The phonetics of voice

Full text

The phonetics of voice1
Marc Garellek, University of California San Diego
Chapter in The Routledge Handbook of Phonetics (W. Katz and P. Assmann, editors)
Revised 9th October 2017
1 Introduction
This chapter focuses on the phonetics of the voice. The term ‘voice’ is used to mean many different
things, with definitions varying both within and across researchers and disciplines. In terms of
voice articulation, definitions can vary from the very narrow – how the vocal folds vibrate – to the
very broad, where ‘voice’ is essentially synonymous with ‘speech’ – how the vocal folds and all
other vocal tract articulators influence how we sound (Kreiman & Sidtis, 2011). In this chapter, I
will use the term ‘voice’ to refer to sound produced by the vocal folds, including but not limited to
vocal fold vibration. I have chosen to focus only on a narrow conception of the voice in order to
constrain the discussion; as we will see, the phonetics of voice – even when it concerns only vocal
fold articulation – is remarkably complex and of great relevance to phonetic and linguistic research.
In contrast, I will use the term ‘voice quality’ to refer to the percept resulting from the voice: in
other words, different vocal fold configurations have specific perceptual ramifications, which we
will call changes in voice quality. The distinction between voice and voice quality adopted here is
therefore analogous to that made between ‘f0’ and ‘pitch’.
Why should we be interested in the phonetics of the voice? Linguists are interested in how
specific forms contribute to linguistic meaning; for spoken languages, phonetic and phonological
research addresses this goal from the point of view of how sounds contribute to meaning. Because
the vocal folds are part of the speech apparatus, a complete understanding of the sound-to-meaning
relationship requires knowledge of how sounds produced by the vocal folds contribute to linguistic
meaning. There are two main contributions of the vocal folds: first, their movements can be used
contrastively in languages. That is, in some languages changes in (a) the presence vs. absence of
vocal fold vibration, (b) the rate of vibration, and (c) the quality of vibration can signal a change
in lexical meaning: compare (a) English /"sl6pi/ ‘sloppy’ vs. /"sl6bi/ ‘slobby’; (b) White Hmong
1The author wishes to thank editors William Katz and Peter Assmann, as well as an anonymous reviewer, Adam
Chong, Pat Keating, Jody Kreiman, and Robin Samlan for their thoughtful comments and suggestions on earlier
versions of this chapter.
1

/tO Ă
£/ ‘pierced’ vs. /tO Ă
£/ ‘wait’; and (c) Jalapa Mazatec breathy-voiced [ndæ
¨
Ă
£] ‘horse’, creaky-
voiced [ndæ
˜
Ă
£] ‘buttock’, and modal-voiced [nthæĂ
£] ‘seed’ (Silverman et al., 1995; Garellek &
Keating, 2011).2
Second, because of their position downstream from the trachea and upstream from the re-
mainder of the vocal tract, all speech is first modulated by the vocal folds. This may seem counter-
intuitive at first, given that phoneticians tend to speak of two sources of sound excitation – voicing
(produced by the vocal folds) and noise (produced in the vocal tract in all obstruents but [P],[h],
and [H]). But even for sounds characterized by acoustic noise generated in the vocal tract (such
as [t] and [s]), the vocal folds still assume a particular articulatory target; for example, voice-
less obstruents (even when unaspirated) often show some degree vocal fold spreading (Munhall &
L¨
ofqvist, 1992), presumably to inhibit voicing and/or to facilitate the high airflow needed to gener-
ate turbulence for voiceless fricatives. The voice is thus part of the production of all speech sounds,
regardless of whether it is used to make phonological contrasts. Aside from phonological mean-
ing, the voice contributes to changes in prosody, syntax, discourse, as well as to speaker identity
and emotional state (Choi et al., 2005; Zhuang & Hasegawa-Johnson, 2008; Gobl & N´
ı Chasaide,
2010; Esling & Edmondson, 2010; Kreiman & Sidtis, 2011; Yanushevskaya et al., 2011; Podesva
& Callier, 2015; Park et al., 2016; Yanushevskaya et al., 2016). Changes in vocal fold vibration
are also associated with differences in singing registers and with voice disorders (Sundberg, 1987;
Titze, 1994; Kempster et al., 2011; Sapienza et al., 2011).
This chapter therefore takes as a starting point that the voice is ultimately used to convey in-
formation, which is transmitted between speakers and hearers in stages commonly known as the
‘speech chain’ (Denes & Pinson, 1993). Although there have been many advancements over the
years in understanding how the vocal folds are innervated, how they move and vibrate, and how
different voice settings are manifested acoustically (for instance, see overviews in Titze, 1994;
Baken & Orlikoff, 2000; Stevens, 2000; Gobl & N´
ı Chasaide, 2010; Hirose, 2010; Story, 2015),
it is still unclear how these stages of the speech chain interact with one another to influence voice
quality. Yet, ultimately the main goals of the study of the voice should be to answer two funda-
2The transcriptions in Hmong and Mazatec include “Chao letters” (Chao, 1930) for differences in lexical tone: the
vertical bar reflects the pitch range from lowest (level 1) to highest (level 5) in one’s normal speaking range, and the
horizontal lines perpendicular to the vertical bar represent pitch values along that range from the beginning to the end
of the tone.
2

mental questions: (1) When we perceive a change in the voice, what caused that change? and (2)
What are the acoustical and perceptual results of a change in voice production? In my research
with colleagues (e.g. Kreiman et al., 2014), we address these questions by modeling how infor-
mation about the voice is transmitted from speaker to hearer. Ideally, a unified theory of voice
production, acoustics, and perception should be able to model any type of information that the
voice can convey, including phonological, prosodic, discourse, sociolinguistic, paralinguistic, and
talker-specific. In this chapter, I focus largely on linguistic (especially phonological) meaning:
how the voice is used to create sounds of the world’s languages. Thus, the first goal of the phonetic
study of the voice can be restated with respect to phonological meaning: (10) When we perceive a
phonologically-relevant change in the voice (such as a switch between two sounds), what caused
that change?
The past decade has seen a rapid increase in research on the role of the voice in sound systems,
much of which will be reviewed in this chapter. This research will help outline the primary voice
dimensions, defined here in articulatory terms, that are used in languages of the world to convey
phonological meaning (Section 2). Readers will see that the study of the voice is essential for
our understanding of phonetics and phonology, because every sound that we make involves an
articulation of the vocal folds, which has specific acoustic attributes that listeners can hear and use
in language.
1.1 What is covered in this chapter?
This chapter is largely theoretical in nature. The focus is on how we can model (within a unified
framework) the three stages of the speech chain as it concerns the vocal folds and their use in
sounds of the world’s languages. Therefore, we will not review fine-grained details pertaining
to voice anatomy and physiology (though see Titze, 1994; Stevens, 2000; Titze, 2006; Reetz &
Jongman, 2008; Hirose, 2010; Kreiman & Sidtis, 2011; Gick et al., 2013; Story, 2015; Zhang,
2016b), as well as voice source modeling (Stevens, 2000; Gobl & N´
ı Chasaide, 2010; Story, 2012;
Samlan et al., 2013; Kreiman et al., 2015; Moisik & Esling, 2014; Moisik et al., 2014; Story, 2015).
Neither will we review instrumentation used to measure muscular and articulatory properties of the
voice; I refer readers to Baken & Orlikoff (2000); Hirose (2010); Gick et al. (2013), among others.
Given that this chapter focuses on linguistic meaning, readers who are especially interested in how
3

the voice varies according to specific voice disorders, emotions, individuals, and singing styles
should consult Laver (1980); Sundberg (1987); Titze (1994); Esling & Edmondson (2010); Gobl
& N´
ı Chasaide (2010); Kreiman & Sidtis (2011); Sapienza et al. (2011), among others. However,
this chapter should also be of use to these readers, in that we will review a theory of the voice that
links vocal production, acoustics, and voice quality perception more generally. Moreover, we will
discuss how the speech chain, as it concerns the vocal folds, relates to meaningful categories that
must be accurately perceived in order for communication to be effective.
In the following Section 2, I classify vocal fold articulations according to their primary phono-
logical dimensions. In Section 3, I review the psychoacoustic voice model by Kreiman et al. (2014)
and how parameters of this model are perceived by listeners and relate to the phonological dimen-
sions outlined in the previous section. In Sections 4 and 5, I discuss recent work showing how the
parameters of Kreiman et al. (2014)’s voice model relate to voice source acoustics and vocal fold
articulation as parameterized in a recent model by Zhang (2015, 2016a). In Section 6, I conclude
with discussion of outstanding questions and areas for future research.
2 Primary linguistic voice dimensions
Vocal fold movements, despite being very complex, can be organized along two articulatory di-
mensions that are especially important in language (Table 1): how far apart the folds are from
each other, and whether they are vibrating. (Further details on these articulations are presented in
Section 5.) In this chapter, I refer to this first dimension as vocal fold approximation, though it
could likewise be called ‘abduction/adduction’ and ‘spreading/constriction’. The minimal vocal
fold approximation in speech can be found during voiceless aspiration (e.g. for [h] and aspirated
stops like [th]), and the maximal vocal fold approximation, when the vocal folds are in full contact,
can be found for a glottal stop [P] as well as glottalized sounds and ejectives (e.g. [>
Pt, t’]). Note
that ‘minimal vocal fold approximation’ is understood within the context of speech sounds; for
example, during active breathing the vocal folds abduct even more than during aspirated speech
sounds (Gick et al., 2013). Other voiceless sounds can be described as having incomplete vocal
fold approximation that is nonetheless greater than that found for aspiration; this state of vocal
fold approximation is sometimes called ‘prephonation’ (Harris, 1999; Esling & Harris, 2003; Ed-
mondson et al., 2011). Likewise, voiced sounds require some degree of vocal fold approximation,
4

Front
Back
Minimal approximation 'Prephonation' Maximal approximation
Figure 1. Schematized laryngoscopic (bird’s-eye) view of the vocal folds showing different degrees
of vocal fold approximation for voiceless sounds. In between minimal approximation (e.g., for a
voiceless glottal fricative) and maximal approximation (e.g., for a glottal stop) is a state sometimes
called ‘prephonation’, which is found for voiceless unaspirated stops (Harris, 1999; Esling &
Harris, 2003; Edmondson et al., 2011).
which can also vary in its degree. Thus, all sounds, regardless of whether they are ‘plain’ voice-
less sounds, aspirated, glottalized, or voiced, involve some degree of vocal fold approximation. A
schematic of different states of vocal fold approximation for voiceless sounds is shown in Figure 1.
The second dimension, vocal fold vibration, is also often called voicing. This dimension
is dependent on vocal fold approximation: voicing can only be initiated with some amount of
vocal fold approximation, and only certain degrees of vocal fold approximation can sustain voicing
once it has begun (Titze, 1992; Kreiman & Sidtis, 2011). Consequently, all sounds with vocal
fold vibration necessarily make use of both dimensions. Voicing is generally treated as being
categorical: sounds produced with vocal fold vibration are ‘voiced’, those without are ‘voiceless’.
(Voicing is sometimes also called ‘phonation’, but this term is also used more generally to describe
any sound generated in the larynx or the remaining components of the vocal tract.) Voicing is very
important in languages’ phonologies; voiced sounds, including vowels, sonorant consonants, and
voiced obstruents, are found in every spoken language, and about 80% of languages have voicing
contrasts in obstruents (Maddieson et al., 2016).
Voicing can be further characterized along two dimensions that are very important for lin-
guistic meaning. The first is rate (or ‘frequency’) of vocal fold vibration, a continuous dimension
that determines the fundamental frequency of the voice and is used to convey lexical tone and
5

