (Praat PSOLA resynthesis of f0 contour, manual editing of f0 contour) ▸ Synthesized stimuli from scratch (e.g., Praat KlattGrid synthesizer, different parameters manipulated in synthesis)? To begin to understand why: we need to better understand natural variation in production (and perception) of “creak”
  24. TYPES OF CREAK 24

  25. UCHICAGO BEIJING: CREAK MANY DIFFERENT TERMS FOR “CREAK”! 25

  26. UCHICAGO BEIJING: CREAK PROPOSED BREAKDOWN OF CREAK TYPES 26 Keating,

    Garellek and Kreiman (2015)
  27. UCHICAGO BEIJING: CREAK THREE ATOMIC PROPERTIES OF CREAK 27 Garellek

    (2019), simplification of Keating et al. (2015) Idea: presence of just one of these properties individually already sufficient to generate perception of creaky voice quality. ▸ Percept: Constricted voicing ▸ Acoustic correlate: spectral tilt ▸ Percept: Irregular pitch ▸ Acoustic correlate: noise ▸ Percept: Low pitch ▸ Acoustic correlate: fundamental frequency (f0)
  28. UCHICAGO BEIJING: CREAK LOW PITCH ALONE INDUCES CREAK PERCEPT 28

    Time (s) 0.1064 0.2067 -0.9012 0.9663 0 0.144962364 Time (s) 0.1064 0.2067 -0.8999 0.9681 0 0.144962364 f0 = 100 Hz f0 = 60 Hz Synthesized with Praat KlattGrid, all parameters other than f0 held constant
  29. UCHICAGO BEIJING: CREAK HYPOTHESIS: DISTINCT TYPES OF CREAK? 29 https://pages.ucsd.edu/~mgarellek/files/Garellek_Phonetics_of_Voice_Handbook_final.pdf

    Garellek (2019: Fig. 3), Keating et al. (2015) Mpi: tense voice (not f0-dependent) “21” tones: low pitch?
  30. UCHICAGO BEIJING: CREAK DISTINCT TYPES OF CREAK: CANTONESE 21 30

    Time (s) 0.8 0.9 -0.4798 0.3565 0 Narrow-band spectrograms Waveforms Time (s) 1.09 1.19 -0.1679 0.2286 0 Time (s) 0.7288 1.043 0 1000 Frequency (Hz) 0.728834954 Time (s) 1.055 1.276 0 1000 Frequency (Hz) 1.05537273 1.27621711 Prototypical Multiply pulsed jiu33 lau lau jak̚3 sub-harmonics
  31. UCHICAGO BEIJING: CREAK DOUBLY PULSED (PERIOD DOUBLED) CREAK: WAVEFORM 31

    A period doubled waveform. Arrows mark the pattern of alternating cycles. Two periods: doubled period results in sub-harmonics Acoustic correlate: high subharmonics-to- harmonics ratio (SHR), a special kind of noise
  32. UCHICAGO BEIJING: CREAK HYPOTHESIS: DISTINCT TYPES OF CREAK? 32 https://pages.ucsd.edu/~mgarellek/files/Garellek_Phonetics_of_Voice_Handbook_final.pdf

    Idea: Creak types can be understood in terms of combinations of the three properties. Garellek (2019) Example: Vocal fry ☑Constricted voicing ☑low pitch ☒NO* irregular pitch (*simplification: see Keating et al. (2015 §§3.1,4.2,4.4)
  33. UCHICAGO BEIJING: CREAK ACOUSTIC MEASURES FOR CREAK PROPERTIES 33 Garellek

    (2019), Keating et al. (2015) ▸ Constricted voicing (spectral tilt) ▸ Lower H1-H2 㱺 lower spectral tilt 㱺 More constricted ▸ Irregular pitch (noise) ▸ Lower harmonics-to-noise-ratio (HNR) 㱺 noisier 㱺 more irregularity in pitch ▸ Low pitch (f0) ▸ Lower f0 㱺 lower pitch Example: Vocal fry ☑Constricted voicing: lower H1-H2 ☒NO irregular pitch: higher HNR ☑low pitch: lower f0
  34. UCHICAGO BEIJING: CREAK 34 VIDEOKYMOGRAPHIC IMAGES OF PHONATION https://www.researchgate.net/figure/Videokymographic-images-of-the-vibration-of-the-vocal-folds-during-phonation-in-the-three_fig3_241682145 Vocal

    cords are mostly constricted during vocal fry. Spectral tilt (H1-H2) is acoustic correlate of proportion of closure during phonation cycle. (review: Kreiman et al. 2014, §4)
  35. LINGUIST 414, SPRING 2019. CLASS 4.1 SPECTRAL TILT: H1-H2 IN

