Understanding voice naturalness

Full text

TICS 2676 No. of Pages 14
Trends in
Cognitive Sciences OPEN ACCESS
Review
Understanding voice naturalness
Christine Nussbaum
1 ,2 ,3 ,
*,
1
Department for General Psychology
and Cognitive Neuroscience, Friedrich
Schiller University Jena, 07743 Jena,
Germany
2
Sascha Frühholz
3,4 ,5
, and
Voice Research Unit, Friedrich Schiller
University, 07743 Jena, Germany
3
The Voice Communication Sciences
(VoCS) MSCA Doctoral Network
4
Department of Psychology, University
of Oslo, 0371 Oslo, Norway
Stefan R. Schweinberger
1,2,3,6 ,7
5
Cognitive and Affective Neuroscience
Unit, University of Zurich, 8050 Zurich,
Switzerland
6
Swiss Center for Affective Sciences,
University of Geneva, 1222 Geneva,
Switzerland
7
German Center for Mental Health
(DZPG), Site Jena-Halle-, Magdeburg,
Germany
The perceived naturalness of a voice is a prominent property emerging from vocal
sounds, which affects our interaction with both human and artificial agents. Despite
its importance, a systematic understanding of voice naturalness is elusive. This is
due to (i) conceptual underspecification, (ii) heterogeneous operationalization,
(iii) lack of exchange between research on human and synthetic voices, and
(iv) insufficient anchoring in voice perception theory. This review reflects on
current insights into voice naturalness by pooling evidence from a wider inter-
disciplinary literature. Against that backdrop, it offers a concise definition of
naturalness and proposes a conceptual framework rooted in both empirical find-
ings and theoretical models. Finally, it identifies gaps in current understanding of
voice naturalness and sketches perspectives for empirical progress.
Naturalness: a prominent aspect of voice perception
Naturalness plays a significant role in how we perceive our environment through sight, sounds,
smell, taste, and touch. For example, perceptions of naturalness influence food choices, environ-
mental preferences, and social trust [1–3]. From a biological perspective, perceptions of natural-
ness may be considered an adaptive norm, where behaviors or traits that significantly deviate
from this norm are considered ‘unnatural.’ Beyond the biological context, the recent emergence
of artificial intelligence (AI)-generated digital and virtual contexts has brought human–machine in-
teractions to everyday life, thus bringing questions about naturalness to the forefront of scientific
research. One of the prime channels for communicative interactions is the voice [4], both in a
purely human context and beyond – with current voice synthesis (se e Glossary) technology
quickly invading everyday life, both in good use (e.g., in customer service calls, public transport,
gaming, or support platforms [5,6]) and in abuse (e.g., deepfakes [7]).
When we hear voices, we form intuitive impressions about them within just a few hundred
milliseconds [8–10]. Crucially, listeners are very sensitive to impressions of voice (un)naturalness.
Unnatural voices may sound nasal or robotic or may differ from the norm in pitch contour, temporal
structure, or spectral composition; in short, there are many ways in which a voice can lack natural-
ness [11]. Importantly, variations in naturalness affect communicative quality [12,13]. Evidence from
speech-language pathologies suggests that individuals with compromised speech naturalness are
often perceived as withdrawn, cold, introverted, or bored [14], which can lead to social isolation
and reduced quality of life [15–17] even when speech intelligibility is preserved [18]. Accordingly,
voice naturalness is a key target of speech therapy across various voice alterations [18–20]. A re-
cent survey on personalized speech synthesis for people who lost their biological voice found that a
majority prefer a more natural-sounding voice, even at the cost of some loss in intelligibility, both as
users and as listeners [21]. Thus, for human-to-human interaction, reduced voice naturalness con-
sistently has negative implications.
However, this is less clear for human–machine interac tion. The Computers-Are-Social-Actors
(CASA) framework proposed in the 1990s [22] assumed that we treat artificial agents like
Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx https://doi.org/10.1016/j.tics.2025.01.010 1
© 2025 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Highlights
Voices elicit impressions about their nat-
uralness, which affect interactions be-
tween humans as well as with artificial
agents.
Despite its intuitive appeal and practical
importance, a systematic understanding
of voice naturalness is elusive – the con-
cept is scientifically ill-defined.
Current voice naturalness research is sit-
uated within different research domains
that resemble echo chambers within
science – they cross-refer neither to
one another nor to current voice percep-
tion theory.
This review offers a concise conceptual
framework by proposing a taxonomy
with two distinct types: deviation-based
naturalness and human-likeness-based
naturalness.
This is compiled into practical recom-
mendations and perspectives for natu-
ralness research, because in a world of
digital agents, understanding the deter-
minants for how humans perceive natu-
ralness in social stimuli is a priority.

*Correspondence:
christine.nussbaum@uni-jena.de
(C. Nussbaum).
humans, fueling an (implicit) naturalness-is-better bias. This spurred efforts to create synthetic
voices that resemble human vocal expression [23,24], even when the link between naturalness
and success in human–machine interactions remains far from fully understood. While initial find-
ings suggested that reduced naturalness in synthetic voices compromises likeability, trustworthi-
ness, and pleasantness [11,25–28], contemporary synthetic voice design questions a ‘one size
fits all’ idea and instead advocates solutions tailored to specific applications [29]. Accordingly,
maximum human-likeness of synthetic voices may not always be required or desirable. Instead,
synthetic voice preferences may depend on the features of the listeners [27,30], the device
[31–33], and its specific function [6,25,31]. Understanding and incorporating such preferences
seems crucial for the success and acceptance of these devices [28].
Trends in Cognitive Sciences
OPEN ACCESS
Given its widespread practical importance, the role of voice naturalness warrants scientific scrutiny.
However, although many recent studies provide useful empirical insights, the current landscape re-
sembles a patchwork rather than a cohesive research field. There are four key issues within the
existing literature: (i) conceptual underspecification, (ii) heterogeneous operationalization, (iii) lack
of exchange between research domains, and (iv) insufficient anchoring in voice perception theory.
These challenges have likely precluded a systematic understanding of vocal naturalness, limited
visibility to a wider audience, obscured crucial research questions, and led to a divergence be-
tween theory and practice. The following sections elaborate on each of these problems before pro-
posing concrete measures to address them.
Current problems in voice naturalness research
Conceptual underspecification
Voice naturalness lacks a consistent definition and terminology in the literature (Figure 1A,B).