Table 1. Primary vocal fold movements and their use in sounds of the world’s languages. The
dashed line between rate vs. quality of voicing indicates that these two dimensions are often de-
pendent on each other.
Dimension Articulatory description Relevant sounds
Spreading-
Constriction
How far apart the vocal folds
are from each other
All voiced sounds
All voiceless sounds, e.g.
Aspirated sounds
Glottalized sounds
Fricatives
Trills
Ejectives
Voicing Whether the vocal folds are vibrating
All voiced sounds, e.g.
Sonorant consonants
(Voiced) vowels
Rate Rate of vibration
Tone
Intonation
Stress
Quality Constriction of vibration
Irregularity/noise
Register
Contrastive voice quality
(‘phonation type’)
intonation, and is one of the main correlates of lexical and phrasal stress (Gordon & Applebaum,
2010; Gordon, 2014; Garellek & White, 2015). The second dimension to voicing is the quality or
‘manner’ of vocal fold vibration. Voice quality is also important for stress, tone, and intonation
(Kreiman, 1982; Sluijter & van Heuven, 1996; Campbell & Beckman, 1997; Garellek & White,
2015; Mooshammer, 2010; Lancia et al., 2016); moreover, it is the primary dimension used for con-
trastive voice quality (‘phonation type’) and is important for voice registers, a multidimensional
linguistic contrast involving a change in voice quality and other (laryngeal and supralaryngeal)
changes (DiCanio, 2009; Brunelle, 2012; Abramson et al., 2015; Brunelle & Kirby, 2016).
The most common voice qualities that are used in language are ‘modal’, ‘breathy’, and
‘creaky’, and we will focus on these for the remainder of the chapter. They are defined relative to
one another; thus, there is no ‘absolute’ breathy voice, but certain voice qualities are ‘breathier’
than others. It is partly due to their relative differences that many names for voice qualities exist:
breathy voice qualities are sometimes called ‘lax, slack, murmured, aspirated’ while creaky quali-
ties are ‘stiff, tense, laryngealized, glottalized, (vocal) fry, pressed’, to name but a few. Other terms,
such as ‘rough’, ‘strident’, ‘sphincteric’, ‘epiglottalized’ or ‘harsh’ voice, tend to be used for voice
6

Most open Most closed
[h] Breathy Modal Creaky [ʔ]
Figure 2. Simplified one-dimensional space for voice quality (based on Ladefoged, 1971; Gordon
& Ladefoged, 2001).
qualities that also necessarily involve supraglottal constriction (Laver, 1980; Traill, 1985; Gerratt
& Kreiman, 2001; Edmondson & Esling, 2006; Miller, 2007; Moisik & Esling, 2011; Moisik,
2013) and thus fall outside the narrow definition of ‘voice’ used here.
Although these different terms can sometimes refer to the same articulation, often researchers
will use different terms to refer to distinct manners of vocal fold vibrations, acoustic characteristics,
or perceptual voice qualities (Batliner et al., 1993; Gerratt & Kreiman, 2001; Redi & Shattuck-
Hufnagel, 2001; Slifka, 2006; Kane et al., 2013; Kuang, 2013b; Keating et al., 2015). From an
articulatory perspective, differences between breathy, modal, and creaky voice can minimally be
described using a one-dimensional model of vocal fold approximation (Ladefoged, 1971; Gordon
& Ladefoged, 2001): as we have already discussed, voiceless sounds can be made with minimal
or maximal vocal fold approximation (as in [h] or [P], respectively). In between these extremes,
there is voicing. Voicing with less vocal fold approximation is ‘breathy’, voicing with more ap-
proximation is ‘creaky’, and voicing that is neither breathy nor creaky is ‘modal’. (‘Lax/slack’
voice is sometimes considered intermediate to breathy and modal, and ‘tense/stiff’ voice to creaky
and modal; Keating et al., 2011; Kuang & Keating, 2014.) This one-dimensional model of vocal
fold approximation is schematized in Figure 2. It is conceptually simple and useful for describing
the phonologically-relevant relationship between voice qualities and categories of voiceless (con-
sonantal) vocal fold approximation (Lombardi, 1991). However, it suffers from certain drawbacks.
First, ‘modal’ voice is defined in articulatory terms, relative to other states of vocal fold approx-
imation; however, many researchers also use this term to refer to a speaker’s default or ‘normal’
voicing. Defined thus, if a speaker’s normal voice quality is quite creaky, that speaker’s ‘modal’
voice would involve quite a bit of vocal fold approximation, and thus should be captured under
7

Table 2. Valves of the throat and their functioning (from Esling & Harris 2005; Edmondson &
Esling 2006; see also Figure 1 in Edmondson & Esling 2006). Valve 1 is similar to the continuum
represented in Figure 2. The remaining valves pertain to structures above the vocal folds, and
therefore would fall outside the narrow definition of ‘voice’ used in this chapter.
Valve Description
Valve 1 vocal fold adduction and abduction
Valve 2 ventricular fold adduction
Valve 3 compression of the arytenoids and aryepiglottic folds
Valve 4 epiglottal-pharyngeal constriction
Valve 5 larynx raising
Valve 6 pharyngeal narrowing
‘creaky’ in this model. The difference between these two definitions of ‘modal’ (one articulatory
and relative to non-modal voice, the other with reference to a speaker’s normal quality) is generally
thought to be of little practical importance, because we assume that speakers can always become
creakier or breathier than their normal voice quality; for instance, a Jalapa Mazatec speaker whose
normal voice quality is very creaky should be able to produce even creakier voice quality than her
default in order to say a word with contrastive creaky voice in the language. But given that we still
know very little about individual differences in voice quality – for example, if the degree of vocal
fold approximation in a particular speaker’s default voice quality has an effect on how that same
speaker will produce non-modal voice – there may be an important distinction to be made between
‘modal’ as defined articulatorily vs. speaker-dependently.
Second, although breathy and creaky voice are (generally) produced with different degrees
of vocal fold approximation, creaky voice can involve additional supraglottal constriction (e.g. of
the ventricular and aryepiglottic folds). The additional constrictions – and how these relate to
linguistically-relevant relationships between voice quality and supralaryngeal articulations (like
tongue root advancement and vowel quality) – are captured more straightforwardly in a ‘valves’
model of voice quality (see Table 2, after Esling & Harris, 2005 and Edmondson & Esling, 2006).
However, only Valve 1 from this model is involved in voice quality if the ‘voice’ is narrowly
defined as including only articulation of the vocal folds.
The third drawback to both continuum and valves models of the voice is that ‘creaky’ voice
– even if it is defined narrowly as a manner of vocal fold vibration with no additional supraglot-
tal articulation – represents a cluster of voice qualities that share some perceptual attributes. In
8

Prototypical
creaky voice
Irregular pitch Low pitch
Constricted voicing
Vocal fry
Tense voice
Unconstricted
creaky voice
Figure 3. Three-dimensional space for creaky voice, with some more common subtypes shown
(based on Keating et al., 2015).
order to be perceived as creaky, the voice must be minimally low in pitch, irregular in pitch, or
constricted-sounding (we will discuss what it means to be ‘constricted-sounding’ in the following
section), though not necessarily all three (Keating et al., 2015). For example, there are cases of
creaky voice that are irregular and low in pitch, but also unconstricted (Slifka, 2000, 2006), which I
call ‘unconstricted creaky voice’. Further, there are two distinct types of creaky voice that are con-
stricted in quality but regular in pitch: vocal fry (which is low and regular in pitch) and tense voice,
which is high-pitched and regular (Keating et al., 2015). A very low-pitched voice (e.g., below 60
Hz) can also be perceived as creaky, despite not having any constriction or irregularity (Keating
et al., 2015). Figure 3 illustrates the differences between prototypical creaky voice, unconstricted
creaky voice, tense voice, and vocal fry.
Therefore, either a low pitch, an irregular pitch, or a constricted quality (regardless of pitch)
is alone sufficient for listeners to perceive a voice as creaky, even if the articulatory origins and
acoustic attributes underlying this percept can differ. However, much more work is needed to
determine how perceptually distinct these subtypes can be (cf. Gerratt & Kreiman, 2001; Garellek,
9

2015), and how linguistically-relevant they are. We have evidence that ‘prototypical’ creaky voice
(low-pitched, irregular in pitch, and constricted) is linguistically important, since it is commonly
found for contrastive and allophonic creaky voice (Gordon & Ladefoged, 2001; Esposito, 2012;
Garellek & Keating, 2011). Both other types of constricted voices listed in Figure 3 are of linguistic
relevance: vocal fry can be used to perceive a glottal stop [P] (Hillenbrand & Houde, 1996; Gerfen
& Baker, 2005), and tense voice can be used for the phonetic realization of contrastive creaky
voice on a vowel with a high lexical tone, and is found more generally with higher-pitched lexical
tones (Garellek & Keating, 2011; Kuang, 2013b). Unconstricted creaky voice is found as a form of
phrase-final creak (Kreiman, 1982; Redi & Shattuck-Hufnagel, 2001; Garellek, 2015) in utterance-
final position when the subglottal pressure is low, at least for some speakers of American English
(Slifka, 2000, 2006). If unconstricted creaky voice occurs due to low subglottal pressure, it should
also occur in similar environments in other languages.
Given that creaky voice represents a cluster of vocal fold articulations and voice qualities,
one might be inclined to ask whether phoneticians should retain the more general term ‘creaky’
at all. I believe we should, because it is useful to have a word for the abstract phonological cate-
gory, which may be realized (in different phonological environments and/or by different speakers)
using different articulations. For instance, it is useful to have a ‘creaky’ phonological category in
Jalapa Mazatec, since this phonation type contrasts with more modal and breathy voice qualities.
However, we know that the ‘creaky’ voice type varies phonetically by lexical tone and preceding
consonant in the language (Garellek & Keating, 2011; Kuang, 2013a), and likely also by prosodic
position, talker, and other factors.
In sum, we can analyze languages’ use of voice primarily in terms of vocal fold approximation
and voicing, which in turn may differ by rate and quality. There are other dimensions of the voice,
including whisper and vocal intensity, but these generally play smaller roles in phonologies of the
world’s languages (cf. Fulop & Golston, 2008 for the role of whispery voice in White Hmong).
Vocal intensity, which is important for stress (Gordon & Applebaum, 2010), is controlled mostly
by the subglottal pressure, but also by vocal fold and supralaryngeal adjustments (Sundberg, 1987;
Zhang, 2016b). Having reviewed the primary dimensions of the voice used to make sounds of the
world’s languages, we will now turn to a model of how language users produce and perceive these
dimensions.
10

3 A psychoacoustic model of the voice
One of the biggest challenges with modeling the voice is that voice articulation, acoustics, and
perception are inherently multidimensional. The vocal folds have a complex structure and vibrate
in a three-dimensional space. Acoustically, one can model the voice source in both temporal and
spectral domains, and within each domain there are many different attributes to the voice that can
be parameterized (Cumming & Clements, 1995; Gobl & N´
ı Chasaide, 2010; Kreiman et al., 2010b,
2015). Voice perception is also extremely complex; there are dozens of ways of characterizing the
voice (Kreiman & Sidtis, 2011), with terminologies and relevant taxonomies varying by discipline.
For example, a whispery voice quality might not be a primary voice dimension in sounds of the
world’s languages, but it is common in speech and is associated with certain speech disorders
(Laver, 1980; Sapienza et al., 2011; Gick et al., 2013). And, as discussed in Section 1, models
of voice articulation and acoustics may not be able to account for all perceptual changes in voice
quality, or may have articulatory and acoustic parameters that are irrelevant for perception (Zhang
et al., 2013; Kreiman et al., 2014; Garellek et al., 2016b). Since our main goal in studying the voice
is to link what speakers do with their voices with what listeners hear, models of the voice should
link to voice perception; acoustic attributes are important only insofar as they are perceptible, and
articulatory attributes only insofar as they result in perceptible acoustic changes.
For these reasons, Kreiman et al. (2014) propose a psychoacoustic model of the voice, shown
in Table 3, which has the following parameters: four pertaining to the harmonic component of the
voice source spectrum (the sound produced by the vocal folds, before being filtered by the vocal
tract), another parameter for modeling the inharmonic component of the source spectrum (i.e.,
the noise), two temporal components of the voice source (f0 and amplitude), and the vocal tract
transfer function (which models the filtering from the vocal tract). Only the first three groups are
relevant for the voice as it concerns the vocal folds; though resonances and zeroes affect overall
voice quality, they are here considered independent of the voice source parameters (cf. Cumming
& Clements, 1995). This model makes several other important assumptions, notably that its param-
eters are both necessary and sufficient to model voice quality; thus, with one of these parameters
missing, voice quality cannot be faithfully modeled, and no other measures are needed to model
the voice.
The temporal parameters (the f0 and amplitude tracks) relate to the presence of voicing, its
11

