Speech Perception
Media
Part of Speech Perception
- Title
- Speech Perception
- extracted text
-
Speech Perception
ATHENA VOULOUMANOS
Abstract
Speech perception is the process by which listeners presented with a distribution of
audible frequencies modulated in amplitude (loudness) and spectral (the frequency
set) content across time turn this sound into a coherent unit of perception that is
interpreted as language. Classic studies established that speech is not perceived by
simply mapping sets of invariant acoustic properties onto different speech sounds. In
fact, speech perception is robust even when the acoustic signal has been dramatically
distorted. Current approaches focus on understanding how we perceive speech by
investigating the neural basis of processing different physical aspects of the speech
signal, the encoding of acoustic information in the speech signal at different time
scales, the developmental of speech perception, and the multimodal representation
of speech. Understanding how humans perceive speech will require the expertise of
psychologists, neuroscientists, and engineers.
INTRODUCTION
Billions of humans use spoken language as a means of communication;
speech perception is a critical, effortless daily activity. Speech is a sound
consisting of a distribution of audible frequencies modulated in amplitude
(akin to loudness) and spectral content (the set of frequencies) across time.
Speech perception is the process by which listeners turn these modulated
audible frequencies into a coherent unit of perception (or percept) that can
be interpreted as language.
FOUNDATIONAL RESEARCH
Two fundamental questions guided early speech perception research. What
do we perceive when we perceive speech? And how do we perceive it?
The answer to these questions originally seemed obvious: speech sounds
appeared to be characterized by specific acoustic properties (which needed to
be discovered), and perceived by mapping these unique acoustic properties
onto each discrete speech sound. But these assumptions were challenged by
Emerging Trends in the Social and Behavioral Sciences. Edited by Robert Scott and Stephen Kosslyn.
© 2015 John Wiley & Sons, Inc. ISBN 978-1-118-90077-2.
1
2
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
the early discovery of four remarkable properties of speech perception: the
lack of acoustic invariance, categorical perception, the lack of segmentability,
and the simultaneous perception of speech and nonspeech. Each of these
points is considered in what follows.
Initial attempts to identify the set of invariant (unchanging) acoustic
properties that allow us to perceive distinct speech sounds, for example,
to identify the specific acoustic properties that characterize a “d” sound,
revealed a major complication. No unique combination of acoustic properties could specify two different “d”s as being one and the same sound
in different speech contexts. Specifically, the acoustic properties of the
“transitions” (rapid frequency changes) that help specify the two “d”
sounds in the syllables “da” and “di,” even syllables spoken by the same
speaker, were completely different. In fact, a given speech sound was
found to be acoustically different depending on the specific speech sounds
that come before and after it, so-called coarticulatory cues. Speech was
discovered to lack acoustic invariance—coarticulation between neighboring
speech sounds gave rise to too many acoustic differences and not enough
similarities between the same speech sound in different speech contexts to
identify a given speech sound based on unique (simple) acoustic properties.
This surprising finding led some to argue that rather than perceiving speech
directly through acoustic properties, speech perception is grounded in the
(less variable) articulatory gestures that generated the speech signal. But
invariant gestures for different speech sounds also proved elusive.
One of the ways people were proposed to adjust for the variability of
speech sounds is through categorical perception. Categorical perception is
the process that allows us to experience a continuously varying property as
two or more discrete categories. For instance, although one of the properties
that differentiates a “da” from a “ta”—voicing—varies continuously, listeners either perceive a “da” or a “ta” sound, never a sound in between the two.
This process allows us to ignore differences within a category boundary and
focus on differences between category boundaries. Perceiving discrete types
of things rather than continuously variable things makes it computationally
easier to think about and act on the world. Moreover, the ability to perceive
speech sounds (especially consonants) categorically was thought to be
uniquely human. But it was quickly shown that categorical perception was
not unique to humans when, in rapid succession, chinchillas, budgerigars,
and rhesus monkeys were shown to perceive human speech sounds categorically. Moreover, recent work has cast doubt on whether humans do, in
fact, ignore within-category variation to the degree previously believed.