    THE SPECTRUM 35 1st harmonic: frequency f0, amplitude H1 2nd harmonic: frequency 2*f0, amplitude H2 H1 H2
  36. LINGUIST 414, SPRING 2019. CLASS 4.1 LOWER H1-H2: MORE CONSTRICTED

    VOICING 36 H1 H2 Spectrum of modal [a] Spectrum of creaky [a] Adapted from Kong (1996, 42) on Hani language H1 H2 Higher H1-H2 Lower H1-H2
  37. UCHICAGO BEIJING: CREAK HARMONICS TO NOISE RATIO (HNR) IN SPECTRUM

    37 Low noise: high HNR High noise: low HNR https://slideplayer.com/slide/13152153/79/images/13/Harmonics-to-noise+Ratio.jpg Cepstral peak prominence (CPP): related measure with same correlations: low noise, high CPP
  38. ACOUSTIC ANALYSIS: “21” ACROSS LANGUAGES AND EXPLORING CREAK TYPES 38

  39. UCHICAGO BEIJING: CREAK DATA OVERVIEW 39 ▸ Tones recorded sentence-medially

    in carrier phrase, ranging over all possible bitone combinations, e.g., ma ma, ma má, etc. ▸ In connected speech, not in isolation ▸ 12 speakers per language, 8 currently analyzed for Beijing Mandarin and 3 for Taiwan Mandarin ▸ Cantonese and Taiwan Mandarin recorded in Los Angeles, California; Hmong recorded in Fresno, California; Beijing Mandarin recorded in Beijing ▸ (Warning: of course results about across-language comparisons are limited by small size of sample of participants)
  40. UCHICAGO BEIJING: CREAK RECALL - TALK OVERVIEW: QUANTITATIVE METHODS 40

    ▸Quantitative methods for studying under-studied languages ▸Intro to quantitative methods for acoustic study of phonation (should be possible to apply to already- existing recordings) ▸Methodological points about: (1) challenges of working with collinearity in high-dimensional data, and (2) the importance of analyzing time-course of measured parameters in addition to parameter means ▸ Exploratory cluster analysis to see what clusters are identified in “creak” space
  41. UCHICAGO BEIJING: CREAK 41 SPECTRAL MEASURES OF 21 VS. {22,

    51} Mandarin Cantonese White Hmong Level 55 55, 33, 22 55, 33, 22 Rise 35 25, 23 24 High fall 51 42, 52 (breathy contrast) Low fall 21 21 21
  42. UCHICAGO BEIJING: CREAK AUTOMATING MEASUREMENT OF SPECTRAL PROPERTIES 42 http://www.phonetics.ucla.edu/voicesauce/

  43. UCHICAGO BEIJING: CREAK VOICESAUCE REQUIRES MATLAB AND/OR WINDOWS 43 http://www.phonetics.ucla.edu/voicesauce/

  44. UCHICAGO BEIJING: CREAK PARAMETERS MEASURED BY VOICESAUCE 44 Huge number

    of parameters measured that are related (high-dimensionality, collinearity). And that’s not even taking into account that each parameter takes place over a time course. Dimensionality reduction helpful. (Faytak tutorial Sunday)
  45. UCHICAGO BEIJING: CREAK ANALYSIS OVERVIEW 45 ▸ What parameters best

    distinguish between 21 and {22,51}? ▸ Functional data analysis (FDA) with principal components analysis (PCA) to capture time course shape in low- dimensional space (see appendix) ▸ Random forests to compute which parameters best discriminate tone categories, while dealing with collinearity in large number of parameters (see appendix) ▸ How do “21” utterances cluster in “creak” space (HNR, H1-H2, f0)? ▸ Exploratory hierarchical cluster analyses to discover potential underlying clusters of creak types in data
  46. UCHICAGO BEIJING: CREAK PRINCIPAL COMPONENTS OF F0 CURVES 46 Functional

    principle components decomposes each f0 contour into a weighted sum of atomic curve shape components.
  47. UCHICAGO BEIJING: CREAK F0 PRINCIPAL COMPONENTS 1 AND 2 47