Table 3. Components of the psychoacoustic model of voice quality and associated parameters
(from Kreiman et al. 2014).
Model component Parameters
Harmonic source
spectral slope (Figure 4)
H1–H2 (see description in text)
H2–H4
H4–H2 kHz
H2 kHz–H5 kHz
Inharmonic
source noise Harmonics-to-noise ratio (HNR)
Time-varying
source characteristics
f0 track
Amplitude track
Vocal tract
transfer function
Formant frequencies and bandwidths
Spectral zeroes and bandwidths
rate of vibration, and its amplitude. The harmonic spectral slope parameters represent differences
in harmonic amplitudes; thus, H1–H2 refers to the difference in amplitude between the first and
second harmonic (Bickley, 1982), H2–H4 is the difference in amplitude between the second and
fourth harmonics (Kreiman et al., 2007), H4–H2 kHz is the difference in amplitude between the
fourth harmonic and the harmonic closest to 2000 Hz, and H2 kHz–H5 kHz is the difference in
amplitude between the harmonic closest to 2000 Hz and the one closest to 5000 Hz (Kreiman et al.,
2011). Together, they characterize the voice’s ‘spectral tilt’ in various harmonic and frequency
bands.
The assumptions regarding the harmonic spectral slope model and its parameters warrant
further discussion. The first main assumption is that the spectral tilt parameters can be articulatorily
and perceptually independent of f0, even though it is clear that f0 changes are associated with
changes in spectral tilt (Kuang, 2013a; Garellek et al., 2016b). Thus, with a constant f0 = 100
Hz, a measure like H2–H4 will have a frequency bandwidth of 200 Hz (between 200 Hz and 400
Hz), whereas with a constant f0 of 200 Hz, the frequency bandwidth of H2–H4 will be 400 Hz
(between 400 and 800 Hz). Frequency bandwidth can vary for H1–H2 and H2–H4, as well as well
as H4–H2 kHz, though the latter parameter is bounded at the high end by a particular frequency of
2000 Hz. (The source spectrum model is therefore used only for f0 values below 500 Hz; higher
than that, and H4 will be equal to or surpass 2 kHz!) The final parameter H2 kHz–H5 kHz depends
on frequency bandwidth alone, not harmonic bandwidth. Overall then, we assume in this model
that voice quality is perceptually dependent on spectral tilt between fixed harmonics (regardless of
12

their frequency) in the lower-frequency end of the spectrum, and that voice quality is perceptually
dependent on spectral tilt between harmonics of fixed frequencies at the higher-frequency end of
the spectrum. It should be noted here that spectral tilt measures over fixed frequency bands have
also been shown to correlate with changes in quality (de Krom, 1995; Hartl et al., 2003; Samlan
et al., 2013; Samlan & Kreiman, 2014).
The second main assumption is that we need four spectral tilt parameters. Kreiman et al.
(2014) and Garellek et al. (2016b) motivate a model of the harmonic source spectrum with four tilt
parameters for several reasons. First, spectral tilt can vary independently according to frequency
band; thus, spectral tilt can be negative in one portion of the spectrum, but positive in another
(Kreiman et al., 2007; Garellek et al., 2016b). Another reason for modeling spectral tilt in terms of
multiple components is that distinct articulations might be responsible for different tilt components
(Zhang et al., 2013). Moreover, source spectra share certain inflection points; for example, it is
common for the harmonic slope to change abruptly around 2000 Hz. Finally, listeners are not
equally sensitive to every component of spectral tilt (Garellek et al., 2016b), and indeed sometimes
the slope of one component can cancel out the perceptual effect of another (Garellek et al., 2013).
We will also discuss the relevance of these parameters for linguistically-relevant dimensions of the
voice in the following section.
This model further assumes that it is the slope between harmonics (rather than amplitudes
of individual harmonics), that is perceptually relevant for quality (Kreiman et al., 2014; Garellek
et al., 2016b). For instance, we assume that it is of no importance whether H3 is louder than either
H2 or H4; the parameter H2–H4 depends only on the amplitudes of the harmonics adjacent to H3.
In our use of this model, we therefore alter the amplitude of the intermediate harmonics to conform
to the slope made between the two harmonic endpoints, as shown in Figure 4.
The final parameter of the model, the harmonics-to-noise ratio (HNR), refers to the differ-
ence in amplitude between the harmonic and inharmonic components of the source spectrum, as
measured in the cepstral domain (de Krom, 1993). The choice of HNR as a means of measur-
ing noise in this model assumes that spectral noise is psychoacoustically important (Kreiman &
Gerratt, 2005; Shrivastav & Sapienza, 2006; Zhang et al., 2013; Garellek et al., 2016b). Although
there are many time-domain noise measures like jitter and shimmer, these are not included because
they are not perceptually relevant independently of HNR (Kreiman & Gerratt, 2005). The current
13

Frequency (Hz)
Amplitude
H1
H2
H4
2 kHz
5 kHz
Figure 4. The four-parameter harmonic source spectrum model, fitted to the spectrum of a natural
voice. The voice source was estimated via inverse filtering, and its spectrum was then calculated
via fast Fourier transform. Differences in the amplitudes of individual harmonics are altered so
that they conform to the slope of the appropriate model segment (based on Kreiman et al., 2014).
version of the model also assumes that HNR over the entire frequency range is sufficient to model
the inharmonic component of the source spectrum. However, it is clear that noise interacts with
harmonic components in different frequency bands in distinct ways (Kreiman & Gerratt, 2012;
Garellek et al., 2016b), and that changes in linguistic uses of voice quality are expressed with more
narrowband noise measures (Garellek, 2012). Therefore, much more work is needed to determine
how best to model noise and its role in causing changes in voice quality.3
A note on how to obtain measurements for these model parameters: Kreiman et al. (2014) and
Garellek et al. (2016b) calculate these model parameters using inverse filtering of audio recordings
followed by analysis-by-synthesis. Although this allows for very accurate source and filter esti-
mation, the main disadvantages of this process are that it is time-consuming and difficult to do on
conversational speech (see Gobl & N´
ı Chasaide, 2010; Kreiman et al., 2010a for more discussion
on the matter). However, these model parameters can be estimated from the audio signal if formant
correction is used, e.g. with a program like VoiceSauce (Shue et al., 2011). This practice is often
used in phonetic studies, but requires accurate formant measurements, which can be problematic
3In fact, some current versions of this psychoacoustic voice model parameterize the spectral slope of the inharmonic
noise in four frequency bands (Kreiman et al., 2016).
14

with certain sounds. This is discussed in more detail in Section 4.2.2.
In the following sections, I will show how the primary linguistic dimensions of the voice can
be expressed and measured using this psychoacoustic model, and how vocal fold articulation as it
concerns these dimensions relates back to the model’s parameters.
4 Acoustic properties of the primary phonological voice dimensions
The primary phonological voice dimensions (vocal fold approximation, voicing, rate of vibration,
and quality of voicing) have clear acoustic ramifications that can be measured using the parameters
of the psychoacoustic voice model outlined in Section 3.
4.1 Vocal fold approximation
Vocal fold approximation during voiceless sounds has both direct and indirect acoustic conse-
quences. Direct consequences include the absence of periodic energy, and, for minimal vocal
fold approximation (i.e., maximal abduction/spreading), the presence of aspiration noise, which in
Kreiman et al. (2014)’s model can be measured using HNR (higher values of the measure indicate
less noise). Aspiration noise can also be seen in spectra and spectrograms as broadband noise
(though filtered by the shape of the vocal tract); in waveforms, aspiration noise is hard to distin-
guish from other sources of noise, such as frication from the vocal tract. Indirect consequences of
vocal fold spreading can be seen on adjacent voiced sounds, which will be breathier, as discussed
in Section 4.2.2.
Complete vocal fold approximation, on the other hand, is hard to measure using parameters
in Kreiman et al. (2014)’s model, and is hard to see directly in either temporal or spectral domains;
if the vocal folds are constricted but not vibrating, then there is no acoustic energy that is produced
during the constriction. However, as vocal folds transition from voicing to and from complete con-
striction, there are indirect consequences present in form of creaky voice on the adjacent sounds. In
sum, vocal fold approximation can be measured indirectly through its effects on adjacent voicing,
which we will discuss more below.
15

Frequency Amplitude
Time
[à] [h] [á]
Figure 5. Waveform and spectrogram of the sequence [`ah´a]. The presence of voicing is indicated
by the glottal pulses in both displays, as well as by the presence of an f0 track (shown in the
dashed line). The intensity track (the solid horizontal line in the display) provides information on
the amplitude of the signal during voicing.
4.2 Voicing
Vocal fold vibration usually produces a complex (quasi-)periodic wave; that is, a wave with many
frequency components (Stevens, 2000). The slowest component is the fundamental frequency (f0),
and the faster components are whole-integer multiples of the f0 (see Figure 4). Thus, if voicing
is present, there must be an f0. Figure 5 illustrates how an f0 track in the sequence [`ah´a] aligns
with the presence of glottal pulses (the vertical striations) in the waveform and spectrogram. The
intensity of voicing (roughly, ‘how loud’ the voice is) can be measured by the psychoacoustic
model using the amplitude track, also shown in Figure 5.
4.2.1 Rate of vibration
Differences in rate (or frequency) of vocal fold vibrations are reflected in the acoustic signal pri-
marily through changes to the fundamental frequency and its change over time. In the temporal
domain, this means that the periodic wave produced during voicing will recur more quickly or
slowly, resulting in more closely- or distantly-spaced glottal pulses. This can be seen in Figure 5:
where the glottal pulses are closer together (i.e. in the second vowel), the f0 track is higher. In the
16

spectral domain, f0 is the frequency of the first harmonic, though here temporal change in f0 is not
calculable. Using Kreiman et al. (2014)’s model, we can measure f0 change using the f0 track.
4.2.2 Voice quality
As discussed above, the three main voice qualities (breathy, modal, and creaky voice), as well as
subtypes of creaky voice, can be described in terms of their degree of constriction (or spreading)
and noise: relative to modal voice, breathy voice is both more spread and noisier (because of the
presence of aspiration noise), whereas prototypical creaky voice is more constricted and noisier
because of irregular pitch.
These two basic dimensions to voice quality – spreading/constriction and noise – have spe-
cific acoustic attributes in the psychoacoustic model proposed by Kreiman et al. (2014). The most
reliable correlate of increased spreading or constriction during voicing is through increased spec-
tral tilt for spreading and decreased spectral tilt for constriction (Klatt & Klatt, 1990; Gordon &
Ladefoged, 2001; Hanson et al., 2001; Kreiman et al., 2012; Chen et al., 2013; Samlan et al., 2013;
Keating et al., 2015; Zhang, 2016a). For the four spectral tilt parameters discussed earlier (H1–H2,
H2–H4, H4–H2 kHz, and H2 kHz–H5 kHz), greater spectral tilt due to breathy voice would be in-
dicated by higher values of these measures. Of these four spectral slopes, it is clear that at least H1–
H2 and H2–H4 are relevant phonologically. For example, native speakers of White Hmong can use
changes in either of these slopes to perceive their contrastive breathy-voiced tone (Garellek et al.,
2013), and many other studies on a variety of languages have shown that breathy and creaky voice
correlate with changes in H1–H2 (Bickley, 1982; Gordon & Ladefoged, 2001; Blankenship, 2002;
Miller, 2007; DiCanio, 2009; Brunelle & Finkeldey, 2011; Garellek & Keating, 2011; Garellek,
2012; Esposito, 2012; Khan, 2012; Berkson, 2013; DiCanio, 2014; Yu & Lam, 2014; Abramson
et al., 2015; Zhang & Yan, 2015). Listeners of languages with contrastive or allophonic non-
modal phonation show differing degrees of sensitivity to H1–H2 (Kreiman et al., 2010b; Kreiman
& Gerratt, 2010), and H2–H4 is also relevant for listeners’ identification of speaker sex (Bishop &
Keating, 2012). On the other hand, it is still unclear whether the higher-frequency slopes H4–H2
kHz and H2 kHz–H5 kHz contribute to linguistically-relevant voice distinctions, though recent
work suggests that they, along with H1–H2 and H2–H4, help differentiate creaky vowels from
non-creaky ones in American English (Garellek & Seyfarth, 2016).
17