At the same time, even a simple consonant-vowel sound such as “da” cannot be divided into its component consonant and vowel segments. No matter
how the syllable is split, it either sounds similar to the full consonant-vowel
Speech Perception
3
syllable or it sounds very like a nonspeech chirp or buzz, suggesting that
syllables, rather than segments, might be more natural percepts. Counter to
intuition, the signal does not have acoustic properties that mark boundaries
between segments that we perceive as different (such as a vowel and a consonant), which is to say that speech lacks “segmentability”—it lacks clearly
marked boundaries between what we hear as distinct speech sounds. (This
problem persists at higher levels of perception. Words also lack clear silences
to mark boundaries between words, in contrast to the blank spaces that mark
word boundaries in written language.)
A phenomenon known as duplex perception showed that a sound could
be simultaneously perceived as speech and a nonspeech, by isolating and
manipulating portions of the sound through a dichotic listening task in
which different parts of the sound are presented to each ear. For example,
if part of the transition (rapid frequency change) that differentiates a “ga”
from a “da” is removed from the syllable and presented to one ear, it sounds
similar to a chirp. If, at the same time, the rest of the syllable is presented to
the other ear, the listener simultaneously hears the nonspeech chirp and the
speech syllable. The ability to perceive a single sound simultaneously as both
speech and nonspeech suggests there may be two distinct and differentiable
ways of perceiving a sound: auditory—the way in which most sounds are
perceived—and phonetic—a special way in which only speech sounds are
perceived. However, later studies showed that duplex perception is also
evident for nonspeech, with a clever study using the sound of slamming
doors in which an analogous dichotic manipulation can induce listeners to
simultaneously perceive a wooden door and a metal door.
Dichotic listening tasks also revealed that people were better at recognizing speech sounds when speech was presented to their right ear than their
left, the so-called right-ear advantage. Because of the neural wiring of the
auditory system, the right ear forms stronger connections to the left hemisphere of the brain, and this right-ear speech perception advantage was taken
as evidence for specialized language processing in the left hemisphere. This
behavioral work complemented the classic neurological work of Broca and
Wernicke, which showed that focal lesions in the left hemisphere caused
aphasias, disorders of speech and language.
These early discoveries highlight the fact that speech perception consists of
more than perceiving and grouping sets of unique acoustic properties into
discrete speech percepts. They fed into major theories of speech perception,
which differ on several major questions that inform current debates. (i) What
are the units of speech perception, acoustic or gestural, segmental or syllabic? (ii) Is speech perception a specialized mode of perception with unique
processes and representations, or is speech perceived as just another environmental sound subject to general auditory processes and mapping onto
4
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
general auditory representations? (iii) Does speech perception rely on specialized neural circuitry, or general-purpose neural substrates?
CUTTING-EDGE RESEARCH
Current approaches address major questions in speech perception by focusing on the nature of the speech signal as broadband frequencies subject to
amplitude and spectral modulations. One current idea about how we turn
time-varying frequencies into speech percepts is that modulations at different time scales reflect different aspects of the speech signal and are analyzed
separately, using different areas of the brain. The rapid modulations that are
important for perceiving segments such as consonants happen over short
time scales (25–50 ms) that are processed in the left hemisphere. The slower
modulations that contribute to the perception of syllables happen over longer
time scales (200–300 ms) that are processed in the right hemisphere. This
hemispheric division of labor of information encoded at different time scales
might even be present in humans from birth. This may provide one answer
to how we perceive the linguistic segments of speech: Information at different time scales in speech may map onto neural circuitry that can decode
information at these different time scales.
One of the remarkable things about speech perception is how very good
humans are at it. Adults can be induced to perceive a speech sound as speech
even if we replace the smear of frequencies in the signal with three simple
sine waves that track the main frequency changes in time, even if we rotate
the sound around 2000 Hz so that all the energy that is usually at high frequencies is now at low frequencies and vice versa, even if we reverse the
sound waves every 50 ms, and even if we band-pass filter the speech signal,
removing much of its spectral information including formant (concentrations of energy around a particular frequency) transitions. Humans show a
remarkable ability to perceive intelligible speech even in severely degraded
signals. Human infants might also be able to perceive atypical speech as
speech when other visual cues, such as the presence of a static human face,
provide a supporting context. Examining how humans perceive a sound as
intelligible speech, and the way intelligibility may be supported by neural
responses reflecting the amplitude and spectral modulations in speech is a
topic of current study.