Many papers do not even provide an explicit definition of naturalness (Box 1). In these studies,
the conceptualization of naturalness must be inferred from the empirical design. If definitions
are provided, they often vary across research contexts (Table 1, for examples). In speech-
language pathology, some researchers refer to the definition provided by Yorkston and
colleagues (1999): ‘Naturalness is defined as c onforming to the listener’s standards of rate,
rhythm, intonation, and stress patterning and to the syntactic structure of the utterance being pro-
duced’ [34]. By contrast, research on synthetic and non-human voices usually defines naturalness
as ‘speech most closely perceived as a human voice’ [35]or ‘the degree to which a user feels a
certain technology or system is human-like’ [36]. Accordingly, many studies using synthetic voices
do not refer to naturalness but to human-likeness or anthropomorphism of voices.
Interestingly, these definitions seem to share two important assumptions. First, voice naturalness is
a perceptual and subjective measure [37]. Second, listeners’ naturalness perception is the result of
a complex multifactorial impression formation, presumably based on the integration and weighting
of many acoustic cues [38]. Beyond this, conceptualizations are highly heterogeneous because
they are tailored to the respective empirical focus. These prevailing inconsistencies alongside het-
erogeneous terminology (discussed next) make it challenging to compare and integrate different in-
sights. Therefore, there is a strong need to unite them under a concise conceptual framework.
Heterogeneous operationalization
A common consequence of inconsistent conceptualization is heterogeneous operationalization.
Primarily, this concerns the studied vocal categories and features, which include human versus
synthetic voices [30,39–42]; cartoon voices [43]; pathological voices such as in individuals with
Parkinson’s disease [44–47], tracheoesophageal speech [48,49], dysarthria [50–53],
Down syndrome [54], or stuttering [19]; acoustically manipulated human voices [55]; vocal fry
[56]; as well as different accents [57,58], dialects [59], age groups [60–62], and gender identities
2Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx

Trends in Cognitive Sciences OPEN ACCESS
TrendsTrends inin CognitiveCognitive SciencesSciences
Figure 1. Terminology and interconnectivity of voice naturalness research. (A) Word cloud depicting synonyms and
closely related concepts from 72 publications that target naturalness in voices (for details, see Box 1 in the main text). Word
size represents number of occurrences. (B) A similar word cloud but generated by ChatGPT (https://chatgpt.com/?oai),
when prompted to generate ten synonyms each for pathological, synthetic/manipulated, and healthy voices, together with
relative occurrence frequency. The full prompt, the generated response, and a reflection on its strengths and limitations
are accessible on the associated OSF repository (https://doi.org/10.17605/OSF.IO/ASFQV). (C) A bibliographic network
visualization using VOSviewer [68], covering publications relatedto voice naturalnessacrossdifferent domainsand ten
basic voice theory papers (see the supplemental information online for a full list of references). Each colored dot represents
a publication and gray links represent citations. Size of the dots indicate the number of links to other publ ications.
Clustering (depicted by different dot colors) is performed automatically in VOSviewer. Closer inspection reveals that green
refers to basic voice theory pape rs, red corresponds predominantly to papers on pathological voices, and blue refers to
synthetized/manipulated voices. A full documentation and an interactive version of the bibliographic network can be found
on the associated OSF repository (https://doi.org/10.17605/OSF.IO/ASFQV).
Glossary
Acoustic cues: physical and
measurable features of sounds (such as
voices); these may include fundamental
frequency, intensity, a range of timbre
cues, or temporal characteristics. Used
by listeners to inform manifold
impressions about voices, such as
emotion, identity, age, gender, or
naturalness.
Anthropomorphism: the attribution of
human characteristics, emotions, or
behaviors to non-human entities.
ChatGPT: a chatbot developed by
OpenAI, based on a large language
model, that generates text on the basis
of input prompts. (GPT stands for
generative pretrained transformer.)
Deepfakes: digitally manipulated
media, such as images, videos, or voice
recordings, created using deep learning
techniques with the goal to convincingly
display the appearance of a specific
individual.
Deviation-based naturalness:
conceptualization as the deviation from a
reference t hat represents maxi mum
naturalness.
Dysarthria: impairments of speech
motor subsystems due to various
neurological conditions such as
Parkinson’s disease, amyotrophic lateral
sclerosis (ALS), developmental
conditions, strokes, or traumatic brain
injury.
Human-likeness-based
naturalness: conceptualization of
naturalness by its resemblance to a real
human voice.
Laryngectomy: surgical removal of the
larynx, typically in the context of laryngeal
cancer treatment.
Prosody: dynamic voice intonation, as
expressed in pitch, loudness, timbre,
and rhythm. Sometimes also referred to
as ‘voice melody.’
Tracheoeso phageal speech: a
method of vocalization following total
laryngectomy via a tracheoesophageal
prosthesis that enables speech through
esophageal vibrations.
Uncanny valley: asudden feeling of
eeriness evoked by humanoid robots
that almost approach but do not entirely
reach a human-like appearance.
Voice synthesis: creation of
computer-generated voices. Common
methods are articulatory synthesis,
concatenative synthesis, and statistical
parametric synthesis, including deep
learning algorithms.
Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx 3

Trends in Cognitive Sciences
OPEN ACCESS
Box 1. A field in numbers
For a more systematic overview of scientific insights into naturalness in voices, a focused literature search on Web of Science
was conducted on April 26, 2023, using the search terms ‘naturalness AND voice’ or ‘human-likeness AND voice,’ which
was repeated on May 28, 2024, to detect the most recent papers. This initial search resulted in 339 articles, to which the
following inclusion criteria were applied: (i) language of publication was English; (ii) papers were published in peer-reviewed
journals or as a conference contribution; (iii) voice naturalness/human-likeness was either measured or manipulated; (iv) pa-
pers reported either a quantitative empirical analysis of human performance/perception data or a literature integration of such
works; thus, works on automatic naturalness classification and mere descriptions of toolboxes or datasets were excluded;
and (v) finally, the search was focused on spoken utterances, excluding singing voices and nonlinguistic vocalizations. Fol-
lowing these criteria, the reference lists of the identified articles were also screened for relevant publications. For a full docu-
mentation of all included papers and a reflection on potential biases in the literature search, please refer to the supplemental
information online and the associated OSF repository (https://doi.org/10.17605/OSF.IO/ASFQV).
In total, 72 articles were identified, covering a time range from 1984 to 2024. Thirty-eight (53%) were published in the past 5 years.