    Principal component 1 mostly shifts curve up/down Principal component 2 tilts the curve like a see-saw http://www.playground-china.com/8165ISeesaws2543.html https://i.redd.it/8yc3gk1g2ev01.jpg
  48. UCHICAGO BEIJING: CREAK 51 5151 51 51 51 51 51

    51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2121 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 5151 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 5151 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 5151 5151 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 -100 -50 0 50 100 -200 0 200 400 Principle component 1 Principle component 2 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 2121 21 21 21 21 21 21 2121 21 21 21 2121 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 5151 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 5151 51 51 51 51 51 51 51 51 51 51 5151 51 21 21 21 21 21 2121 2121 21 21 2121 2121 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 5151 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 51 51 51 5151 51 51 51 51 51 51 51 51 51 51 51 51 51 51 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 21 51 51 51 51 51 51 51 51 51 5151 51 51 51 51 51 -50 0 50 100 150 200 -100 0 100 200 Principle component 1 Principle component 2 21 VS. 51 IN BEIJING VS. TAIWAN MANDARIN: F0 48 Taiwan Beijing Principal component 2 best discriminates 21/51 Principal component 1 best discriminates 21/51 You miss this distinction if you look only at mean f0.
  49. UCHICAGO BEIJING: CREAK 21 VS. 22 IN HMONG VS. CANTONESE:

    H1 49 21 21 21 21 21 2121 21 21 21 22 22 22 22 22 22 2222 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 2222 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 2222 22 22 21 21 21 21 21 22 22 2222 22 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 2222 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 2222 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 21 21 21 2121 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 2222 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 2222 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 2121 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 2121 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 2222 22 22 21 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 2122 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 2222 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 2222 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 2222 22 2222 2222 22 22 22 22 22 22 22 2222 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 2121 21 22 22 22 22 22 22 22 2222 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 2222 22 2222 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 2222 22 2222 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 21 21 21 21 21 22 2222 22 2222 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 2121 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 2222 21 21 21 21 21 22 2222 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 22 21 22 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 2222 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 2222 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 2222 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 2222 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 2222 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 2222 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 2222 22 22 22 21 21 21 21 21 2222 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 2222 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 2222 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 2222 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 2222 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 21 21 21 21 21 2222 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 2121 21 22 22 22 2222 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 212222 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 2221 22 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 2222 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 2222 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 2222 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 2121 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 2122 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 2222 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 2222 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2122 21 22 2122 21 22 21 22 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 2222 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 2222 22 22 22 22 22 22 22 22 22 22 22 22 22 2121 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 21 21 21 21 21 22 22 22 22 22 22 2222 22 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 2222 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 2222 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 -20 -10 0 10 20 -60 -40 -20 0 20 Principle component 1 Principle component 2 21 21 21 21 21 22 2222 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 2122 21 22 22 22 22 22 22 22 22 22 2222 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 2222 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 22 22 2222 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 2222 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 22 22 22 22 2222 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 22 22 22 2222 22 22 2222 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 2222 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 2222 22 22 22 22 2222 22 22 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 22 22 2222 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 22 22 22 22 2222 22 22 22 22 22 21 22 21 22 21 22 21 22 21 2222 22 22 22 22 22 2222 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 2222 2222 21 21 21 21 21 21 21 2121 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 2122 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 2222 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 22 22 22 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 22 22 21 22 21 22 21 22 21 22 21 22 22 22 22 22 22 22 22 22 22 -10 0 10 20 -50 -25 0 25 Principle component 1 Principle component 2 White Hmong Cantonese Principal component 1 best discriminates 21/22 Principal component 2 best discriminates 21/22 You miss this distinction if you look only at mean H1.
  50. UCHICAGO BEIJING: CREAK CLUSTER ANALYSIS OVERVIEW 50 ▸ How do

    the “creak space” measures determine dimensions underlying discovered clusters? ▸ Principal components analysis reveal independent dimensions: are the three types of creak space measures independent? ▸ Noise: Harmonics to noise ratio (HNR), Cepstral peak prominence (CPP), Subharmonics-to-harmonics ratio (SHR) for multiply pulsed/doubly pulsed creak ▸ Fundamental frequency (f0) ▸ Spectral tilt (H1-H2) ▸ How do “21” utterances cluster in creak space (HNR, H1-H2, f0)?
  51. UCHICAGO BEIJING: CREAK 3D CREAK SPACE: (HNR, H1-H2, F0) 51