Some important speech perception biases are present from birth. Even
neonates discriminate between and prefer speech to many nonspeech
sounds including backwards speech (which has scrambled temporal properties that make it unintelligible), and synthetic sounds that mimic some
properties of speech by using sinusoidal waves to track the main spectral
and temporal changes across time of natural speech sounds. This bias
Speech Perception
5
for speech is somewhat broadly tuned, as newborns appear to listen to
other vocalizations such as rhesus monkey calls as much as speech. By 3
months, human infants prefer listening to speech, even speech in a foreign
language they have not previously heard, to many other sounds including
environmental sounds (running water, bells), rhesus monkey calls, and
even human emotional vocalizations such as laughter. Important questions
remain on what specific properties allow infants to recognize a sound as
speech and prefer it to other sounds.
Although we think of speech as primarily an auditory medium, speech perception is inherently multimodal. There is a significant visual component to
speech perception. Not only can adults and infants match vowels, gender,
and affect in voices and faces but classic studies also showed that when adults
and infants hear a voice producing “ba” and see a silent face producing “ga,”
they integrate the information to perceive a sound that is never actually specified by either stimulus, a “da.” Visual information alone is sufficient to allow
infants and adults to detect when a bilingual speaker switches to a different
language. But vision is not the only integrated sense. By applying a puff of air
to irrelevant parts of the body such as the ankle, adults perceive a “pa” sound
where previously they had perceived a “ba” because the puff of air provides
tactile information consistent with aspiration (an expulsion or air) produced
during a “pa” sound. Even infants’ perception of speech is affected by their
own lip movements. By sucking on differently shaped objects, infants produced lip movements consistent with “ee”-like lip spreading or “oo”-like lip
rounding, and these lip movements affected their perception of those vowels.
Speech processing is inherently multimodal from early in infancy. The multimodal representation of speech including visual, tactile, and somatosensory
information is a key area of current study.
Many of these approaches tackle how we perceive speech by building up
from acoustic properties of signals, but another approach examines the influences of higher level information from words, syntax, semantics, and cognition on how we perceive individual speech sounds. When a portion of a
speech sound is replaced by a nonspeech sound such as a cough, listeners
are very poor at detecting which speech sound has been replaced, suggesting that they were able to automatically “restore” the missing speech portion.
At the same time, when a particular speech segment is made ambiguous, say,
by being halfway between a “b” and a “p,” listeners perceive it unambiguously as one or the other depending on whether it forms a real word (“pork”
rather than “bork”), or whether the sentence makes sense with one rather
than the other (“I ate a pear for dessert”). These observations suggest that
speech perception is not based entirely on the acoustic information in the signal but is instead modulated by the higher order linguistic context. Speech
perception is a dynamic process that allows us to perceive, segment, and
6
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
recognize words from different speakers and in different contexts. Current
research examines whether the mechanisms that support speech perception
in simple perceptual tasks contribute to our understanding of dynamic spoken language.
KEY ISSUES FOR FUTURE RESEARCH
Speech perception has been studied intensively since the 1930s. But key
aspects of how we perceive speech are incompletely understood: What do we
perceive when we perceive speech? Do we perceive speech differently than
we perceive other sounds? How is speech perception instantiated in neural
circuitry and do physical aspects of the auditory signal map neatly onto
distinct neural processes or regions? Is the perception of speech different in
humans than in other animals? Advances in understanding human speech
perception will likely require the collaboration of psychologists specializing
in perception, neuroscientists specializing in functional neuroanatomy,
and engineers specializing in digital signal processing. The way in which
humans perceive speech is still very much a matter of debate.
FURTHER READING
Blumstein, S. E., & Stevens, K. N. (1981). Phonetic features and acoustic invariance
in speech. Cognition, 10(1–3), 25–32.
Giraud, A. L., & Poeppel, D. (2012). Cortical oscillations and speech processing:
Emerging computational principles and operations. Nature Neuroscience, 15(4),
511–7.