Other measures of spectral tilt over different frequency bands, such as H1–A1, H1–A2, and
H1–A3 (differences in amplitude between the first harmonic and the harmonic closest to the first,
second, and third formants) have proved useful for distinguishing voice qualities or registers in
languages of the world (N´
ı Chasaide & Gobl, 1993; Hanson, 1997; Gordon & Ladefoged, 2001;
Hanson et al., 2001; Wayland & Jongman, 2003; Gerfen & Baker, 2005; Andruski, 2006; DiCanio,
2009; Esposito, 2010a,b; Avelino, 2010; Garellek & Keating, 2011; Brunelle, 2012; Berkson, 2013;
Kirby, 2014; Abramson et al., 2015). There are several important remarks to make on the similar-
ities and differences between these measures and the spectral tilt parameters proposed by Kreiman
et al. (2014). First, both types of measures use harmonic-based bandwidths, rather than fixed fre-
quency ranges (see discussion in Section 3). Morevoer, the harmonics closest to a formant (i.e.,
A1, A2, and A3) are not defined in a source model. One consequence of this is that measures
like H1–A1, H1–A2, and H1–A3 are correlated and can even overlap with the source spectral tilt
measures like H1–H2. For instance, the vowel [i] spoken by the average adult male speaker of
American English has an F1 at around 340 Hz (Hillenbrand et al., 1995). So if an adult male
speaker of American English says [i] with an f0 of 150 Hz, the harmonic closest to F1 is H2,
which means that, for this token, H1–H2 would be equal to H1–A1 (see the left panel of Figure 6).
Thus, depending on context, it might not make sense to measure both H1–H2 and H1–A1 for the
same token. However, this issue would not occur when using source spectral parameters like the
kind found in Kreiman et al. (2014)’s model, because the model parameters can never overlap in
frequency. Of more theoretical relevance is the fact that, for measure like H1–A1, what counts as
‘A1’ will vary by context. For instance, if the same adult male speaker from the previous example
were to say the vowel [A] with an f0 = 150 Hz, then H1–A1 would be equal to the difference in
amplitude between the first harmonic and the harmonic closest to roughly 770 Hz, which is H5
(at roughly 750 Hz, see the right panel of Figure 6). Measures like H1–A1, H1–A2, and H1–A3
vary in their harmonic bandwidth depending on the formant frequencies; on the other hand, source
spectral tilt measures like H1–H2 and H2–H4 have fixed harmonic bandwidths and therefore do
not depend on vowel quality. Thus, use of measures like H1–A1 assumes that the ways in which
spectral tilt can determine voice quality necessarily depend on the vowel quality. For instance, if
I compare Mazatec breathy vs. modal vowels of different qualities using H1–A1 (as in Garellek
& Keating 2011), then I assume that a relevant way of distinguishing these voice qualities is by
18

Frequency (Hz)
03500
Sound pressure level (dB/Hz)
0
20
40
60
Frequency (Hz)
03500
Sound pressure level (dB/Hz)
0
20
40
60
Frequency (Hz)
03500
Sound pressure level (dB/Hz)
0
20
40
60
A1 A1
Figure 6. FFT and LPC spectra for synthetic [i] (left) and [A] (right) with f0 = 150 Hz and
formants equal to that of an average adult male speaker of American English. For [i] (left), the
second harmonic (H2) is also the harmonic closest to the first formant (A1), so H1–H2 equals
H1–A1. For [A] (right), H1–A1 equals H1–H5.
comparing the spectral tilt between H1 and the first formant, regardless of the harmonic that is
most affected by that formant. On the other hand, using measures such as H1–H2, H2–H4, H4–H2
kHz, and H2 kHz–H5 kHz implies that voice source characteristics are relevant independent of the
filter; although Kreiman et al. (2014) assume that both the source and filter can influence voice
quality, they further assume that measuring spectral tilt within fixed harmonic or frequency bands
is a relevant way of distinguishing voice quality. It is still unclear which method of represent-
ing spectral tilt more closely reflects perception of voice quality; i.e., whether listeners perceive
changes in quality more as a function of formant- or harmonic-based differences in spectral tilt.
The second important dimension to voice quality is noise, which in the psychoacoustic voice
model is calculated in the spectral domain as harmonics-to-noise ratio (HNR). Since breathy voice
has aspiration noise (due to the increase in vocal fold spreading or lower vocal fold thickness, see
Section 5), this will lower the HNR. Numerous studies have shown that HNR measures are useful
for distinguishing breathy vs. non-breathy voice qualities used in language (Gordon & Ladefoged,
2001; Blankenship, 2002; Brunelle, 2012; Berkson, 2013; Esposito, 2012; Garellek, 2012; Khan,
2012; Kuang, 2012; Simpson, 2012). Creaky voice also lowers HNR, but usually this is because
of its irregular pitch: if the f0 is not regular, the signal’s noise will increase. Various studies also
provide evidence for the use of HNR in distinguish creaky vs. non-creaky voice qualities in lan-
guage (Blankenship, 2002; Esposito, 2012; Garellek, 2012, 2015; Keating et al., 2015; Garellek &
19

Table 4. Sample H1–H2 and HNR values for two groups of Vowels A vs. B and A0vs. B0. Even
though Vowels A and A0share the same H1–H2 (as do B and B0), the HNR differences suggest that
Vowel B is creakier than Vowel A, whereas Vowel A0is breathier than Vowel B0.
Vowel A Vowel B Vowel A0Vowel B0
H1–H2 10 5 10 5
HNR 20 10 10 20
Interpretation A has higher tilt, less noise than B. A0has higher tilt, more noise than B0.
A = modal, B = creaky A0= breathy, B0= modal
Seyfarth, 2016). ‘Harsh’ voice qualities, which are not reviewed in detail here because they neces-
sarily involve supraglottal constriction (Edmondson et al., 2001; Edmondson & Esling, 2006), can
also have lower HNR (Miller, 2007), due at least in part to the supralaryngeal noise.
Phoneticians often use acoustic measures to categorize different voice qualities. For example,
the lower values of H1–H2 in Category A compared with Category B might be used to justify
an analysis in which Category A is creaky and Category B is modal. There are, however, two
important things to keep in mind with regard to this practice. First, spectral tilt measures like
H1–H2 vary continuously between more constricted creaky voice qualities (which have lower H1–
H2) and less constricted or breathier voice qualities (which have higher H1–H2); modal voice’s
spectral tilt is somewhere in between that of constricted creaky and breathy voice. But raw values
of spectral tilt measures do not index a precise voice quality; one person’s creaky voice can have
an average H1–H2 of -2 dB while another person’s creaky voice averages 5 dB. Thus, if Category
A has a higher H1–H2 than Category B, we cannot know whether the difference between A and B
is one between more modal vs. creaky voice, between more breathy vs. modal voice, or between
more breathy vs. more creaky voice. This is why spectral tilt measures are often interpreted with
respect to noise measures like HNR (Blankenship, 2002; Garellek, 2012; Simpson, 2012; Garellek
& White, 2015): if Category A has both a higher H1–H2 and a higher HNR than Category B,
then we can assume A is more modal than B, because modal voice generally has higher H1–H2
and HNR values than creaky voice. But if Category A0has a higher H1–H2 and a lower HNR
than Category B, then we can assume A0is breathier than B0, because breathy voice generally
has higher H1–H2 and lower HNR values than modal voice (see Table 4). Figure 7 illustrates the
relationship between breathy, modal, and (prototypical) creaky voice in a two-dimensional acoustic
space consisting of spectral tilt and HNR.
20

Lower spectral tilt Higher spectral tilt
Higher HNR
⃝ Prototypical
creaky voice
⃝ Breathy
Lower spectral tilt Higher spectral tilt
⃝ Modal
⃝ Prototypical
creaky voice
⃝ Breathy
Lower HNR
Figure 7. Relative acoustic differences between breathy, modal, and (prototypical) creaky voice, in
terms of both spectral tilt and noise. The difference in position of breathy voice and prototypical
creaky voice on the HNR scale is arbitrary; cf. Blankenship 2002 and Garellek 2012.
The second caveat pertains to the fact that lower spectral tilt measures correlate with increased
constriction. As we have discussed earlier, not all subtypes of creaky voice are more constricted or
noisy than modal voice. For instance, tense voice is constricted, high in pitch, and regular in pitch,
which means that we would not expect to find a decrease in HNR for tense voice relative to modal
voice. The acoustic characteristics (in terms of both spectral tilt and noise) of these types of creaky
voice are shown in Table 5.
Another difficulty with measuring voice quality is that when speech sounds travel through air,
they bear evidence both of voice characteristics and of the supralaryngeal modulations of the vocal
tract. It can sometimes be challenging to infer whether a particular acoustic property is due to
voice or supralaryngeal vocal tract articulations: for instance, nasalized vowels (caused by velum
lowering) and breathy vowels (caused by vocal fold spreading during voicing) both have higher
spectral tilt (Klatt & Klatt, 1990; Simpson, 2012; Garellek et al., 2016a). There are several options
for researchers who wish to disentangle confounding effects of manner of voicing and suprala-
ryngeal articulation. First, one can use ‘inverse filtering’ to remove the acoustic effects of other
articulators, and measure manner of voicing from the source waveform (often with the additional
step of modeling the waveform using a variety of voice source models, N´
ı Chasaide & Gobl, 1993;
21

Table 5. Summary of psychoacoustic voice model’s parameters according to primary phonological
dimensions of voice.
Dimension Relevant model parameters
Vocal fold approximation
Absence of f0 track
Aspiration noise (if vocal folds are spread)
Voice quality changes on adjacent voiced sounds
Voicing Presence of f0 track
Rate of vibration Frequency of f0 track
Voice quality
(compared with modal)
Breathy voice:
Higher H1–H2, H2–H4, H4–H2 kHz, H2 kHz–H5 kHz
Lower HNR
Unconstricted creaky voice:
Higher H1–H2 H2–H4, H4–H2 kHz, H2 kHz–H5 kHz
Lower HNR
Lower f0
Constricted creaky voice qualities
(Prototypical creaky, tense voice, and vocal fry):
Lower H1–H2 H2–H4, H4–H2 kHz, H2 kHz–H5 kHz
Lower HNR (prototypical creaky voice)
Lower f0 (prototypical creaky and vocal fry)
Epstein, 2002; Gobl & N´
ı Chasaide, 2010; Kreiman et al., 2015). Another option is to measure
spectral tilt and noise from the audio output spectrum, but to ‘correct for’ or undo the effects of
the supralaryngeal articulators (Hanson, 1995, 1997; Iseli et al., 2007). However, it should be
noted that formant corrections are meant to undo the effects of vowel formants specifically; thus,
they cannot remove the effects of nasalization or consonantal resonances (Simpson, 2012; Garellek
et al., 2016a). Of course, formant corrections will also fail whenever the formants are mistracked.
This is especially common with high-pitched and breathy voices where the high-frequency and
high-energy of the first harmonic can be misidentified as a formant. It is thus recommended that
researchers check the f0 and formant frequencies to ensure that the spectral slope measures are not
mistracked due to inaccurate tracking of f0 or (for ‘corrected’ measures) of the formants . Spec-
tral slope measures whose harmonic amplitudes have been corrected for effects of formants and
bandwidths are usually denoted with asterisks, e.g. H1*–H2*. Readers should therefore be aware
that different researchers will use a label such as ‘H1–H2’ to refer either to uncorrected spectral
tilt derived from the output audio spectrum, or to the measure calculated directly from the source
spectrum.
22

As mentioned at the start of this section, it is impossible to measure voice quality during a
voiceless sound, which makes use of vocal fold approximation as its primary voice dimension.
However, differences in vocal fold approximation can have ramifications for voice quality on an
adjacent voiced sound. Numerous linguistic studies have shown that vowels adjacent to aspirated
sounds are breathier than vowels adjacent to non-aspirated ones; conversely, vowels adjacent to
glottalized sounds (including ejectives) tend to be creaky compared to vowels adjacent to non-
glottalized ones (L¨
ofqvist & McGowan, 1992; N´
ı Chasaide & Gobl, 1993; Blankenship, 2002;
Vicenik, 2010; DiCanio, 2012; Esposito & Khan, 2012; Garellek, 2012). This relationship between
voiceless consonants and voice quality during vowels is captured easily in Ladefoged’s continuum
model, because both voiceless vocal fold spreading/constriction and voice quality are modeled
along a single dimension (analogous to our ‘vocal fold approximation’ dimension, see Figure 2).
For instance, in the sequence [ah], the vocal folds will have to transition from modal voicing to
minimal vocal fold approximation, necessarily ‘passing through’ breathy voice. This transition
can be seen during the first vowel of [`ah´a] from Figure 5. Using VoiceSauce (Shue et al., 2011) to
measure the four spectral tilt parameters from the audio spectra (but correcting for vowel formants),
as well as harmonics-to-noise ratio (HNR), it is clear that the vowel is breathier in the second half,
nearest the [h] (see Figure 8): H1*–H2*, H2*–H4*, and H4*–H2 kHz* are higher, whereas HNR is
lower (indicating greater noise). On the other hand, H2 kHz*–H5 kHz* is lower in the second half,
which is likely due to interactions with high-frequency noise (Kreiman & Gerratt, 2012; Garellek
et al., 2016b).
Therefore, we can use various parameters of the psychoacoustic model described in Section 3
to measure acoustic changes in the voice associated with vocal fold approximation, voicing, and its
rate and quality. The crucial parameters are f0 (its presence vs. absence, and its value when present)
as well as the source spectral tilt parameters and HNR. The formant parameters of Kreiman et al.
(2014)’s model are also important for correcting for spectral tilt parameters like H1*–H2*, which
are measured from the output audio spectrum, and when measuring a parameter like H1–A1, which
makes reference to formants. In the next section, we show how a model of voice articulation can
account for changes in these parameters.
23