    High noise: low harmonics-to-noise ratio (HNR) Low f0 Low H1-H2 347 462 337 340 350 472 2 16 162 330 373 344 372 249 374 334 460 365 335 370 336 341 371 345 346 343 322 352 321 386 435 377 453 376 392 443 461 368 393 375 383 348 353 361 349 390 356 358 469 359 466 339 387 354 465 326 431 328 409 362 367 384 467 473 366 363 369 364 405 433 360 437 327 455 157 160 79 81 46 6 30 158 9 155 10 76 82 62 110 40 108 48 106 58 114 69 86 89 90 27 28 146 143 149 132 51 54 7 20 151 124 44 72 338 439 457 159 1 153 83 60 67 8 29 39 41 45 56 49 42 43 138 80 66 156 36 38 37 152 136 50 57 59 75 144 140 122 134 17 19 47 468 21 147 26 104 102 141 223 238 300 306 196 226 203 243 298 186 292 282 308 210 307 211 244 241 271 279 188 190 217 245 272 278 305 274 187 242 261 207 208 201 216 220 284 281 288 290 310 246 309 294 296 311 318 224 228 199 164 198 247 230 303 178 166 234 213 218 248 286 277 280 221 185 227 222 179 302 319 259 255 182 317 169 165 233 173 192 239 287 168 167 194 267 289 265 269 275 263 195 283 314 174 312 299 240 297 237 231 291 295 253 252 254 232 236 163 316 170 175 206 214 215 209 293 205 262 202 191 229 212 189 264 193 204 270 285 250 273 219 315 276 266 268 258 320 397 171 161 172 260 177 251 180 197 304 176 183 313 444 430 452 428 424 458 420 456 388 391 385 395 422 432 438 378 426 448 440 446 396 181 464 470 380 415 417 450 454 324 416 474 400 429 381 411 410 419 394 418 351 389 331 200 301 398 441 445 412 471 333 355 184 256 257 225 235 342 329 413 382 323 332 408 401 406 414 357 449 402 407 451 459 399 463 436 421 379 442 425 434 447 403 427 404 423 65 118 130 23 116 55 125 120 52 53 105 133 73 61 63 126 128 97 70 87 94 64 68 33 148 150 84 88 92 22 96 34 31 35 5 78 91 145 93 85 95 109 129 111 103 139 101 127 24 25 113 121 123 137 107 131 135 112 77 154 98 15 100 99 32 142 18 71 12 14 3 4 74 13 11 325 119 115 117 0 10 20 30 40 50 Cluster Dendrogram hclust (*, "ward.D2") d Height
  52. UCHICAGO BEIJING: CREAK WHICH MEASURES DETERMINE PCA DIMENSIONS? 52 Hmong

    0 0.1 0.2 0.29 0.39 0.49 0.59 0.69 0.78 0.88 0.98 Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Beijing Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Taiwan Dim.1 Dim.2 Dim.3 Dim.4 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Cantonese The darker the circle, the more that measure dominates the PCA dimension
  53. UCHICAGO BEIJING: CREAK WHICH MEASURES DETERMINE PCA DIMENSIONS? 53 Hmong

    0 0.1 0.2 0.29 0.39 0.49 0.59 0.69 0.78 0.88 0.98 Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Beijing Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Taiwan Dim.1 Dim.2 Dim.3 Dim.4 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Cantonese Noise measures make up one PCA dimension.
  54. UCHICAGO BEIJING: CREAK WHICH MEASURES DETERMINE PCA DIMENSIONS? 54 Hmong

    0 0.1 0.2 0.29 0.39 0.49 0.59 0.69 0.78 0.88 0.98 Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Beijing Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Taiwan Dim.1 Dim.2 Dim.3 Dim.4 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Cantonese SHR (subharmonics, correlate of doubly pulsed creak) mostly independent of other noise measures.
  55. UCHICAGO BEIJING: CREAK WHICH MEASURES DETERMINE PCA DIMENSIONS? 55 Hmong

    0 0.1 0.2 0.29 0.39 0.49 0.59 0.69 0.78 0.88 0.98 Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Beijing Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Taiwan Dim.1 Dim.2 Dim.3 Dim.4 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Cantonese F0 and H1-H2 often travel together in a dimension.
  56. UCHICAGO BEIJING: CREAK WHICH MEASURES DETERMINE PCA DIMENSIONS? 56 Hmong