Liberman, A., & Whalen, D. (2000). On the relation of speech to language. Trends in
Cognitive Sciences, 4(5), 187–196.
McGettigan, C., & Scott, S. K. (2012). Cortical asymmetries in speech perception:
What’s wrong, what’s right and what’s left? Trends in Cognitive Sciences, 16(5),
269–276.
Samuel, A. G. (2011). Speech perception. Annual Review of Psychology, 62, 49–72.
Vouloumanos, A., & Werker, J. F. (2007). Listening to language at birth: Evidence for
a bias for speech in neonates. Developmental Science, 10(2), 159–164.
ATHENA VOULOUMANOS SHORT BIOGRAPHY
Athena Vouloumanos is a Professor of Psychology at New York University
where she directs the NYU Infant Cognition and Communication Lab.
Vouloumanos’s research addresses fundamental questions on speech perception, language acquisition, and the development of communication.
With funding from the Natural Sciences and Engineering Research Council
Speech Perception
7
of Canada, the Fonds québécois de recherche sur la société et la culture,
and the National Institutes of Health, Vouloumanos has been exploring
the linguistic and cognitive abilities of adults and young infants, including
newborns and infants at high risk for autism spectrum disorder. Her findings
have been published in journals such as Science, Cognition, Cognitive Science,
Child Development, Developmental Science, and the Proceedings of the National
Academy of Sciences.
Personal webpage: http://www.psych.nyu.edu/vouloumanos/
Laboratory webpage: http://www.psych.nyu.edu/niccl/
RELATED ESSAYS
Mental Models (Psychology), Ruth M. J. Byrne
Misinformation and How to Correct It (Psychology), John Cook et al.
Construal Level Theory and Regulatory Scope (Psychology), Alison Ledgerwood et al.
Resource Limitations in Visual Cognition (Psychology), Brandon M. Liverence
and Steven L. Franconeri
Neural and Cognitive Plasticity (Psychology), Eduardo Mercado III
Attention and Perception (Psychology), Ronald A. Rensink
-
Speech Perception
ATHENA VOULOUMANOS
Abstract
Speech perception is the process by which listeners presented with a distribution of
audible frequencies modulated in amplitude (loudness) and spectral (the frequency
set) content across time turn this sound into a coherent unit of perception that is
interpreted as language. Classic studies established that speech is not perceived by
simply mapping sets of invariant acoustic properties onto different speech sounds. In
fact, speech perception is robust even when the acoustic signal has been dramatically
distorted. Current approaches focus on understanding how we perceive speech by
investigating the neural basis of processing different physical aspects of the speech
signal, the encoding of acoustic information in the speech signal at different time
scales, the developmental of speech perception, and the multimodal representation
of speech. Understanding how humans perceive speech will require the expertise of
psychologists, neuroscientists, and engineers.
INTRODUCTION
Billions of humans use spoken language as a means of communication;
speech perception is a critical, effortless daily activity. Speech is a sound
consisting of a distribution of audible frequencies modulated in amplitude
(akin to loudness) and spectral content (the set of frequencies) across time.
Speech perception is the process by which listeners turn these modulated
audible frequencies into a coherent unit of perception (or percept) that can
be interpreted as language.
FOUNDATIONAL RESEARCH
Two fundamental questions guided early speech perception research. What
do we perceive when we perceive speech? And how do we perceive it?
The answer to these questions originally seemed obvious: speech sounds
appeared to be characterized by specific acoustic properties (which needed to
be discovered), and perceived by mapping these unique acoustic properties
onto each discrete speech sound. But these assumptions were challenged by
Emerging Trends in the Social and Behavioral Sciences. Edited by Robert Scott and Stephen Kosslyn.
© 2015 John Wiley & Sons, Inc. ISBN 978-1-118-90077-2.
1
2
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
the early discovery of four remarkable properties of speech perception: the
lack of acoustic invariance, categorical perception, the lack of segmentability,
and the simultaneous perception of speech and nonspeech. Each of these
points is considered in what follows.