Frequency (Hz)
05000
Sound pressure level (dB/Hz)
-20
0
20
40
5000500000
Frequency (Hz)
First half Second half
H1*–H2* (dB) 1.8 2.6
H2*–H4* (dB) 2.3 10.1
H4*–H2 kHz* (dB) 19.1 27.4
H2 kHz*–H5 kHz* (dB) 0.9 –5.7
CPP (a measure of HNR, dB) 21.3 19.0
Figure 8. Top: audio spectra of the first vowel in the sequence [`ah´a].The left panel is the spectrum
taken over the first half of the vowel; the right panel over the second half. Bottom: spectral tilt
parameters and CPP (Cepstral Peak Prominence, a measure of HNR). All acoustic measures but
H2 kHz*–H5 kHz* indicate greater breathiness in the second half compared with the first.
5 Voice production
Voice production is remarkably complex and depends on lung pressure and several muscles that,
working together, alter the shape and stiffness of the vocal folds, the distance between them and
other laryngeal structures, and the position of the larynx as a whole. There currently exist numerous
excellent sources on laryngeal anatomy and physiology as they pertain to speech (e.g., Titze, 1994;
Stevens, 2000; Reetz & Jongman, 2008; Hirose, 2010; Kreiman & Sidtis, 2011 and Gick et al.,
2013), but in this section we will focus on the articulations of the vocal folds that are associated
with the primary dimensions of the voice that are used in language (vocal fold approximation and
voicing, the latter of which can be further characterized by its rate and manner), and how these
relate back to the psychoacoustic voice model discussed in Section 3.
Measuring vocal fold articulation is fraught with challenges. The first main challenge is that
the vocal fold dynamics are multidimensional, which makes it hard to determine what aspects of
vocal fold articulation we should be measuring in the first place. For instance, many researchers
have noted (via scoping) that individuals with voice disorders have asymmetric vocal fold vibra-
24

tion. Crucially though, asymmetric vibration is also very common in individuals with no voice
disorders (Bonilha et al., 2012), and not all vocal fold asymmetries produce relevant changes in
voice quality (Zhang et al., 2013; Samlan et al., 2014). The second main challenge is method-
ological: it is very hard to see and thus measure the vocal folds. Direct observation (e.g. via
laryngoscopy) is invasive and limits the types of sounds speakers can make while being scoped
and the types of sounds we can observe during imaging. And while direct observation of the vocal
folds provides extremely important information about vocal kinematics, it also only allows for a
two-dimensional bird’s-eye view of the superior part of the vocal folds.
Because of these challenges, there is a longstanding tradition (e.g., van den Berg & Tan, 1959)
of using physical and, more recently, computational models of vocal fold articulation, which enable
researchers to lower the degrees of freedom and determine how individual parameters affect vocal
production, acoustics, and quality (Flanagan, 1972; for a recent overview, see Zhang, 2016b). In
the remainder of this section, I will outline one such model, and show how its parameters can be
used to understand how vocal articulation leads to linguistically-relevant changes in voice quality.
5.1 Modeling voice articulation
Although the vocal folds are anatomically and physiologically complex, we can describe their
linguistically-relevant dimensions using a simplified model (Figure 9, after Zhang (2015, 2016a)).
This three-dimensional model of the vocal folds has been used to simulate computationally the
effects of various vocal fold parameters on voicing. Although there are many other models of
vocal fold vibration (Isogai et al., 1988; Titze et al., 1995; Titze, 2006; Samlan & Story, 2011;
Story, 2012; see also a recent overview in Zhang, 2016b), Zhang’s recent model systematically
relates model parameters to acoustic ones, some of which crucially appear in the psychoacoustic
voice model described earlier. Therefore, it is particularly useful for assessing the cause-and-effect
relationship between voice articulation, acoustics, and perception.
The relevant parameters are vocal fold stiffness in the front-back dimension (represented by
the oblique arrow in Figure 9), medial surface thickness in the vertical direction, the angle be-
tween the vocal folds (the horizontal arrow), and subglottal pressure. This model makes several
assumptions (which are described in detail in Zhang 2015, 2016a), and at present does not include
interactions between the subglottal and supraglottal tracts. And though the vocal folds have multi-
25

Front
Stiffness
Top
Left
Glottal width
Medial
thickness
Figure 9. Simplified model of the vocal folds, after (Zhang, 2015, 2016a). The primary voice
dimensions are influenced mainly by glottal width (the angle between the folds, thin unbroken
arrow), their medial vertical thickness (thin dashed arrow), their stiffness from front to back (thick
dashed arrow), and the interactions of these parameters with the subglottal pressure.
ple layers (Hirose, 1997; Kreiman & Sidtis, 2011; Gick et al., 2013; Zhang, 2016b) and are often
simplified as a two-layered structure composed of a body and a cover (Hirano & Katika, 1985),
Zhang currently models the folds as a one-layer structure because the vocal fold body and cover
rarely differ in stiffness. Moreover, only front-back stiffness is modeled because different degrees
of muscular activation had strong effects on this dimension but much smaller effects on transverse
stiffness (Yin & Zhang, 2013).
All of these parameters can vary in degree with changes in activation of laryngeal muscles
as well as respiration. In the remainder of this section, I will review how these model parameters
have been shown to, or might eventually be shown to, produce the primary linguistic dimensions
of the voice that are used in sounds of the world’s languages. A summary is shown in Table 6.
5.2 Vocal fold approximation
Zhang (2015, 2016a) models vocal fold articulation during voicing, and not voiceless vocal fold
approximation used for aspirated and glottalized sounds. Nonetheless, we know that vocal fold ap-
proximation can be described by a continuum of glottal width, following the continuum and valves
models described in Section 2. This dimension is indeed parameterized in Zhang (2015, 2016a)’s
26

model as the angle between the two folds, or ‘glottal width’ (the green arrow in Figure 9). This
should therefore be the primary parameter responsible for changes in vocal fold approximation that
are used to make voiceless aspirated and glottalized consonants.
As we reviewed in Section 2, other voiceless sounds can be described as being in a ‘prephona-
tion’ state, during which the vocal folds are close together but not completely adducted, nor spread
enough to produce much aspiration noise (Harris, 1999; Esling & Harris, 2003; Edmondson et al.,
2011). This too can be modeled articulatorily using the glottal width parameter in Zhang (2015,
2016a)’s model. Yet we also know that voiceless unaspirated (and voiceless aspirated) stops are
often followed by a rise in f0 (Hombert et al., 1979). This may imply that the mechanisms in-
volved in f0 control (Section 5.3.1) can also be involved in the production of voiceless stops. On
the other hand, increased activation of the cricothyroid muscle (which results in increased vocal
fold stiffness) is not clearly associated with production of voiceless stops (Hirose & Gay, 1972;
Hombert et al., 1979); thus, Hombert et al. (1979) speculate that raising of the whole larynx, which
indirectly affects vocal fold stiffness and thus f0, can be responsible for this effect (see also Honda
et al., 1999; Stevens, 2000; Brunelle, 2010).
5.3 Voicing
The mechanism of voicing is usually characterized by the myoelastic aerodynamic theory (van den
Berg, 1958) and its more recent extensions (Titze, 2006). According to this theory, the combination
of tissue elasticity (e.g. altering the stiffness of the intrinsic laryngeal muscles) and aerodynamic
forces is responsible for initiating, sustaining, and ending the vibration of the folds. Vibration
usually cannot start until after the vocal folds are brought together or nearly so, in order to build
up a subglottal pressure (3-5 cm H20 for modal voice, Titze, 1992). In his model simulations,
Zhang (2016a) found that the most important parameter to voicing onset is the angle between the
vocal folds: the greater the angle between the folds, the higher the pressure must be to initiate
voicing (see also Titze, 1992). Vocal fold thickness also matters: the thinner the folds, the higher
the pressure needed to initiate voicing, which Zhang (2016a) attributes to the difficulty of very
thin vocal folds to maintain a degree of glottal opening that is conducive to voicing against the
subglottal pressure. When the vocal folds are thin, their prephonatory glottal opening is much
larger than the resting glottal opening, which makes it difficult to initiate voicing. But as vocal
27

fold thickness increases, the resting glottal opening is easier to maintain. This enables voicing to
begin at lower pressure, unless the vocal folds are so thick that additional pressure is needed to
keep them from remaining closed.
The findings regarding voicing initiation pressure are also important for sounds of language,
because at edges of utterances (when the subglottal pressure is low), voicing is accompanied by
specific vocal fold changes. For instance, stressed word-initial vowels in English and other lan-
guages are often glottalized (as in saying the word ‘after’ with a glottal stop [PæftÄ]), especially
phrase- and utterance-initially and when stressed (Nakatani & Dukes, 1977; Umeda, 1978; Pier-
rehumbert & Talkin, 1992; Dilley et al., 1996; Davidson & Erker, 2014; Garellek, 2013, 2014).
Since glottalization involves vocal fold approximation, it would be parameterized with a smaller
angle between the vocal folds in this model. The smaller angle between the vocal folds is also
associated with lower voicing initiation pressure, which would be beneficial for utterance-initial
stressed vowels; these must be strongly voiced (to mark stress) despite the low subglottal pressure
(Garellek, 2014).
5.3.1 Rate of vibration
Consistent with earlier work (e.g. Stevens, 2000), Zhang (2016a) found that rate of vibration de-
pends on three parameters; f0 increases with greater vocal fold stiffness, subglottal pressure, and
vocal fold approximation. The role of increased vocal fold approximation in achieving higher f0
is interesting for linguistic tonal patterns, because sounds with increased vocal fold adduction are
often accompanied by higher f0 (e.g., Korean fortis stops), and high tones are often accompanied
by increased constriction (Kingston, 2005; Kuang, 2013b).
However, the articulatory parameters interact with one another in complicated ways that affect
f0 (and voice quality, which we discuss in the following section), because in human voices the
vocal fold stiffness will covary with the other model parameters. For example, Zhang (2016a)
finds that the f0 can be raised by decreasing the glottal width (with or without an increase in
the subglottal pressure), but that this is likely to come with greater vocal fold thickness and low
front-to-back stiffness. The acoustic result is a higher f0 with more constriction and decreased
spectral tilt – essentially, the characteristics of tense voice. On the other hand, just raising the
vocal fold stiffness parameter will likely be accompanied by decreased vocal fold thickness, which
28