    0 0.1 0.2 0.29 0.39 0.49 0.59 0.69 0.78 0.88 0.98 Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Beijing Dim.1 Dim.2 Dim.3 Dim.4 Dim.5 Dim.6 Dim.7 Dim.8 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Taiwan Dim.1 Dim.2 Dim.3 Dim.4 H1H2c_mean strF0_mean SHR_mean HNR05_mean HNR15_mean HNR25_mean HNR35_mean CPP_mean Cantonese “Creak dimensions” least independent in Beijing Mandarin and Cantonese.
  57. UCHICAGO BEIJING: CREAK HMONG 21 CLUSTERS IN 3D CREAK SPACE

    57 Low f0, H1-H2, HNR: Not irregular pitch? High f0, high HNR: prototypical/unconstricted High H1-H2, Lower f0, HNR: unconstricted? High f0, H1-H2, HNR: Modal
  58. UCHICAGO BEIJING: CREAK CANTONESE 21 CLUSTERS IN 3D CREAK SPACE

    58
  59. UCHICAGO BEIJING: CREAK 2D CREAK SPACE: CONSTRICTION BY F0 59

    Hmong Cantonese -4 -2 0 2 -2 0 2 4 strF0_mean H1H2c_mean cluster 1 2 3 4 Cluster plot -3 -2 -1 0 1 2 -2 0 2 strF0_mean H1H2c_mean cluster 1 2 3 4 Cluster plot
  60. UCHICAGO BEIJING: CREAK 2D CREAK SPACE: NOISE BY F0 60

    Hmong Cantonese -2 -1 0 1 2 -2 0 2 4 strF0_mean HNR15_mean cluster 1 2 3 4 Cluster plot -2 -1 0 1 2 -2 0 2 strF0_mean HNR15_mean cluster 1 2 3 4 Cluster plot
  61. UCHICAGO BEIJING: CREAK 2D CREAK SPACE: NOISE BY CONSTRICTION 61

    Hmong Cantonese -2 -1 0 1 2 -4 -2 0 2 H1H2c_mean HNR15_mean cluste 1 2 3 4 Cluster plot -2 -1 0 1 2 -3 -2 -1 0 1 2 H1H2c_mean HNR15_mean cluster 1 2 3 4 Cluster plot
  62. UCHICAGO BEIJING: CREAK CONCLUSION 62 ▸ Idea: recognizing types of

    creaky voice can help us understand how variation in creaky voice is related to linguistic structure: ▸ Vocal Fry ⊂ Creaky Voice ▸ 3D “creak space” of noise (harmonics-to-noise-ratios, etc.), fundamental frequency (f0), and constricted voice (H1-H2) does emerge in cluster discovery ▸ Methodological points ▸ Intro to quantitative methods for acoustic study of phonation: dimensionality reduction can help us understand important features of time- course of highly collinear high-dimensional voice quality measures for tone classification ▸ Further challenges: Is f0 even well-defined? Spectral measures largely based on f0! Non-visual inspection/automatic creak detection (see appendix)
  63. UCHICAGO BEIJING: CREAK REFERENCES 63 ▸ Cao, R. (2012). Perception