Initial attempts to identify the set of invariant (unchanging) acoustic
properties that allow us to perceive distinct speech sounds, for example,
to identify the specific acoustic properties that characterize a “d” sound,
revealed a major complication. No unique combination of acoustic properties could specify two different “d”s as being one and the same sound
in different speech contexts. Specifically, the acoustic properties of the
“transitions” (rapid frequency changes) that help specify the two “d”
sounds in the syllables “da” and “di,” even syllables spoken by the same
speaker, were completely different. In fact, a given speech sound was
found to be acoustically different depending on the specific speech sounds
that come before and after it, so-called coarticulatory cues. Speech was
discovered to lack acoustic invariance—coarticulation between neighboring
speech sounds gave rise to too many acoustic differences and not enough
similarities between the same speech sound in different speech contexts to
identify a given speech sound based on unique (simple) acoustic properties.
This surprising finding led some to argue that rather than perceiving speech
directly through acoustic properties, speech perception is grounded in the
(less variable) articulatory gestures that generated the speech signal. But
invariant gestures for different speech sounds also proved elusive.
One of the ways people were proposed to adjust for the variability of
speech sounds is through categorical perception. Categorical perception is
the process that allows us to experience a continuously varying property as
two or more discrete categories. For instance, although one of the properties
that differentiates a “da” from a “ta”—voicing—varies continuously, listeners either perceive a “da” or a “ta” sound, never a sound in between the two.
This process allows us to ignore differences within a category boundary and
focus on differences between category boundaries. Perceiving discrete types
of things rather than continuously variable things makes it computationally
easier to think about and act on the world. Moreover, the ability to perceive
speech sounds (especially consonants) categorically was thought to be
uniquely human. But it was quickly shown that categorical perception was
not unique to humans when, in rapid succession, chinchillas, budgerigars,
and rhesus monkeys were shown to perceive human speech sounds categorically. Moreover, recent work has cast doubt on whether humans do, in
fact, ignore within-category variation to the degree previously believed.
At the same time, even a simple consonant-vowel sound such as “da” cannot be divided into its component consonant and vowel segments. No matter
how the syllable is split, it either sounds similar to the full consonant-vowel
Speech Perception
3
syllable or it sounds very like a nonspeech chirp or buzz, suggesting that
syllables, rather than segments, might be more natural percepts. Counter to
intuition, the signal does not have acoustic properties that mark boundaries
between segments that we perceive as different (such as a vowel and a consonant), which is to say that speech lacks “segmentability”—it lacks clearly
marked boundaries between what we hear as distinct speech sounds. (This
problem persists at higher levels of perception. Words also lack clear silences
to mark boundaries between words, in contrast to the blank spaces that mark
word boundaries in written language.)
A phenomenon known as duplex perception showed that a sound could
be simultaneously perceived as speech and a nonspeech, by isolating and
manipulating portions of the sound through a dichotic listening task in
which different parts of the sound are presented to each ear. For example,
if part of the transition (rapid frequency change) that differentiates a “ga”
from a “da” is removed from the syllable and presented to one ear, it sounds
similar to a chirp. If, at the same time, the rest of the syllable is presented to
the other ear, the listener simultaneously hears the nonspeech chirp and the
speech syllable. The ability to perceive a single sound simultaneously as both
speech and nonspeech suggests there may be two distinct and differentiable
ways of perceiving a sound: auditory—the way in which most sounds are
perceived—and phonetic—a special way in which only speech sounds are
perceived. However, later studies showed that duplex perception is also
evident for nonspeech, with a clever study using the sound of slamming
doors in which an analogous dichotic manipulation can induce listeners to
simultaneously perceive a wooden door and a metal door.
Dichotic listening tasks also revealed that people were better at recognizing speech sounds when speech was presented to their right ear than their
left, the so-called right-ear advantage. Because of the neural wiring of the
auditory system, the right ear forms stronger connections to the left hemisphere of the brain, and this right-ear speech perception advantage was taken
as evidence for specialized language processing in the left hemisphere. This
behavioral work complemented the classic neurological work of Broca and
Wernicke, which showed that focal lesions in the left hemisphere caused
aphasias, disorders of speech and language.