Zhang (2016a, p.1506) says will be falsetto-like in quality. Falsetto voice is used as a singing
register (Sundberg, 1987), but can also be used in language to index various emotions and types
of sociolinguistic meaning (Callier, 2013; Stross, 2013; Zimman, 2013; Podesva & Callier, 2015;
Starr, 2015). It may also be used for phonological purposes, e.g. as the phonetic implementation of
very high-pitched tones such as the high level tone in Black Miao and the falsetto tones of Hubei
Chinese (Kuang, 2013b; Wang & Tang, 2012; Wang, 2015).
5.3.2 Voice quality
In Zhang (2016a)’s model, changes in voice quality are driven by vocal fold approximation, sub-
glottal pressure, vocal fold thickness, and their interactions. Not surprisingly, lower vocal fold
approximation (i.e., a greater angle of glottal width) is associated with higher noise (as measured
by HNR); within limits, the more the vocal folds are spread, the more turbulent airflow is gener-
ated at the glottis. Lower subglottal pressure is also associated with higher noise, because voicing
is weaker and less regular in this condition; not surprisingly, languages often have breathy and/or
irregular creaky voicing at utterance edges, where the subglottal pressure is low (Rodgers, 1999;
Ogden, 2001; Redi & Shattuck-Hufnagel, 2001; Slifka, 2003, 2006; Esling & Edmondson, 2010;
Garellek, 2014, 2015; Di Napoli, 2015; Garellek & Seyfarth, 2016).
Interestingly, increasing vocal fold thickness has a strong influence on spectral tilt. Zhang
(2016a) measures tilt using H1–H2, as well as larger frequency bands with reference to H1 (H1–
H4, H1–H2 kHz, and H1–H5 kHz); aside from H1–H2, the other bands are not equivalent to
the parameters from Kreiman et al. (2014)’s psychoacoustic model, though his results should be
comparable to some extent. In his simulations, increasing vocal fold thickness results in lower
spectral tilt in all frequency bands. Therefore, we would expect measures like H2–H4, H4–H2
kHz, and H2 kHz–H5 kHz to also be lower with increasing thickness. The combination of thick
vocal folds with tight approximation, and very low stiffness and subglottal pressure, produces a
voice quality that Zhang (2016a) describes as vocal fry-like, with a low f0, spectral tilt, and noise.
Of the articulatory model’s four parameters, only vocal fold thickness had a sizable effect
on spectral tilt measures. This has implications for our understanding of linguistic voice quality,
because breathy, modal, and creaky voices are usually analyzed and modeled in terms of glottal
width (see the continuum and valves models in Section 2). On the other hand, Zhang (2016a)’s
29

results imply that vocal fold thickness should matter more than width in producing changes in
voice quality associated with changes in spectral tilt. But because glottal width influences f0, this
parameter might be especially relevant for voice quality changes associated with specific lexical
tones (Kingston, 2005; Kuang, 2013b). And even though vocal fold thickness might contribute
more to changes in spectral tilt in model simulations, in human voices this parameter is likely to
covary with others in ways which we have yet to fully understand.
Table 6. Summary of Zhang (2015, 2016a)’s articulatory voice model’s parameters according
to primary phonological dimensions of voice. The initiation and sustaining of voicing depends
on complex interactions between the four parameters (for more details, see Zhang 2016a). The
precise model parameters required for some voice qualities are at present speculative (and are
marked by an asterisk), but are based on articulatory and acoustic characteristics described in
previous studies, as described in Section 2.
Dimension Relevant model parameters
Approximation Glottal width
Voicing
Glottal width
Medial vocal fold thickness
Vocal fold stiffness
Subglottal pressure
Rate of vibration
To raise f0:
Greater vocal fold stiffness
Greater subglottal pressure
Smaller glottal width
Voice quality
(compared with modal)
Breathy voice:
Less vocal fold thickness
Larger glottal width
*Unconstricted creaky voice:
Less vocal fold thickness
Larger glottal width
Less vocal fold stiffness
Lower subglottal pressure
Constricted creaky voice qualities
(Prototypical creaky, tense voice, and vocal fry):
More vocal fold thickness
Smaller glottal width (Tense voice, vocal fry, and *prototypical creaky)
Less vocal fold stiffness (Vocal fry and *prototypical creaky)
Lower subglottal pressure (Vocal fry)
30

6 Summary of chapter and future work
This chapter provides an overview of the phonetics of voice, as defined narrowly by the activity
of the vocal folds. I took as starting points an assumption and descriptive fact regarding the voice:
first, I assume that the study of voice should be driven by what humans can hear (rather than
what we can do); and second, that the multidimensionality of voice production, acoustics, and
perception necessitates a unified model of the voice that is driven by what we can hear.
In this chapter I narrowed considerably the discussion of ‘what we can hear’ by focusing ex-
clusively on linguistic properties of the voice, especially those that we know matter for languages’
sound systems: vocal fold approximation, voicing, rate of vibration, and quality of voicing. A
combination of these four dimensions can be found in all sounds of language. Using the psychoa-
coustic model of the voice developed in Kreiman et al. 2014 and further (regarding the source
spectrum) in Garellek et al., 2016b, I showed how these linguistically-relevant vocal dimensions
can be modeled and measured. Recent articulatory modeling, such as that proposed by Zhang
(2015, 2016a), also bring us closer to understanding the links between voice production, acous-
tics, and voice quality perception. By demonstrating how different aspects of vocal fold dynamics
condition changes in the acoustic signal, articulatory and psychoacoustic models also enable us
to better understand sound changes involving the different voice dimensions of language (Kirby,
2013, 2014; Kuang & Liberman, 2015; Ratliff, 2015; Brunelle & Kirby, 2016).
The articulatory and psychoacoustic models of the voice reviewed here are continously being
refined and improved; as I mention at various points in this chapter, many of their assumptions,
and the predictions they make, have yet to be confirmed. Moreover, we are still far from knowing
the relevant articulatory and psychoacoustic properties of all linguistically-relevant aspects of the
voice. For example, Keating et al. (2015) discuss other subtypes of creaky voice not reviewed
here, and modeling the temporal characteristics of voice quality is still in its early stages, though
temporal properties of the voice are extremely important in language (Nellis & Hollenbach, 1980;
Silverman, 2003; DiCanio, 2009; Brunelle et al., 2010; Esposito & Khan, 2012; Garellek, 2012;
Remijsen, 2013; Yu, to appear). Much further work is also needed to determine how to integrate
other laryngeal and supralaryngeal structures with these models, since it is clear that the phonetics
of the voice and its linguistic patterns cannot be separated completely from the rest of the vocal
tract, or even the rest of the larynx (Edmondson & Esling, 2006; Moisik & Esling, 2011; Kuang,
31

2011; Brunelle, 2012; Story, 2012; Moisik & Esling, 2014; Samlan & Kreiman, 2014; Brunelle
& Kirby, 2016; Garellek et al., 2016a; Kuang & Cui, 2016; Carignan, 2017). For instance, recent
modeling work by Story (2012) involves synthesis of voice (including aspiration) and supralaryn-
geal articulations, along with their time-varying characteristics. Moreover, the role of the voice
in indexing non-phonological meaning is also an important component to linguistic studies of the
voice (Zhuang & Hasegawa-Johnson, 2008; Esling & Edmondson, 2010; Kreiman & Sidtis, 2011;
Mendoza-Denton, 2011; Callier, 2013; Zimman, 2013; Podesva & Callier, 2015; Starr, 2015; Park
et al., 2016). Finally, through interdisciplinary research we stand to learn much more about how
our voices convey different types of meaning.
32

References
Abramson, A. S., Tiede, M. K., & Luangthongkum, T. (2015). Voice register in Mon: acoustics
and electroglottography. Phonetica,72, 237–256.
Andruski, J. E. (2006). Tone clarity in mixed pitch/phonation-type tones. Journal of Phonetics,
34, 388–404.
Avelino, H. (2010). Acoustic and electroglottographic analyses of nonpathological, nonmodal
phonation. Journal of Voice,24, 270–280.
Baken, R. J., & Orlikoff, R. F. (2000). Clinical Measurement of Speech and Voice. San Diego:
Singular Publishing Group.
Batliner, A., Burger, S., Johne, B., & Kießling, A. (1993). M ¨
USLI: A classifcation scheme for
laryngealizations. In Proceedings of ESCA workshop on prosody (pp. 176–179). Lund.
Berkson, K. H. (2013). Phonation Types in Marathi: An Acoustic Investigation. Ph.D. thesis
University of Kansas.
Bickley, C. (1982). Acoustic analysis and perception of breathy vowels. MIT Speech Communica-
tion Working Papers,1, 71–81.
Bishop, J., & Keating, P. (2012). Perception of pitch location within a speaker’s range: Fundamen-
tal frequency, voice quality and speaker sex. Journal of the Acoustical Society of America,132,
1100–1112.
Blankenship, B. (2002). The timing of nonmodal phonation in vowels. Journal of Phonetics,30,
163–191.
Bonilha, H. S., Deliyski, D. D., Whiteside, J. P., & Gerlach, T. T. (2012). Vocal fold phase asym-
metries in patients with voice disorders: A study across visualization techniques. American
Journal of Speech-Language Pathology,21, 3–15.
Brunelle, M. (2010). The role of larynx height in the Javanese tense ∼lax stop contrast. In
R. Mercado, E. Potsdam, & L. d. Travis (Eds.), Austronesian and Theoretical Linguistics (pp.
7–24). Amsterdam: John Benjamins.
33

Brunelle, M. (2012). Dialect experience and perceptual integrality in phonological registers: Fun-
damental frequency, voice quality and the first formant in Cham. Journal of the Acoustical
Society of America,131, 3088–3102.
Brunelle, M., & Finkeldey, J. (2011). Tone perception in Sgaw Karen. In Proceedings of the 17th
International Congress of Phonetic Sciences (pp. 372–375).
Brunelle, M., & Kirby, J. (2016). Tone and phonation in Southeast Asian languages. Language
and Linguistics Compass,10, 191–207.
Brunelle, M., Nguyễn, D. D., & Nguyễn, K. H. (2010). A laryngographic and laryngoscopic study
of Northern Vietnamese tones. Phonetica,67, 147–169.
Callier, P. (2013). Linguistic context and the social meaning of voice quality variation. Ph.D. thesis
Georgetown University.
Campbell, N., & Beckman, M. (1997). Stress, prominence, and spectral tilt. In Intonation: Theory,
Models, and Applications (pp. 67–70). Athens, Greece: International Speech Communication
Association.
Carignan, C. (2017). Covariation of nasalization, tongue height, and breathiness in the realization
of F1 of Southern French nasal vowels. Journal of Phonetics,63, 87–105.
Chao, Y. R. (1930). A system of “tone-letters”. Le Maˆ
ıtre Phonetique,45, 24–27.
Chen, G., Kreiman, J., Gerratt, B. R., Neubauer, J., Shue, Y.-L., & Alwan, A. (2013). Development
of a glottal area index that integrates glottal gap size and open quotient. Journal of the Acoustical
Society of America,133, 1656–1666.
Choi, J.-Y., Hasegawa-Johnson, M., & Cole, J. (2005). Finding intonational boundaries using
acoustic cues related to the voice source. Journal of the Acoustical Society of America,118,
2579–2587.
Cumming, K. E., & Clements, M. A. (1995). Glottal models for digital speech processing: A
historical survey and new results. Digital Signal Processing,5, 21–42.
34

Davidson, L., & Erker, D. (2014). Hiatus resolution in American English: the case against glide
insertion. Language,90, 482–514.
de Krom, G. (1993). A cepstrum-based technique for determining harmonics-to-noise ratio in
speech signals. Journal of Speech and Hearing Research,36, 254–266.
de Krom, G. (1995). Some spectral correlates of pathological breathy and rough voice quality for
different types of vowel fragments. Journal of Speech and Hearing Research,38, 794–811.
Denes, P. B., & Pinson, E. N. (1993). The speech chain: The physics and biology of spoken
language. (2nd ed.). New York: W. H. Freeman.
Di Napoli, J. (2015). Glottalization at phrase boundaries in Tuscan and Roman Italian. In
J. Romero, & M. Riera (Eds.), The Phonetics–Phonology Interface: Representations and
methodologies. Amsterdam: John Benjamins.
DiCanio, C. (2014). Cue weight in the perception of Trique glottal consonants. Journal of the
Acoustical Society of America,119, 3059–3071.
DiCanio, C. T. (2009). The phonetics of register in Takhian Thong Chong. Journal of the Interna-
tional Phonetic Association,39, 162–188.
DiCanio, C. T. (2012). Coarticulation between tone and glottal consonants in Itunyoso Trique.
Journal of Phonetics,40, 162–176.
Dilley, L., Shattuck-Hufnagel, S., & Ostendorf, M. (1996). Glottalization of word-initial vowels
as a function of prosodic structure. Journal of Phonetics,24, 423–444.
Edmondson, J. A., Chang, Y., Hsieh, F., & Huang, H. J. (2011). Reinforcing voiceless finals in
Taiwanese and Hakka: Laryngoscopic case studies. In Proceedings of the 17th International
Congress of Phonetic Sciences. Hong Kong.
Edmondson, J. A., & Esling, J. H. (2006). The valves of the throat and their functioning in tone,
vocal register and stress: laryngoscopic case studies. Phonology,23, 157–191.
35