Sixty-seven report behavioral empirical data, of which 48 are predominantly ratings. Two are literature reviews, and three used
neurophysiological measures. Regarding voice category, 33 used synthetic, 18 human-pathological, six human-manipulated,
and five healthy human voices. Ten used more than one of these voice categories. In only 32 papers, an explicit de finition of nat-
uralness could be identified (see Table 1 in the main text for examples and the supplemental information online for a full list). These
articles presented a large variability in wording and vocabulary. In an attempt to capture this verbal space, all articles were
scanned for synonyms and closely related concepts of naturalness. The output is captured in the word cloud in Figure 1A
in the main text. Subsequently, these were compared with the articles’ keywords: 58 papers provided keywords, but only
32 had keywords related to naturalness or any of its synonyms. Finally, the conceptualization of naturalness was coded ac-
cording to the taxonomy we proposed. In case no definition of naturalness was provided, the ‘implicit’ conceptualization was
inferred from the research design. With this approach, we concluded that 26 employed a deviation-based conceptualization,
35 used human-likeness, and 11 used a combination of both.
Table 1. Example definitions of deviation-based and human-likeness-based voice naturalness
a
Conceptualization Definition Refs
Deviation-based
naturalness
‘Naturalness was defined as conforming to the listener’s standards of rate,
rhythm, intonation, and stress patterning and to the syntactic structure of the
utterance being produced.’ (p. 4687)
[44]
‘Speech naturalness can be described as how the speech of a person with a
speech disorder compares with that of typical speech or, in the case of an
acquired disorder, how an individual’s speech compares to its premorbid state’
(p. 1134)
[14]
‘Speech naturalness refers to a rather broad perceptual impression
representing the overall quality of a person’s speech output in relation to what is
conceptualized as normal or natural’ (p. 1633/1634)
[51]
‘[…] degree to which individuals sound ‘different’ from healthy peers’ (p. 1265) [53]
Human-likeness-based
naturalness
‘Human likeness has been used […] to describe how accurately the machine is
able to imitate a human.’ (p. 2864)
[26]
‘Naturalness refers to whether synthetic speech is perceived as uniquely
human, despite being computer-generated.’ (p. 5)
[21]
‘Natural speech is the speech most closely perceived as a human voice.’
(p. 10)
[35]
‘Naturalness refers to how closely the output sounds like human speech.’
(p. 389.e1)
[42]
Combination of both ‘By naturalness, we understand the voice stimulus to be perceived as a
plausible outcome of the human speech production system’ (p. 1)
[74]
‘[…] voices which sound like they could come from an actual human being
(which should be rated as more natural) and voices that sound more fictitious,
such as a cartoon character or a monster (which should be rated as less
natural).’ (p.429)
[57]
a
Note: definitions are all original quotes from the respective references. The full compilation of extracted definitions can be found
in the supplemental information online. Note that the mapping of definitions to the conceptualization of naturalness was carried
out by us and not the authors of the original publications.
4Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx

[20,63,64]. In addition, it concerns the experimental designs and measurements, especially rating
scales that differ in the number of levels and denomination s of end points. For example, in one
study, participants we re asked ‘How natural is the audio?’ from ‘1 – natural’ to ‘5 – unnatural’
[65]; in another one, they rated voices on a 10-point scale from ‘very natu ral, human-like’ to
‘very mechanical, robot-like’ [58] or made a binary classification of voices as either human or
computer-generated [37]. In principle, such empirical heterogeneity can be a powerful source
of insight, potentially revealing the degree to which methodological aspects affect results. For ex-
ample, there is recent evidence from face perception that differences in rating scales may not
have a large impact on outcome [66]. However, it cannot be concluded that this generalizes to
naturalness ratings, and the insufficient report of empirical details impedes a meaningful compar-
ison of findings. Specifically, it is often not stated how naturalness and the related experimental
task were explained to the listeners, but instructions can be crucial determinants of study out-
come. Furthermore, the precise acoustic properties of voice material often remain elusive, bearing
a risk for potential undetected confounds. Finally, only a few studies provide measurements of in-
terrater reliability [67]. To help address these issues, Box 2 provides a compilation of practical rec-
ommendations as guidance for future research.
Trends in Cognitive Sciences OPEN ACCESS
Lack of exchange between different research domains
Research on voice naturalness is inherently interdisciplinary, with two main domains: speech-
language pathology and synthetic voices. However, while the scientific findings are acknowl-
edged and referenced within each domain, these domains are poorly interconnected.
Figure 1C illustrates this via a cross-citation analysis using VOSviewer [68], showing several dis-
tinct clusters of studies reminiscent of echo chambers that are frequently discussed in social
media [69]. Poor interconnectivity is not unique to naturalness but can affect many other research
domains within person perception. Consider fields with different research traditions, such as
Box 2. Practical recommendations for voice naturalness research
Research on voice naturalness is highly interdisciplinary. To make future research accessible to a wider readership across
disciplines and to allow comparability and integration of findings, awareness of this interdisciplinarity is crucial. Here is a
compilation of some practical recommendations as a tentative roadmap for future research:
. Offer a concise definition of voice naturalness to both participants and readers. With the taxonomy of naturalness, this
article offers a conceptual framework that can be tailored to any empirical design, such as by specifying the reference
and the type of deviation under study. If used consistently, this taxonomy offers a quick orientation for readers and fos-
ters comparability across findings.
. Use consistent keywords to make relevant research findable across disciplines. We recommend ‘naturalness’;
‘human-likeness’; or, in appropriate cases, ‘authenticity.’
. Include full reports on methodological details. Specifically, this concerns acoustic manipulations that target voice nat-
uralness, measurements (i.e., rating scales used to assess naturalness impressions), instructions to raters, and reports
on reliability. For synthetic voices, be as specific as possible on synthesis methods, toolboxes and their settings, and
any additional processing that is applied.
. Wherever possible, provide stimulus examples. This is important because readers may have a clear idea of how a male
versus female voice sounds or how an angry voice differs from a happy one, but their imagination of an (un)natural or
synthetic voice could be quite vague and differ tremendously from the actual audio material. Often, direct auditory im-
pressions can be co mplementary to and more insightful than a list of acoustic measures and descriptions. In some
cases (i.e., when very different synthesis methods were used), differences in audio material may offer a straightforward
explanation for different empirical outcomes.
. Communicate findings inclusively enough for readerships from diverse backgrounds. Provide explicit definitions (e.g., for
terms such as ‘prosody,’‘dysarthria,’ or ‘anthropomorphism’), avoid technical jargon, including abbreviations unfamiliar
to other fields (e.g., synthesis algorithms, machine learning approaches, or acoustic measures), adopt scientific stan-
dards from other fields where appropriate, and discuss findings against the wider interdisciplinary literature (i.e., linking
insights into pathological voices to synthetic ones and vice versa).
. Quantify naturalness whenever it could have important implications for the ecological validity of the stimulus material, even
when naturalness is not the primary focus of the study. This is especially important when using acoustic manipulations
that could have unintended side effects on perceived naturalness [74,116].
Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx 5

impression formation according to social psychological models of intergroup perception versus
face/voice perception models. The se models were developed for different types of perceptual
cues, and different two-factor models with different labels have been proposed in both cases
(e.g., warmth vs. competence, e.g., [70]; or trus tworthiness vs. dominance, e.g., [71]).
More recently, though, these fields arguably benefited from interconnectivity, with substantial
research to link these distinct clusters and uncover both these specific taxonomies and their empir-
ical relationships [72,73]. In the case of voice naturalness, however, two recent systematic
literature reviews on pathological [17] and synthetic voices [23] do not have a single reference
in common. One might argue that this is not problematic, because the different disciplines simply
have different interests and readerships. However, some intriguing commonalities and systematic
patterns only emerge when pooling evidence from all available angles. For example, across syn-
thetic, pathological, and acoustically manipulated voices, converging evidence emerges for a
strong effect of pitch variation on perceived naturalness [14,26,74]. Furthermore, although sev-
eral studies failed to find an uncanny valley [75] effect for synthetic voices [11,76], a recent
study suggests it might exist for pathological voices [77]. This lack of exchange between research
fields not only has precluded relevant insights but also has likely impeded the visibility and impact
of voice naturalness research as a whole.
Trends in Cognitive Sciences
OPEN ACCESS
Insufficient anchoring in voice perception theory
The majority of naturalness research comes from applied fields, aiming to optimize artificial agents
or to improve the quality of life in patients with voice disorders. These findings provide valuable
practical knowledge, but they are insufficiently anchored in voice perception theory. As an illustra-
tion, we added ten influential, theory-building voice perception publications to the VOSviewer
analysis (Figure 1C), with the outcome suggesting that these tend to be ignored by most previous
naturalness research. Indeed, several authors have pointed out that research on voice natural-
ness lacks theoretical perspectives on vo ice perception and voice analysis [17,23]. This leaves
us with an intriguing divergence between increasing applied knowledge in rapidly developing
branches (especially synthetic voices) on the one hand and a simultaneous lack of understanding
of basic mechanisms on the other hand. To fully understand how naturalness affects our percep-
tion and response to voices, this void needs to be filled.
Toward a concise framework for voice naturalness
After identifying key problems that impede a systematic understanding of naturalness in voices, a
logical next step is to propose concrete measures to address them, starting with a conceptual
framework for the explicit definition of naturalness in voices.
Definitions of naturalness
We propose a taxonomy with two distinct types: deviation-based naturalness and human-
likeness-based naturalness (Figure 2, Key figure). In deviation-based naturalness, naturalness is
defined as the deviation from a reference that represents maximum naturalness. Example instruc-
tions for raters could be ‘Does this voice sound distorted?,’‘Does this voice sound unusual?,’ or
just ‘Does this voice sound natural?’ This conceptualization needs two important specifications:
the reference representing maximum naturalness and the type of deviation. In some cases, the
reference is explicitly provided for example, through a comparison or baseline stimulus (see
[78]). However, in many studies, raters are instructed to use an inner implicit reference that is
based on their experience and expectations for exampl e, judge whether ‘it conforms to the ex-
pected standard of unimpaired speech’ [52]. Thetypeof deviation is specified through the
vocal material. It can virtually cover all acoustic features, ranging from specific man ipulations
(e.g., spectral features or speech rate [79–81]) to complex multivariate vocal patterns (e.g., in
distorted or pathological voices [82]).
6Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx

Trends in Cognitive Sciences OPEN ACCESS
Key figure
A conceptual framework for the definition of voice naturalness
TrendsTrends inin CognitiveCognitive SciencesSciences
Figure 2. Assessing the naturalness of voices requires a reference frame (left panel), which is most commonly represented by the voice production system of humans. This
human production system sets the reference either as individual voice samples (explicit target voice) or as prototype voice representations (implicit prototype voice), against
which test voice samples (right panel) are assessed for naturalness. Two types of naturalness assessments are proposed (middle panel). The deviation-based approach
assesses naturalness in terms of distance away from the reference, while the human-l ikeness-based approach assesses naturalness according to its similar ity to the
reference. Deviation in voice naturalness can occur, for example, due to clinical conditions, voice manipulations, and acoustic artifacts. Human-likeness-based
naturalness defines naturalness by its resemblance to a real human voice. Human-likeness can be assessed from audio samples by judging whether they lie within the
perimeter of an acceptable human voice border.
Human-likeness-based naturalness defines naturalness by its resemblance to a real human
voice. Instructions for raters could be ‘Does this voice sound like a real human speaker?’ or
‘How human-like does the voice sound to you?’ Compared with the deviation-based definition,
the concept of human-likeness-based naturalness requires an additional obligatory assumption:
the existence of a non-human voice space. This highlights the notion of a categorical boundary to
human voices, although the transition between categories can be continuous. In other words, a
definition of human-likeness is only meaningful if we assume that voices can be non-human in
principle. Although deviation-based naturalness may, in certain cases, cross the boundary to
the non-human voice space, this boundary is not essential for its definition. Apart from this critical
distinction, however, human-likeness-based naturalness may represent a special case of
deviation-based naturalness: the reference is a human voice (or listeners´ representation of a
human voice), and the deviation is assessed along the human–non-human spectrum. The
above considerations suggest that the human-likeness–based conceptualization is partic ularly
well-suited for research into synthetic voices.
With this taxonomy, we provide a flexible and intuitive reference for the explicit definition of natural-
ness alongside its underlying assumptions. With future research committed to one conceptual
framework, systematic integration and comparison of findings could be greatly facilitated. In fact,
both conceptualizations seem already prevalent (Table 1), but they often remain implicit through
certain design choices only (Box 1). For example, comparing human with synthetic voices typically
implies human-likeness-based naturalness, whereas assessment of pathological voices often em-
ploys the deviation-based approach. One study deserves particular mention: Diel and Lewis [77]
studied the uncanny valley effect in different types of unnatural voices. They found that impressions
of uncanniness resulted from ‘deviation from familiar categories’ rather than ‘categorical ambiguity.’
This could reflect initial empirical observations in line with our proposed conceptual distinction.
Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx 7

Trends in Cognitive Sciences
OPEN ACCESS
Delimiting distinctiveness and authenticity
The following section briefly discusses the demarcation of the proposed definitions of naturalness
from two established concepts in perception research, starting with distinctiveness. Distinctive-
ness, as opposed to typicality, has been defined as the degree to which faces or voices stick
out d ue to rare or unusual features, and this concept is commonly used to refer to identity
[83,84]. According to face or voice space models, individual instances are represented along mul-
tiple perceptual dimensions, and they appear distinctive if they deviate substantially from a central
tendency or norm in that space. Our deviation-based definition of naturalness is closely related to
the concept of distinctiveness, as both share two critical features: a norm/reference and a devi-
ation. However, distinctiveness, as a different concept, can capture multiple forms of deviations
beyond naturalness. Accordingly, while unnatural voices would commonly be perceived as
somewhat distinctive, natural voices can be distinct or typical. However, one may speculate
that impressions of human-based naturalness could be quite independent from impressions of
distinctiveness under certain conditions. For instance, a person who is very accustomed to a
smart-speaker device may not rate synthetic voices as very distinctive but still clearly non-
human. In that vein, the link between distinctiveness and naturalness may be not primarily a con-
ceptual but an empirical matter requiring future inspection.
A second concept that deserves particular c onsideration is a uthenticity. I n the scientific litera-
ture, authenticity is an established term with meaning that may refer to vocal emotion, identity,
or gender – rather than the holistic impression of a voice. Emotional authenticity, for example,
refers to the distinction between a posed and a ‘real’/spontaneous emotional expression,
which leads to differential behavioral and neural outc omes [85–87]. In the con text of voice clon-
ing and the now very prevalent challenge of deepfakes [7], identity authenticity is assessed with
regard to a specific speaker. In principle, authenticity c an be assessed with regard to manifold
social signals, including age, gender, or even personality [88,89]. At first sight, the concepts of
authenticity and naturalness appear highly similar. In fact, when ChatGPT was prompted for
synonyms of naturalness, authenticity was its first reply (Figure 1B), which may suggest that
in openly accessible online sources, these two terms are indeed frequently occurring in an in-
terchangeable manner. Accordingly, i t might be argued that authenticity is just a special form of
deviation-based natur alness, with a more sp eci fic reference. For example, ‘Does this sound
like a natural voice?’ is converted into ‘Does this sound like a natural emotional ex pression?’
However, if considered against the backdrop of voice perception theory, it becomes apparent
that assessments of naturalness an d authenticity appear at different stages of voice processing
(Figure 3). Thus, it would be preferable to keep the concepts of naturalness and authenticity
rather separate.
Converging evidence
In our view, understanding voice naturalness requires pooling evidence from all relevant fields.
Even when these may nurture different perspectives on voice naturalness, they are united by
overarching questions: how do we form an impression about voice naturalness? Which
acoustic features affect this impression? How does naturalness impact perception, interac-
tion, and communication? Can we understand differences across individuals and listening
contexts?
In principle, conceptual progress for disintegrated – but also highly interdisciplinary – naturalness
research can be achieved by two measures: (i) converting empirical heterogeneity from an imped-
iment into an advantage and (ii) fostering mutually beneficial exchange between fields. Awareness
of the interdisciplinary nature of the field is crucial for implementing both measures. First, publica-
tions need to be findable and accessible, preferably through the establishment of common
8Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx

terminology that feeds into common keywords. Second, findings need to be communicated in-
clusively for readerships from diverse background s. Finally, conceptual and empirical aspects
need to be reported with sufficient detail to promote comparability. In Box 2, these suggestions
were converted into practical recommendations.
Trends in Cognitive Sciences OPEN ACCESS
TrendsTrends inin CognitiveCognitive SciencesSciences
Figure 3. Rooting voice naturalness in voice-processing theory. Theories of voice perception suggest a multilevel processing approach for voice samples
(left panel), which involves analyzing these samples on the basis of their features and auditory object patterns (middle panel), followed by an analysis of the
information conveyed by the voice signals (right panel). Assessing the nat uralness of voices appears at the level of voice features (low-level auditory analysis) and
voice object analysis (voice structural analysis) and includes the assessment of acoustic deviations and acoustic likeness, as well as the assessment of pattern
deviations and pattern likeness to reference voice samples. Unlike naturalness assessments, authenticity judgments mainly concern the assessmentof
communicative and social content carried by the voice signal at the level of voice information analysis. Such voice content either can be expressed spontaneously
(authentic) or can be enacted (nonauthentic), or it could be real or fake in relation to person-related identity information. Naturalness and authenticity assessments
may have mutual influences.
Progress along these lines not only will enhance mutual inspiration between clinicians and engi-
neers but also could foster innovative health t echnology. For instance, voice naturalness is a
key objective for cochlear implant (CI) research, where a sensory prosthesis restores hearing in
people with sensorineural deafness by resynthesizing auditory signals for direct electrical stimu-
lation of the cochlea [90], and real-time synthesis in CI sound processors could be modified to
achieve better perceptual outcomes, ultimately benefiting quality of life [91]. For people who are
predicted to lose their personal voice due to progressive disorders such as amyotrophic lateral
sclerosis (ALS) or due to planned laryngectomy, current voice banking technology already
allows personalized speech synthesis with the patient´s former individual voice, often with remark-
ably high ratings of both naturalness and authenticity [21,92].
Naturalness research rooted in voice perception theory
So far, no considerable efforts have been made to link naturalness perception to distinct stages
of voice processing. As discussed earlier, the topic of voice naturalness is highly influenced by
research perspectives from applied sciences and seemingly less by basic voice research and
its theoretical approaches. However, neurocognitive models of voice perception can provide
process-related perspectives on multilevel voice perception and voice information analysis.
This allow s us to link the mechanisms underlying voice naturalness assess ments to the appro-
priate level of voice analysis. Influential theories of voice perception propose sequential and
partly hierarchical stages of voice processing, including a major distinction between mecha-
nisms for voice object analysis (i.e., perception of an auditory stimulus as a voice) as initial
stages that are followed by the analysis of communicative and social content carried by the
voice signal [4,93–95].
Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx 9

Trends in Cognitive Sciences
OPEN ACCESS
This processing distinction between voice object analysis and voice content analysis is relevant to
the conceptual distinction between the assessment of voice naturalness on the one hand and the
assessment of the authenticity of expressed voice content on t he other hand (Figure 3).
Assessing the naturalness of voices is conceptually associated with the initial levels of voice object
analysis, including the stages of low-level auditory analysis and the analysis of structural voice
patterns. Humans presumably assess acoustic feature deviations and acoustic feature likeness
as low-level naturalness assessments [96], whereas assessing pattern deviations and pattern
likeness concerns the assessments of natural or unnatural spectrotemporal voice profiles [97].