    of Mandarin Chinese Tone 2/Tone 3 and the role of creaky voice. Doctoral dissertation. University of Florida. ▸ Gårding, E., Kratochvil, P., Svantesson, J. O., and Zhang, J. (1986). Tone 4 and Tone 3 discrimination in modern standard Chinese. Language and Speech, 29:281- 293. ▸ Garellek, M. (2019). The phonetics of voice. In W. F. Katz & P. F. Assmann eds., The Routledge Handbook of Phonetics: 75-106. http:// idiom.ucsd.edu/~mgarellek/files/Garellek_Phonetics_of_Voice_Handbook_final.pdf ▸ Garellek, M., Keating, P., Esposito, C. M., Kreiman, J. 2013. Voice quality and tone identification in White Hmong. JASA 133, 1078-1089. ▸ Huang, Yaqian. (2019). Low f0 as a creak attribute in Mandarin tone perception. LSA 2019 https://drive.google.com/open? id=10iURFJ_fu9rPPYgcdRVYOHVjtA-ac1vZ ▸ Kreiman and Sidtis (2011). Foundations of Voice Studies. Wiley Blackwell. https://onlinelibrary.wiley.com/doi/book/ 10.1002/9781444395068 ▸ Gerratt, B. R., and Kreiman, J. (2001). “Toward a taxonomy of nonmodal phonation,” J. Phonetics 29, 365–381. ▸ Kong, Jiangping. (1996). Hani yu fa sheng lei xing sheng xue yanjiu ji yin zhi gai nian de tao lun [Study on phonation types and timbre of Hani language]. Min zu yu wen [Minority Languages of China] 1, 40-46. ▸ Kong, Jiangping and Gaowu Wang. (2013). Contemporary voice research: a China perspective. ▸ Kreiman, J., Gerratt, B. R., Garellek, M., Samlan, R., and Zhang, Z. (2014). Toward a unified theory of voice production and perception. Loquens, 1(1), e009. doi: http://dx.doi.org/10.3989/loquens.2014.009 ▸ Kuang, Jianjing. (2017) Covariation between voice quality and pitch: Revisiting the case of Mandarin creaky voice. JASA https://doi.org/ 10.1121/1.5003649 ▸ Kuang, Jianjing. (2018). The influence of tonal categories and prosodic boundaries on the creakiness in Mandarin. JASA 143, EL509 (2018); doi: 10.1121/1.5043094 ▸ Yang, Ruoxiao. (2015). The role of phonation cues in Mandarin tonal perception. J. Chin. Ling. 43(2): 453-472, https://www.jstor.org/stable/ 24774954 ▸ Yu, K. M., and Lam, H. W. (2014). The role of creaky voice in Cantonese tonal perception. JASA, 136(3), 1320-1333. 38 ▸ Zhang, Yubin and James Kirby (2018). Variation in Cantonese Tone 4 and Tone 6 perception. The 30th North American Conference on Chinese Linguistics (NACCL-30) ▸ Zhu, Xiaonong. (2012). Multiregisters and Four Levels: A New Tonal Model. https://www.jstor.org/stable/23754196 ▸ Zhu, Xianong. (2015). Types of falling tones. J. Chin. Ling. 43(2): 605-637. https://www.jstor.org/stable/24774975 ▸ Zhu Xiaonong 朱晓农,Zhang Ting 章婷 and Yi Li 衣莉. (2012). 凹调的种类 (A classification of dipping tones). Zhongguo Yuwen 中国语⽂文 5: 420-436.
  64. APPENDIX 64

  65. UCHICAGO BEIJING: CREAK IPA VOICE QUALITY SYMBOLS EXTENSION 65 https://doi.org/10.1017/S0025100317000159

  66. UCHICAGO BEIJING: CREAK FUNCTIONAL PRINCIPAL COMPONENTS RESOURCES 66 ▸ http://lands.let.ru.nl/FDA/

    (Michele Gubian) ▸ Gubian workshop materials: https://github.com/ uasolo/FPCA-phonetics-workshop ▸ http://www.psych.mcgill.ca/misc/fda/ (older, but can still be useful) ▸ faculty.bscb.cornell.edu/~hooker/ ShortCourseHandout.pdf, http://faculty.bscb.cornell.edu/ ~hooker/FDAWorkshop/ (Giles Hooker)
  67. UCHICAGO BEIJING: CREAK WORKING WITH COLLINEAR DATA 67 ▸ Strategies

    for addressing collinearity in multivariate linguistic data (Tomaschek, Peter Hendrix, R. Haraald Baayen, J. Phonetics, 2018) ▸ https://www.sciencedirect.com/science/article/pii/ S0095447017302292 ▸ Beware Default Random Forest Importances ▸ https://explained.ai/rf-importance/index.html
  68. UCHICAGO BEIJING: CREAK AUTOMATIC CREAK DETECTION 68 ▸ Drugman, T.,

    Kane, J., Gobl, C. (requires Matlab) ▸ https://github.com/covarep/covarep/tree/master/ glottalsource/creaky_voice_detection ▸ See also Kuang (2018)
  69. EXTRA SLIDES 69

  70. UCHICAGO BEIJING: CREAK 70 PITCH-DEPENDENT VOICE QUALITY CONTINUUM Zhu (2015)

  71. UCHICAGO BEIJING: CREAK EXAMPLE OF DAMPING IN WAVEFORM 71 https://www.toppr.com/guides/physics/oscillations/damped-simple-harmonic-motion/

  72. UCHICAGO BEIJING: CREAK 72 MULTI-REGISTER AND FOUR-LEVEL MODEL Zhu (2012)