Edmondson, J. A., Esling, J. H., Harris, J. G., Li, S., & Ziwo, L. (2001). The aryepiglottic folds
and voice quality in the Yi and Bai languages: laryngoscopic case studies. Mon-Khmer Studies,
(pp. 83–100).
Epstein, M. A. (2002). Voice quality and prosody in English. Ph.D. thesis UCLA.
Esling, J. H., & Edmondson, J. A. (2010). Acoustical analysis of voice quality for sociophonetic
purposes. In M. D. Paolo, & M. Yaeger-Dror (Eds.), Sociophonetics: A student’s guide chap-
ter 11. London: Routledge.
Esling, J. H., & Harris, J. G. (2003). An expanded taxonomy of states of the glottis. In Proceedings
of the International Congress of Phonetic Science (pp. 1049–1052). Barcelona.
Esling, J. H., & Harris, J. G. (2005). States of the glottis: An articulatory phonetic model based
on laryngoscopic observations. In W. J. Hardcastle, & J. M. Beck (Eds.), A Figure of Speech: a
Festschrift for John Laver (pp. 347–383). Mahwah, NJ: Erlbaum.
Esposito, C. M. (2010a). The effects of linguistic experience on the perception of phonation.
Journal of Phonetics,38, 306–316.
Esposito, C. M. (2010b). Variation in contrastive phonation in Santa Ana Del Valle Zapotec.
Journal of the International Phonetic Association,40, 181–198.
Esposito, C. M. (2012). An acoustic and electroglottographic study of White Hmong phonation.
Journal of Phonetics,40, 466–476.
Esposito, C. M., & Khan, S. (2012). Contrastive breathiness across consonants and vowels: A
comparative study of Gujarati and White Hmong. Journal of the International Phonetic Associ-
ation,42, 123–143.
Flanagan, J. L. (1972). Speech Analysis, Synthesis, and Perception. Berlin: Springer.
Fulop, S. A., & Golston, C. (2008). Breathy and whispery voice in White Hmong. Proceedings of
meetings on acoustics,4, 060006.
Garellek, M. (2012). The timing and sequencing of coarticulated non-modal phonation in English
and White Hmong. Journal of Phonetics,40, 152–161.
36

Garellek, M. (2013). Production and perception of glottal stops. Ph.D. thesis UCLA.
Garellek, M. (2014). Voice quality strengthening and glottalization. Journal of Phonetics,45,
106–113.
Garellek, M. (2015). Perception of glottalization and phrase-final creak. Journal of the Acoustical
Society of America,137, 822–831.
Garellek, M., & Keating, P. (2011). The acoustic consequences of phonation and tone interactions
in Jalapa Mazatec. Journal of the International Phonetic Association,41, 185–205.
Garellek, M., Keating, P., Esposito, C. M., & Kreiman, J. (2013). Voice quality and tone identifi-
cation in White Hmong. Journal of the Acoustical Society of America,133, 1078–1089.
Garellek, M., Ritchart, A., & Kuang, J. (2016a). Breathy voice during nasality: a cross-linguistic
study. Journal of Phonetics,59, 110–121.
Garellek, M., Samlan, R., Gerratt, B. R., & Kreiman, J. (2016b). Modeling the voice source in
terms of spectral slopes. Journal of the Acoustical Society of America,139, 1404–1410.
Garellek, M., & Seyfarth, S. (2016). Acoustic differences between English /t/ glottalization and
phrasal creak. In Proceedings of Interspeech 2016 (pp. 1054–1058). San Francisco.
Garellek, M., & White, J. (2015). Phonetics of Tongan stress. Journal of the International Phonetic
Association,45, 13–34.
Gerfen, C., & Baker, K. (2005). The production and perception of laryngealized vowels in Coat-
zospan Mixtec. Journal of Phonetics,33, 311–334.
Gerratt, B. R., & Kreiman, J. (2001). Toward a taxonomy of nonmodal phonation. Journal of
Phonetics,29, 365–381.
Gick, B., Wilson, I., & Derrick, D. (2013). Articulatory Phonetics. Oxford: Wiley-Blackwell.
Gobl, C., & N´
ı Chasaide, A. (2010). Voice source variation and its communicative functions. In
W. J. Hardcastle, J. Laver, & F. E. Gibbon (Eds.), The Handbook of Phonetic Sciences chapter 11.
(pp. 378–423). Oxford: Wiley-Blackwell. (2nd ed.).
37

Gordon, M. (2014). Disentangling stress and pitch accent: Toward a typology of prominence at
different prosodic levels. In H. van der Hulst (Ed.), Word Stress: Theoretical and Typological
Issues (pp. 83–118). Oxford: Oxford University Press.
Gordon, M., & Applebaum, A. (2010). Acoustic correlates of stress in Turkish Kabardian. Journal
of the International Phonetic Association,40, 35–58.
Gordon, M., & Ladefoged, P. (2001). Phonation types: a cross-linguistic overview. Journal of
Phonetics,29, 383–406.
Hanson, H. M. (1995). Glottal characteristics of female speakers. Ph.D. thesis Harvard University.
Hanson, H. M. (1997). Glottal characteristics of female speakers: Acoustic correlates. Journal of
the Acoustical Society of America,101, 466–481.
Hanson, H. M., Stevens, K. N., Kuo, H.-K. J., Chen, M. Y., & Slifka, J. (2001). Towards models
of phonation. Journal of Phonetics,29, 451–480.
Harris, J. G. (1999). States of the glottis for voiceless plosives. In Proceedings of the International
Congress of Phonetic Sciences (pp. 2041–2044). San Francisco.
Hartl, D. M., Hans, S., Vessi`
ere, J., & Brasnu, D. F. (2003). Objective acoustic and aerodynamic
measures of breathiness in paralytic dysphonia. European Archives of Oto-Rhino-Laryngology,
260, 175–182.
Hillenbrand, J., Getty, L. A., Clark, M. J., & Wheeler, K. (1995). Acoustic characteristics of
American English vowels. Journal of the Acoustical Society of America,97, 3099–3111.
Hillenbrand, J. M., & Houde, R. A. (1996). Role of F0 and amplitude in the perception of glottal
stops. Journal of Speech and Hearing Research,39, 1182–1190.
Hirano, M., & Katika, Y. (1985). Cover-body theory of vocal fold vibration. In R. G. Daniloff
(Ed.), Speech science: recent advances (pp. 1–46). San Diego: College-Hill Press.
Hirose, H. (1997). Investigating the physiology of laryngeal structures. In W. J. Hardcastle, &
J. Laver (Eds.), The Handbook of Phonetic Sciences (pp. 116–136). Oxford: Blackwell.
38

Hirose, H. (2010). Investigating the physiology of laryngeal structures. In W. J. Hardcastle,
J. Laver, & F. E. Gibbon (Eds.), The Handbook of Phonetic Sciences (pp. 130–152). Oxford:
Blackwell. (2nd ed.).
Hirose, H., & Gay, T. (1972). The activity of the intrinsic laryngeal muscles in voicing control.
Phonetica,25, 140–164.
Hombert, J.-M., Ohala, J. J., & Ewan, W. G. (1979). Phonetic explanations for the development of
tones. Language,55, 37–58.
Honda, K., Hirai, H., Masaki, S., & Shimada, Y. (1999). Role of vertical larynx movement and
cervical lordosis in F0 control. Language and Speech,42, 401–411.
Iseli, M., Shue, Y.-L., & Alwan, A. (2007). Age, sex, and vowel dependencies of acoustic measures
related to the voice source. Journal of the Acoustical Society of America,121, 2283–2295.
Isogai, Y., Horiguchi, S., Honda, K., Aoki, Y., Hirose, H., & Saito, S. (1988). A dynamic simula-
tion model of vocal fold vibration. In O. Fujimura (Ed.), Vocal Physiology: Voice Production,
Mechanisms and Functions (pp. 191–206). New York: Raven Press.
Kane, J., Drugman, T., & Gobl, C. (2013). Improved automatic detection of creak. Computer
Speech and Language,27, 1028–1047.
Keating, P., Esposito, C., Garellek, M., Khan, S., & Kuang, J. (2011). Phonation contrasts across
languages. In Proceedings of the International Congress of Phonetic Sciences (pp. 1046–1049).
Hong Kong.
Keating, P., Garellek, M., & Kreiman, J. (2015). Acoustic properties of different kinds of creaky
voice. In Proceedings of the 18th International Congress of Phonetic Sciences. Glasgow.
Kempster, G. B., Gerratt, B. R., Abbott, K. V., Barkmeier-Kraemer, J., & Hillman, R. E. (2011).
Consensus auditory-perceptual evaluation of voice: Development of a standardized clinical pro-
tocol. American Journal of Speech-Language Pathology,18, 124–132.
Khan, S. (2012). The phonetics of contrastive phonation in Gujarati. Journal of Phonetics,40,
780–795.
39

Kingston, J. (2005). The phonetics of Athabaskan tonogenesis. In S. Hargus, & K. Rice (Eds.),
Athabaskan prosody (pp. 137–184). Amsterdam: John Benjamins.
Kirby, J. (2013). The role of probabilistic enhancement in phonologization. In A. C. L. Yu (Ed.),
Origins of sound change: approaches to phonologization (pp. 228–246). Oxford: Oxford Uni-
versity Press.
Kirby, J. P. (2014). Incipient tonogenesis in Phnom Penh Khmer: Acoustic and perceptual studies.
Journal of Phonetics,43, 69–85.
Klatt, D. H., & Klatt, L. C. (1990). Analysis, synthesis, and perception of voice quality variations
among female and male talkers. Journal of the Acoustical Society of America,87, 820–857.
Kreiman, J. (1982). Perception of sentence and paragraph boundaries in natural conversation.
Journal of Phonetics,10, 163–175.
Kreiman, J., Anto˜
nanzas-Barroso, N., & Gerratt, B. R. (2010a). Integrated software for analysis
and synthesis of voice quality. Behavior Research Methods,42, 1030–1041.
Kreiman, J., Anto˜
nanzas-Barroso, N., & Gerratt, B. R. (2016). The UCLA voice synthesizer,
version 2.0. Journal of the Acoustical Society of America,140, 2961.
Kreiman, J., Garellek, M., Chen, G., Alwan, A., & Gerratt, B. R. (2015). Perceptual evaluation of
voice source models. Journal of the Acoustical Society of America,138, 1–10.
Kreiman, J., Garellek, M., & Esposito, C. (2011). Perceptual importance of the voice source
spectrum from H2 to 2 kHz. Journal of the Acoustical Society of America,130, 2570.
Kreiman, J., Gerratt, B., & Anto˜
nanzas-Barroso, N. (2007). Measures of the glottal source spec-
trum. Journal of Speech, Language, and Hearing Research,50, 595–610.
Kreiman, J., & Gerratt, B. R. (2005). Perception of aperiodicity in pathological voice. Journal of
the Acoustical Society of America,117, 2201–2211.
Kreiman, J., & Gerratt, B. R. (2010). Perceptual sensitivity to first harmonic amplitude in the voice
source. Journal of the Acoustical Society of America,128, 2085–2089.
40

Kreiman, J., & Gerratt, B. R. (2012). Perceptual interaction of the harmonic source and noise in
voice. Journal of the Acoustical Society of America,131, 492–500.
Kreiman, J., Gerratt, B. R., Garellek, M., Samlan, R., & Zhang, Z. (2014). Toward a unified theory
of voice production and perception. Loquens,e009.
Kreiman, J., Gerratt, B. R., & Khan, S. (2010b). Effects of native language on perception of voice
quality. Journal of Phonetics,38, 588–593.
Kreiman, J., Shue, Y.-L., Chen, G., Iseli, M., Gerratt, B. R., Neubauer, J., & Alwan, A. (2012).
Variability in the relationships among voice quality, harmonic amplitudes, open quotient, and
glottal area waveform shape in sustained phonation. Journal of the Acoustical Society of Amer-
ica,132, 2625–2632.
Kreiman, J., & Sidtis, D. (2011). Foundations of Voice Studies. Oxford: Wiley-Blackwell.
Kuang, J. (2011). Production and perception of the phonation contrast in Yi. Master’s thesis
UCLA.
Kuang, J. (2012). Registers in tonal contrasts. UCLA Working Papers in Phonetics,110, 46–64.
Kuang, J. (2013a). Phonation in tonal contrasts. Ph.D. thesis UCLA.
Kuang, J. (2013b). The tonal space of contrastive five level tones. Phonetica,70, 1–23.
Kuang, J., & Cui, A. (2016). Relative cue weighting in perception and production of a sound
change in progress. Talk presented at LabPhon 2016, Cornell, NY.
Kuang, J., & Keating, P. (2014). Vocal fold vibratory patterns in tense versus lax phonation con-
trasts. Journal of the Acoustical Society of America,136, 2784–2797.
Kuang, J., & Liberman, M. (2015). Influence of spectral cues on the perception of pitch height. In
Proceedings of the 18th International Congress of Phonetic Sciences. Glasgow, UK.
Ladefoged, P. (1971). Preliminaries to linguistic phonetics. Chicago: University of Chicago.
Lancia, L., Voigt, D., & Krasovitskiy, G. (2016). Characterization of laryngealization as irregular
vocal fold vibration and interaction with prosodic prominence. Journal of Phonetics,54, 80–97.
41