These early discoveries highlight the fact that speech perception consists of
more than perceiving and grouping sets of unique acoustic properties into
discrete speech percepts. They fed into major theories of speech perception,
which differ on several major questions that inform current debates. (i) What
are the units of speech perception, acoustic or gestural, segmental or syllabic? (ii) Is speech perception a specialized mode of perception with unique
processes and representations, or is speech perceived as just another environmental sound subject to general auditory processes and mapping onto
4
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
general auditory representations? (iii) Does speech perception rely on specialized neural circuitry, or general-purpose neural substrates?
CUTTING-EDGE RESEARCH
Current approaches address major questions in speech perception by focusing on the nature of the speech signal as broadband frequencies subject to
amplitude and spectral modulations. One current idea about how we turn
time-varying frequencies into speech percepts is that modulations at different time scales reflect different aspects of the speech signal and are analyzed
separately, using different areas of the brain. The rapid modulations that are
important for perceiving segments such as consonants happen over short
time scales (25–50 ms) that are processed in the left hemisphere. The slower
modulations that contribute to the perception of syllables happen over longer
time scales (200–300 ms) that are processed in the right hemisphere. This
hemispheric division of labor of information encoded at different time scales
might even be present in humans from birth. This may provide one answer
to how we perceive the linguistic segments of speech: Information at different time scales in speech may map onto neural circuitry that can decode
information at these different time scales.
One of the remarkable things about speech perception is how very good
humans are at it. Adults can be induced to perceive a speech sound as speech
even if we replace the smear of frequencies in the signal with three simple
sine waves that track the main frequency changes in time, even if we rotate
the sound around 2000 Hz so that all the energy that is usually at high frequencies is now at low frequencies and vice versa, even if we reverse the
sound waves every 50 ms, and even if we band-pass filter the speech signal,
removing much of its spectral information including formant (concentrations of energy around a particular frequency) transitions. Humans show a
remarkable ability to perceive intelligible speech even in severely degraded
signals. Human infants might also be able to perceive atypical speech as
speech when other visual cues, such as the presence of a static human face,
provide a supporting context. Examining how humans perceive a sound as
intelligible speech, and the way intelligibility may be supported by neural
responses reflecting the amplitude and spectral modulations in speech is a
topic of current study.
Some important speech perception biases are present from birth. Even
neonates discriminate between and prefer speech to many nonspeech
sounds including backwards speech (which has scrambled temporal properties that make it unintelligible), and synthetic sounds that mimic some
properties of speech by using sinusoidal waves to track the main spectral
and temporal changes across time of natural speech sounds. This bias
Speech Perception
5
for speech is somewhat broadly tuned, as newborns appear to listen to
other vocalizations such as rhesus monkey calls as much as speech. By 3
months, human infants prefer listening to speech, even speech in a foreign
language they have not previously heard, to many other sounds including
environmental sounds (running water, bells), rhesus monkey calls, and
even human emotional vocalizations such as laughter. Important questions
remain on what specific properties allow infants to recognize a sound as
speech and prefer it to other sounds.
Although we think of speech as primarily an auditory medium, speech perception is inherently multimodal. There is a significant visual component to
speech perception. Not only can adults and infants match vowels, gender,
and affect in voices and faces but classic studies also showed that when adults
and infants hear a voice producing “ba” and see a silent face producing “ga,”
they integrate the information to perceive a sound that is never actually specified by either stimulus, a “da.” Visual information alone is sufficient to allow
infants and adults to detect when a bilingual speaker switches to a different
language. But vision is not the only integrated sense. By applying a puff of air
to irrelevant parts of the body such as the ankle, adults perceive a “pa” sound
where previously they had perceived a “ba” because the puff of air provides
tactile information consistent with aspiration (an expulsion or air) produced
during a “pa” sound. Even infants’ perception of speech is affected by their
own lip movements. By sucking on differently shaped objects, infants produced lip movements consistent with “ee”-like lip spreading or “oo”-like lip
rounding, and these lip movements affected their perception of those vowels.
Speech processing is inherently multimodal from early in infancy. The multimodal representation of speech including visual, tactile, and somatosensory
information is a key area of current study.