Whereas voice naturalne ss assessments likely take place at the earlier stages of voice object
analysis, authenticity assessments likely take place at later stages involving voice information
analysis. Voices are used as carriers to express communicative and social content. For example,
voices are used for speech communication, emotional expressions, and to produce individual
voice characteristics. Such voice content could be either spontaneous and authentic, or it
could be acted and thus rather nonauthentic [98]. This authentic/nonauthentic distinction specif-
ically also concerns person-specific identity information in voices, which could be real or fake [7].
Such authenticity assessments might be independent of naturalness assessments, although
there is also a possibility of mutual influences. For instance, perceiving a voice as unnatural
might bias nonauthenticity judgments of voice content and vice versa.
Perspectives for future research
Our theoretical considerations on the processing of voice naturalness call for investigations of its
time course and underlying brain mechanisms – relative to authenticity as sessment but also to
other voice characteristics. Initial evidence suggests that voice naturalness affects the brain re-
sponse as early as 200 ms after voice onset and interacts with the processing of vocal emotions
[99–101]. However, comparably early effects have been found for authenticity assessments
[86,102,103]. Although the interpretability of these findings is limited by the potential influence
of acoustic confounds, the findings suggest that naturalness and authenticity assessments are
both fast and fundamental parts of voice perception. However, electrophysiological insights di-
rectly comparing the time course of naturalness and authenticity are elusive, as is their interplay
with impressions of age, gender, or pers onality traits. A recent electroencephalographic (EEG)
study suggests that many first impressions formed from voices are highly intercorrelated [8],
but for naturalness, we are currently limited to behavioral data that point toward interactions
with age, gender, and emotion perception [60,63,74]. In a broad sense, naturalness impressions
are always formed against a specific context, whether that context refers to the voice itself or to
the properties of the interaction. Accordingly, whether the same voice is assessed in an all-human
or human–machine interaction context could make a crucial difference.
In that vein, while this review focuses on understanding naturalness in voices from an interdisci-
plinary perspective, we wish to emphasize the multisensory perspective of naturalness research.
In fact, substantial research in the domain of faces has compared the perceived naturalness or
realism of synthesized versus real faces (for a systematic review and meta-analysis, see [104]).
Recent research even demonstrated conditions in which synthesized faces can be perceived
as more human than genuine human faces. Moreover, an attempt to identify the visual features
that trigger such a paradoxical facial ‘hyperrealism’ effect suggested contributions of typicality,
familiarity, attractiveness, and low memorability [105]. Although this interpretation was based
on qualitative reports and requires converging evidence, such research can inspire the systematic
search for commonalities or differences between mechanisms that trigger judgments of voice or
face naturalness. Ultimately, naturalness research should also systematically consider interac-
tions between vocal and visual aspects o f naturalness in combination. Indeed, accumulating
10 Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx

evidence suggests a complex interplay of visual appearance, vocal features, behavior, and the
interactional context for the acceptance of virtual agents [28,31–33,106–113].
Trends in Cognitive Sciences OPEN ACCESS
Beyond humans, vocalizations are abundant in the animal kingdom. Many animals can manipulate
and adapt their vocal calls to specific situations or needs. For instance, birds living in urban environ-
ments modify their song in frequency or amplitude to avoid masking by constant anthropogenic
noise [114]. While this reduces the risk of not being heard by conspecifics, the degree to which
such urban-induced changes to natural patterns of vocalization may have other consequences
to communication seems unclear at present. Potentially, with appropriate adaptations, the present
taxonomy could be useful to promote an understanding of animal voice naturalness as well.
Finally, very recent fMRI research has uncovered a cortical-striatal brain network that is involved
when listeners try to distinguish deepfake from real speaker identities [7]. Such research is rele-
vant also because the accelerating spread of misinformation via social media is now considered
a major problem that compromises societal cohesion [69,115]. While large-scale misinformation
is still mostly text-based as of today, next-generation deepfakes likely will be even more efficient
vehicles of misinformation. This is because they efficiently instrumentalize person-related trust via
high-level perceptual deception. From that perspective, better understanding of the characteris-
tics of ‘successful’ vocal deepfakes and their processing in the brain may be one important com-
ponent for strengthening human resilience to fake information of the future.
Concluding remarks
Naturalness in voices is a highly intuitive concept, but one that is scientifically underspecified and
far from systematically understood, despite considerable research efforts. To address this, we
propose a conceptual framework for voice naturalness. Our taxonomy, composed of deviation-
based naturalness and human-likeness-based naturalness, is rooted in voice perception theory
and is inspired by interdisciplinary empirical findings. The new framework offers the flexibility
that is necessary to be applicable across diverse empirical designs while at the same time pro-
moting comparability across research domains. This conceptual groundwork is complemented
with several practical recommendations to bridge previously unconnected approaches and bet-
ter integrate this highly interdisciplinary field. This provides a foundation for conjoined efforts to-
ward more systematic future research on numerous open questions regarding voice
naturalness (see Outstanding questions). While the focusisonvoiceshere, we ultimately opt
for a multisensory perspective on naturalness research. In a world that is increasingly dominated
by digitally synthesized age nts, it seems important to identify the multifaceted det erminants for
human perception of naturalness in social stimuli.
Acknowledgments
We thank Simone Dahmen and Fatma Bilem for their support with the literature analysis and the members of the Jena Voice
Research Unit (https://www.voice.uni-jena.de/) for helpful suggestions on this project. The authors gratefully acknowledge
the award of funding through an EU's Marie Skłodowska-Curie Actions Doctoral Network ‘Voice Communication Sciences’
(action 101168998, https://www.vocs.eu.com/).
C.N.: I dedicate this work to our stillborn son. Thanks for changing our lives
Declaration of interests
The authors declare no competing interests.
Supplemental information
Supplemental information associated with this article can be found online at https://doi.org/10.1016/j.tics.2025.01.010.
Outstanding questions
Vocal communication is abundant in
the animal kingdom, and many
animals manipulate their vocal
behavior in an adaptive manner – is
there demand for a comparative per-
spective on voice naturalness?
How is a listener´s perception of
naturalness shaped through experience
(e.g., with voice assistants, smart home
devices, or patients with voice
disorders)?
With respect to the present conceptual
framework, (how) are human-likeness
based naturalness and deviation-
based naturalness dissociable in the
brain?
In the trade-off between precise exper-
imental control and open field record-
ings, can w e identify converging
evidence for how and when reduced
naturalness in voices critically affects
the ecological validity of research? In
depth, will we need a dynamic definition
of ecological validity in view of an ever
more digital world of social interaction?
Are natural voices always preferred, or
is naturalness preference context
dependent? Can natural voices impede
rather than promote communication
success in some situations?