Laver, J. (1980). The Phonetic Description of Voice Quality. Cambridge: Cambridge University
Press.
L¨
ofqvist, A., & McGowan, R. S. (1992). Influence of consonantal envelope on voice source aero-
dynamics. Journal of Phonetics,20, 93–110.
Lombardi, L. (1991). Laryngeal features and laryngeal neutralization. Ph.D. thesis University of
Massachusetts.
Maddieson, I., Flavier, S., Marsico, E., & Pellegrino, F. (2016). LAPSyD: Lyon-Albuquerque
Phonological Systems Databases, Version 1.0. http://www.lapsyd.ddl.ish-lyon.
cnrs.fr/lapsyd/. (Last checked 2017-07-22).
Mendoza-Denton, N. (2011). The semiotic hitchhiker’s guide to creaky voice: Circulation and
gendered hardcore in a Chicana/o gang persona. Journal of Linguistic Anthropology,21, 261–
280.
Miller, A. L. (2007). Guttural vowels and guttural co-articulation in Ju|’hoansi. Journal of Pho-
netics,35, 56–84.
Moisik, S. R. (2013). Harsh voice quality and its association with blackness in popular American
media. Phonetica,69, 193–215.
Moisik, S. R., & Esling, J. H. (2011). The ‘whole larynx’ approach to laryngeal features. In
Proceedings of the 17th International Congress of Phonetic Sciences. Hong Kong.
Moisik, S. R., & Esling, J. H. (2014). Modeling the biomechanical influence of epilaryngeal
stricture on the vocal folds: a low-dimensional model of vocal-ventricular fold coupling. Journal
of Speech, Language, and Hearing Research,57, S687–S704.
Moisik, S. R., Lin, H., & Esling, J. H. (2014). A study of laryngeal gestures in Mandarin cita-
tion tones using simultaneous laryngoscopy and laryngeal ultrasound (SLLUS). Journal of the
International Phonetic Association,44, 21–58.
42

Mooshammer, C. (2010). Acoustic and laryngographic measures of the laryngeal reflexes of lin-
guistic prominence and vocal effort in German. Journal of the Acoustical Society of America,
127, 1047–1058.
Munhall, K., & L¨
ofqvist, A. (1992). Gestural aggregation in speech-laryngeal gestures. Journal
of Phonetics,20, 111–126.
Nakatani, L. H., & Dukes, K. D. (1977). Locus of segmental cues for word juncture. Journal of
the Acoustical Society of America,62, 714–719.
Nellis, D. G., & Hollenbach, B. E. (1980). Fortis versus lenis in Cajonos Zapotec phonology.
International Journal of American Linguistics,46, 92–105.
N´
ı Chasaide, A., & Gobl, C. (1993). Contextual variations of the vowel voice source as a function
of adjacent consonants. Language and Speech,36, 303–330.
Ogden, R. (2001). Turn transition, creak and glottal stop in Finnish talk-in-interaction. Journal of
the International Phonetic Association,31, 139–152.
Park, S. J., Sigouin, C., Kreiman, J., Keating, P., Guo, J., Yeung, G., Kuo, F.-Y., & Alwan, A.
(2016). Speaker identity and voice quality: Modeling human responses and automatic speaker
recognition. In Proceedings of Interspeech 2016 (pp. 1044–1048). San Francisco.
Pierrehumbert, J., & Talkin, D. (1992). Lenition of /h/ and glottal stop. In G. J. Docherty, &
D. R. Ladd (Eds.), Papers in Laboratory Phonology II (pp. 90–117). Cambridge: Cambridge
University Press.
Podesva, R. J., & Callier, P. (2015). Voice quality and identity. Annual Review of Applied Linguis-
tics,35, 173–194.
Ratliff, M. (2015). Tonoexodus, tonogenesis, and tone change. In P. Honeybone, & J. Salmons
(Eds.), The Oxford Handbook of Historical Phonology (pp. 245–261). Oxford: Oxford Univer-
sity Press.
Redi, L., & Shattuck-Hufnagel, S. (2001). Variation in the realization of glottalization in normal
speakers. Journal of Phonetics,29, 407–429.
43

Reetz, H., & Jongman, A. (2008). Phonetics: transcription, production, acoustics, and perception.
Oxford: Wiley-Blackwell.
Remijsen, B. (2013). Tonal alignment is contrastive in falling contours in Dinka. Language,89,
297–327.
Rodgers, J. (1999). Three influences on glottalization in read and spontaneous German speech.
Arbeitsberichte des Instituts f¨
ur Phonetik und digitale Sprachverarbeitung der Universit¨
at Kiel,
34, 177–284.
Samlan, R. A., & Kreiman, J. (2014). Perceptual consequences of changes in epilaryngeal area
and shape. Journal of the Acoustical Society of America,136, 2798–2806.
Samlan, R. A., & Story, B. H. (2011). Relation of structural and vibratory kinematics of the vocal
folds to two acoustic measures of breathy voice based on computational modeling. Journal of
Speech, Language, and Hearing Research,54, 1267–1283.
Samlan, R. A., Story, B. H., & Bunton, K. (2013). Relation of perceived breathiness to laryn-
geal kinematics and acoustic measures based on computational modeling. Journal of Speech,
Language, and Hearing Research,56, 1209–1223.
Samlan, R. A., Story, B. H., Lotto, A. J., & Bunton, K. (2014). Acoustic and perceptual effects
of left–right laryngeal asymmetries based on computational modeling. Journal of Speech, Lan-
guage, and Hearing Research,57, 1619–1637.
Sapienza, C., Hicks, D. M., & Ruddy, B. H. (2011). Voice disorders. In N. B. Anderson, & G. H.
Shames (Eds.), Human Communication Disorders: An Introduction (pp. 202–237). Boston:
Pearson. (8th ed.).
Shrivastav, R., & Sapienza, C. M. (2006). Some difference limens for the perception of breathiness.
Journal of the Acoustical Society of America,120, 416–423.
Shue, Y.-L., Keating, P. A., Vicenik, C., & Yu, K. (2011). VoiceSauce: A program for voice
analysis. In Proceedings of the International Congress of Phonetic Sciences (pp. 1846–1849).
Hong Kong.
44

Silverman, D. (2003). Pitch discrimination between breathy vs. modal phonation. In J. Local,
R. Ogden, & R. Temple (Eds.), Phonetic Interpretation (Papers in Laboratory Phonology 6)
(pp. 293–304). Cambridge: Cambridge University Press.
Silverman, D., Blankenship, B., Kirk, P., & Ladefoged, P. (1995). Phonetic structures in Jalapa
Mazatec. Anthropological Linguistics,37, 70–88.
Simpson, A. (2012). The first and second harmonics should not be used to measure breathiness in
male and female voices. Journal of Phonetics,40, 477–490.
Slifka, J. (2000). Respiratory constraints on speech production at prosodic boundaries. Ph.D.
thesis MIT.
Slifka, J. (2003). Respiratory constraints on speech production: Starting an utterance. Journal of
the Acoustical Society of America,114, 3343–3353.
Slifka, J. (2006). Some physiological correlates to regular and irregular phonation at the end of an
utterance. Journal of Voice,20, 171–186.
Sluijter, A. M. C., & van Heuven, V. J. (1996). Spectral balance as an acoustic correlate of linguistic
stress. Journal of the Acoustical Society of America,100, 2471–2485.
Starr, R. L. (2015). Sweet voice: The role of voice quality in a Japanese feminine style. Language
in Society,44, 1–34.
Stevens, K. N. (2000). Acoustic Phonetics. Cambridge: MIT Press.
Story, B. H. (2012). Phrase-level speech simulation with an airway modulation model of speech
production. Computer Speech and Language,27, 989–1010.
Story, B. H. (2015). Mechanisms of voice production. In M. A. Redford (Ed.), Handbook of
Speech Production (pp. 34–58). Oxford: Wiley-Blackwell.
Stross, B. (2013). Falsetto voice and observational logic: Motivational meanings. Language in
Society,42, 139–162.
45

Sundberg, J. (1987). The Science of the Singing Voice. DeKalb, IL: Northern Illinois University
Press.
Titze, I. R. (1992). Phonation threshold pressure: A missing link in glottal aerodynamics. Journal
of the Acoustical Society of America,91, 2926–2935.
Titze, I. R. (1994). Principles of Voice Production. Engelwood Cliffs, NJ: Prentice Hall Inc.
Titze, I. R. (2006). The Myoelastic Aerodynamic Theory of Phonation. Iowa City, IA: National
Centre for Voice and Speech.
Titze, I. R., Schmidt, S. S., & Titze, M. R. (1995). Phonation threshold pressure in a physical
model of the vocal fold mucosa. Journal of the Acoustical Society of America,97, 3080–3084.
Traill, A. (1985). Phonetic and Phonological Studies of the !X ´
o˜
o Bushmen. Hamburg: Buske.
Umeda, N. (1978). Occurrence of glottal stops in fluent speech. Journal of the Acoustical Society
of America,64, 88–94.
van den Berg, J. (1958). Myoelastic aerodynamic theory of voice production. Journal of Speech
and Hearing Research,1, 227–244.
van den Berg, J., & Tan, T. S. (1959). Results of experiments with human larynxes. Practica
Oto-Rhino-Laryngologica,21, 425–450.
Vicenik, C. (2010). An acoustic study of Georgian stop consonants. Journal of the International
Phonetic Association,40, 59–92.
Wang, C. (2015). Multi-register Tone Systems and Their Evolution on the Jianghan Plain. Ph.D.
thesis Hong Kong University of Science and Technology.
Wang, C.-Y., & Tang, C.-J. (2012). The falsetto tones of the dialects in Hubei province. In
Proceedings of the 6th International Conference on Speech Prosody.
Wayland, R., & Jongman, A. (2003). Acoustic correlates of breathy and clear vowels: The case of
Khmer. Journal of Phonetics,31, 181–201.
46

Yanushevskaya, I., Murphy, A., Gobl, C., & N´
ı Chasaide, A. (2016). Perceptual salience of voice
source parameters in signaling focal prominence. In Proceedings of Interspeech 2016 (pp. 3161–
3165). San Francisco.
Yanushevskaya, I., N´
ı Chasaide, A., & Gobl, C. (2011). Universal and language-specific perception
of affect from voice. In Proceedings of the 17th International Congress of Phonetic Sciences
(pp. 2208–2211). Hong Kong.
Yin, J., & Zhang, Z. (2013). The influence of thyroarytenoid and cricothyroid muscle activation
on vocal fold stiffness and eigenfrequencies. Journal of the Acoustical Society of America,133,
2972–2983.
Yu, K. M. (to appear). The role of time in phonetic spaces: temporal resolution in Cantonese tone
perception. Journal of Phonetics, .
Yu, K. M., & Lam, H. W. (2014). The role of creaky voice in Cantonese tonal perception. Journal
of the Acoustical Society of America,136, 1320–1333.
Zhang, J., & Yan, H. (2015). Contextually dependent cue weighting for a laryngeal contrast in
Shanghai Wu. In Proceedings of the International Congress of Phonetic Sciences (pp. 1–5).
Glasgow, UK.
Zhang, Z. (2015). Regulation of glottal closure and airflow in a three-dimensional phonation
model: Implications for vocal intensity control. Journal of the Acoustical Society of America,
137, 898–910.
Zhang, Z. (2016a). Cause-effect relationship between vocal fold physiology and voice production
in a three-dimensional phonation model. Journal of the Acoustical Society of America,139,
1493–1507.
Zhang, Z. (2016b). Mechanics of human voice production and control. Journal of the Acoustical
Society of America,140, 2614–2635.
Zhang, Z., Kreiman, J., Gerratt, B. R., & Garellek, M. (2013). Acoustic and perceptual effects of
changes in body layer stiffness in symmetric and asymmetric vocal fold models. Journal of the
Acoustical Society of America,133, 453–462.
47

Zhuang, X., & Hasegawa-Johnson, M. (2008). Towards interpretation of creakiness in Switch-
board. In Proceedings of Speech Prosody (pp. 37–40).
Zimman, L. (2013). Hegemonic masculinity and the variability of gay-sounding speech: The
perceived sexuality of transgender men. Journal of Language & Sexuality,2, 1–39.
48