Many of these approaches tackle how we perceive speech by building up
from acoustic properties of signals, but another approach examines the influences of higher level information from words, syntax, semantics, and cognition on how we perceive individual speech sounds. When a portion of a
speech sound is replaced by a nonspeech sound such as a cough, listeners
are very poor at detecting which speech sound has been replaced, suggesting that they were able to automatically “restore” the missing speech portion.
At the same time, when a particular speech segment is made ambiguous, say,
by being halfway between a “b” and a “p,” listeners perceive it unambiguously as one or the other depending on whether it forms a real word (“pork”
rather than “bork”), or whether the sentence makes sense with one rather
than the other (“I ate a pear for dessert”). These observations suggest that
speech perception is not based entirely on the acoustic information in the signal but is instead modulated by the higher order linguistic context. Speech
perception is a dynamic process that allows us to perceive, segment, and
6
EMERGING TRENDS IN THE SOCIAL AND BEHAVIORAL SCIENCES
recognize words from different speakers and in different contexts. Current
research examines whether the mechanisms that support speech perception
in simple perceptual tasks contribute to our understanding of dynamic spoken language.
KEY ISSUES FOR FUTURE RESEARCH
Speech perception has been studied intensively since the 1930s. But key
aspects of how we perceive speech are incompletely understood: What do we
perceive when we perceive speech? Do we perceive speech differently than
we perceive other sounds? How is speech perception instantiated in neural
circuitry and do physical aspects of the auditory signal map neatly onto
distinct neural processes or regions? Is the perception of speech different in
humans than in other animals? Advances in understanding human speech
perception will likely require the collaboration of psychologists specializing
in perception, neuroscientists specializing in functional neuroanatomy,
and engineers specializing in digital signal processing. The way in which
humans perceive speech is still very much a matter of debate.
FURTHER READING
Blumstein, S. E., & Stevens, K. N. (1981). Phonetic features and acoustic invariance
in speech. Cognition, 10(1–3), 25–32.
Giraud, A. L., & Poeppel, D. (2012). Cortical oscillations and speech processing:
Emerging computational principles and operations. Nature Neuroscience, 15(4),
511–7.
Liberman, A., & Whalen, D. (2000). On the relation of speech to language. Trends in
Cognitive Sciences, 4(5), 187–196.
McGettigan, C., & Scott, S. K. (2012). Cortical asymmetries in speech perception:
What’s wrong, what’s right and what’s left? Trends in Cognitive Sciences, 16(5),
269–276.
Samuel, A. G. (2011). Speech perception. Annual Review of Psychology, 62, 49–72.
Vouloumanos, A., & Werker, J. F. (2007). Listening to language at birth: Evidence for
a bias for speech in neonates. Developmental Science, 10(2), 159–164.
ATHENA VOULOUMANOS SHORT BIOGRAPHY
Athena Vouloumanos is a Professor of Psychology at New York University
where she directs the NYU Infant Cognition and Communication Lab.
Vouloumanos’s research addresses fundamental questions on speech perception, language acquisition, and the development of communication.
With funding from the Natural Sciences and Engineering Research Council
Speech Perception
7
of Canada, the Fonds québécois de recherche sur la société et la culture,
and the National Institutes of Health, Vouloumanos has been exploring
the linguistic and cognitive abilities of adults and young infants, including
newborns and infants at high risk for autism spectrum disorder. Her findings
have been published in journals such as Science, Cognition, Cognitive Science,
Child Development, Developmental Science, and the Proceedings of the National
Academy of Sciences.
Personal webpage: http://www.psych.nyu.edu/vouloumanos/
Laboratory webpage: http://www.psych.nyu.edu/niccl/
RELATED ESSAYS
Mental Models (Psychology), Ruth M. J. Byrne
Misinformation and How to Correct It (Psychology), John Cook et al.
Construal Level Theory and Regulatory Scope (Psychology), Alison Ledgerwood et al.
Resource Limitations in Visual Cognition (Psychology), Brandon M. Liverence
and Steven L. Franconeri
Neural and Cognitive Plasticity (Psychology), Eduardo Mercado III
Attention and Perception (Psychology), Ronald A. Rensink