Many domains of social perception are
characterized by individual variability,
but it is unclear whether there are
substantial individual differences in the
tolerance of or preference for
unnatural voice features. If so, can
these be re lated to other domains of
auditory cognition or to other person
traits?
To what extent is naturalness
perception affected by factors such as
age, gender, or cultural background?
Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx 11

Trends in Cognitive Sciences
OPEN ACCESS
References
1. Román, S. et al. (2017) The importance of food naturalness for
consumers: results of a systematic review. Trends Food Sci.
Technol. 67, 44–57
2. Meier, B.P. et al. (2019) Naturally better? A review of the natural-
is-better bias. Soc. Personal. Psychol. 13, e12494
3. Ode, A. et al. (2009) Indicators of perceived naturalness as drivers
of landscape preference. J. Environ. Manag. 90, 375–383
4. Young, A.W. et al. (2020) Face and voice perception: under-
standing commonalities and differences. Trends Co gn. Sci.
24, 398–410
5. Rodero, E. and Lucas, I. (2023) Synthetic versus human voices
in audiobooks: the human emotion al intimacy effect. New
Media Soc. 25, 1746–1764
6. Rodero, E. (2017) Effectiveness, attention, and recall of human
and artificial voices in an advertising story. Prosody influence
and functions of voices. Comput. Hum. Behav. 77, 336–346
7. Roswandowitz, C. et al. (2024) Cortical -striatal brain network
distinguishes d eepfake from real speaker identity. Com mun.
Biol. 7, 711
8. Lavan, N. et al. (2024) The time course of person perc eption
from voices in the brain. Proc. Natl.Acad.Sci.U.S.A. 121,
e2318361121
9. Lavan, N. (2023) How do we describe other people from voices
and faces? Cognition 230, 105253
10. Ji ang, Z. et al. (20 24) Comparison of face-based and voice-
based first impressions in a Chinese sample. Br.J.Psychol.
115, 20–39
11. Kühne, K. et al. (2020) The human takes it all: humanlike synthe-
sized voice s are perceived as less eerie and more likabl e. Evi-
dence from a subjective ratings study. Front. Neurorobot. 14,
593732
12. Ilves, M. and Surakka, V. (2013) Subjective responses to syn-
thesised speech with lexi cal emotional content: the effect of
the naturalness of the synthetic voice. Behav. Inform. Technol.
32, 117–131
13. Ilves, M. et al. (2011) The e ffects of emotionally worded syn-
thesized s peech on the rat ings of emotion s an d vo ice qual-
ity. In Affective Computing and Intelligent Interaction. ACII
2011. Lecture Notes in Computer Science (6974) (D’Mello,
S. et al ., eds), pp. 588–598, Spr inger
14. Anand, S. and Stepp, C.E. (2015) Listener perception of mono-
pitch, naturalness, and intelligibility for speakers with Parkinson’s
disease. J. Speech Lang. Hear. Res. 58, 113 4–1144
15. Moya-Galé, G. and Levy, E.S. (2019) Parkinson’s disease-
associated dysarthria: prevalence, impact and management
strategies. Res. Rev. Parkinsonism 9, 9–16
16. Damico, J.S., Ball, M.J., eds (2019) The SAGE Encyclopedia of
Human Co mmunicati on Sciences and Disorders,SAGE
Publications
17. Klopfenstein, M. et al. (2020) The study of speech naturalness in
communication disorders: a systematic review of the literature.
Clin. Linguist. Phon. 34, 327–338
18. Frankford, S.A. et al. (2024) Contributions of speech timing and
articulatory precision to listener perceptions of intelligibility and
naturalness in Parkinson’s disease. J. Speech Lang. Hear.
Res. 67, 2951–2963
19. Euler, H.A. et al. (2021) Speech restructuring group treatment
for 6-to-9-year-old children who stutter: a therapeutic trial.
J. Commun. Disord. 89, 106073
20. Hardy, T.L.D. et al. (2020) Acoustic predictors of gender attribu-
tion, masculinity-femininity, and vocal naturalness ratings
amongst transgender and cisgender speakers. J. Voice 34,
300.e11–300.e26
21. Hyppa-Martin, J. et al. (2024) A large-scale comparison of two
voice synthesis techniques on intelligibility, naturalness, prefer-
ences, and attitudes toward voices banked by individuals with
amyotrophic lateral scleros is. Augment . Altern. Commun . 40,
31–45
22. Nass, C. et al. (1994) Computers are social actors. In Proceedings
of the SIGCHI Conference on Human Factors in Computing Sys-
tems Celebrating Interdependence- CHI ‘94, ACM Press
23. Seaborn, K. et al. (2021) Voice in human–agent interaction.
ACM Comput. Surv. 54, 1–43
24. Triantafyllopoulos, A. et al. (2023) An overview of affective
speech synthesis and conversion in the deep learning era.
Proc. IEEE 1355–1381
25. Schreibelmayr, S. and Mara, M. (2022) Robot voices in daily life:
vocal human-likeness and application context as determinants
of user acceptance. Front. Psychol. 13, 787499
26. Baird, A. et al. (2018) The perception and analysis of
the likeability and human likeness of synthesized speech. In
Interspee ch 2018 ,pp.2863–2867, ISCA
27. Lee, E.-J. (2010) The more humanlike, the better? How
speech type and users’ cognitive style affect social responses
to computers. Comput. Hum. Behav. 26, 665–672
28. Lu, L. et al. (2021) Leveraging ‘human-likeness’ of robotic ser-
vice at restaurants. Int. J. Hosp. Manag. 94, 1–9
29. Cambre,J.and Kulkarni,C.(2019) Onevoice fits all? Proc.
ACM Hum.-Comput. Interact. 3, 1–19
30. Eyssel, F. and et al.Yanco, H. et al. (2012) If you sound like me,
you must be mor e human. In HRI’ 12. Proceedings of the Seventh
Annual ACM/ IE EE Conference on Human-Robot Int eraction:
March 5-8, 2012 Boston, Massachusetts, USA, pp. 125–126,
Association for Computing Machinery
31. Im, H. et al. (2023) Let voice assistants sound like a machine:
voice and task type effects on perceived fluency, compe-
tence, and consumer attitude. Comput. Hum. Behav. 145,
107791
32. McGinn, C. and Torre, I. (2019 - 2019) Can you tell the robot by
the voice? An exploratory study on the role of voice in the percep-
tion of robots. In 2019 14th ACM/IEEE International Conference
on Human-Robot Interaction (HRI),pp. 211–221, IEEE
33. Mitchell, W. J. et al. (2011) Amismatchinthe humanrealismof
face and voice produces an uncanny valley. i-Perception 2,
10–12
34. Yorkston, K.M. et al. (1999) Management of Motor Speech
Disorders in Children and Adults, PRO-ED
35. Mawalim, C.O. et al. (2022) Speaker anonymization by modify-
ing fundamental frequency and x-vector singular value.
Comput. Speech Lang. 73, 1–17
36. Hu, P. et al. (2021) Dual humanness and trust in conversational
AI: a pers on-centered a pproach. Comp ut. Hum. Behav. 119,
106727
37. Nusbaum, H.C. et al. (1997) Measuring the naturalness of syn-
thetic speech. Int. J. Speech Technol. 2, 7–19
38. Mayo, C. et al. (2011) Listeners’ weighting of acoustic cues to
synthetic speech naturalness: a multidimensional scaling analysis.
Speech Comm. 53, 311–326
39. Abdulrahman, A. and Richards, D. (2022) Is natural necessary?
Human voice versus synthetic voice for intelligent virtual agents.
MTI 6, 51
40. Urakami, J. et al. (2020) The effect of naturalness of voice and
empathic responses on enjoyment, attitudes and motivation
for interacting with a voice user interface. In Human-Computer
Interaction. Mu ltimodal and Natural Interaction (Kurosu, M.,
ed.), pp. 244–259, Springer International Publishing
41. Velner, E. et al. (2020) Intonation in robot speech. In Proceedings
of the 2020 ACM/IEEE Inter national Conferen ce on Human-
Robot Interaction (Belpaeme, T. et al., eds), pp. 569–578, ACM
42. Yamasaki, R. et al. (2017) Perturbation measuremen ts on the
degree of naturalness of synthesized vowels. J. Voice 31,
389.e1–389.e8
43. Ko, S. et al. (2023) The effects of robot voices and appearances
on users’ emotion recogni tion and subj ective pe rception. Int.
J. Human. Robot. 20, 2350001
44. Abur , D. et al. (2021) Feedback and feedforward audit ory-
motor processes for voice an d articulation in Parkinson’s
disease. J. Speech Lang. Hear. Res. 64, 4682–4694
45. Klopfen stein, M. (2015) Relationship between acoustic
measures and speech naturalness ratings in Parkinson’s
disease: a within-speaker approach. Clin. Linguist. Phon.
29, 938–954
46. K lopfenstein, M. (2016) Speech nat uralness ratings and per-
ceptual corr elates of highly natu ral and unnatural speech in
hypokinetic dysarthria secondary t o Parkinson’s disease.
JIRCD 7, 123–146
12 Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx

Trends in Cognitive Sciences OPEN ACCESS
47. Moya-Galé, G. et al. (2024) Perceptual consequences of online
group speech treatment for individuals with Parkinson’sdis-
ease: a pilot study cas e serie s. Int. J. Speech L ang. Pathol.,
Published onlin e May 1, 2024. https://doi.org/10.1080/
17549507.2024.2330538
48. Eadie, T.L. and Doyle, P.C. (2002) Direct magnitude estimation
and interval scaling of naturalness and severity in trache o-
esophageal (TE) speakers. J. Speech Lang. Hear. Res. 45,
1088–1096
49. Eadie, T.L. et al. (2008) Influence of speaker gender on listener
judgments of tracheoesophageal speech. J. Voice 22, 43–57
50. Yorkston, K.M. et al. (1990) The effect of rate control on the in-
telligibility and naturalness of dysarthric speech. J. Speec h
Hear. Disord. 55, 550–560
51. Schölderle, T. et al. (2023) Speech naturalness in the assess-
ment of childhood dysarthria. Am. J. S peech-Lang. Pathol.
32, 1633–1643
52. Lehner, K. and Ziegler, W. (2022) Clinical measures of commu-
nication limitations in dysarthr ia assessed through
crowdsourcing: specificity, sensitivity, and r etest-reliability.
Clin. Linguist. Phon. 36, 988–1009
53. Vogel, A.P. et al. (2019) Speech treatment improves dysarthria
in multisystemic ataxi a: a rater-bli nded, co ntrolled p ilot-study
in ARSACS. J. Neurol. 266, 1260–1266
54. Jones, H.N. et al. (2019) Auditory-perceptual speech features in
children with Down syndrome. Am. J. Intellect. Dev. Disabil.
124, 324–338
55. Assmann, P.F. et al. (2006) Effects of frequency shifts on per-
ceived naturalness a nd gender information in speech. In
INTERSPEECH
56. Venkatraman, A. and Sivasankar, M.P. (2018) Continuous vocal
fry simulated in laboratory subjects: a preliminary report on
voice production and listener ratings. Am. J. Speech-Lang.
Pathol. 27, 1539–1545
57. Kapolowicz, M.R. et al. (2022) Effects of spectral envelope and
fundamental frequency shifts on the perception of foreign-
accented speech. Lang. Speech 65, 418–443
58. Tamagawa, R. et al. (2011) The effects of synthesized voice
accents on user perceptions of robots. Int. J. Soc. Robot. 3,
253–262
59. Mackey, L.S. et al. (1997) Effe ct of speech dialect on speech
naturalness ratings: a systematic replication of Martin,
Haroldson, and Triden (1984). J. Speech Lang. Hear. Res. 40,
349–360
60. Goy, H. et al. (2016) Effects of age on speech and voice quality
ratings. J. Acoust. Soc. Am. 139, 1648
61. Coughlin-Woods, S. et al. (2005) Ratings of speech naturalness
of children ages 8-16 years. Percept. Motor Skills 100,
295–304
62. Baird, A. et al. (2017) Perception of paralinguistic traits in syn-
thesized voices. In Proceedings of the 12th International Audio
Mostly Conference on Augmented and Participato ry Sound
and Music Experiences (Fazekas, G. et al., eds), pp. 1–5, ACM
63. Merritt, B. and Bent, T. (2020) Perceptual evaluation of speech
naturalness in speakers of varying gender identities. J. Speech
Lang. Hear. Res. 63, 2054–2069
64. Baird, A. et al. (2018) The perception of vocal traits in synthe-
sized voice s: age, gender, and human likeness. J. Audio Eng.
Soc. 66, 277–285
65. Aylett, M.P. et al. (2020) Speech synthesis for the generation of
artificial personality. IEEE Trans. Affect. Comput. 11, 361–372
66. Kramer, R.S.S. et al. (2024) The psychometrics of rating facial
attractiveness using different response scales. Percep tion 53,
645–660
67. Martin, R.R. et al. (1984) Stuttering and speech naturalness.
J. Speech Hear. Disord. 49, 53–58
68. van Eck, N.J. and Waltman, L. (2010) Software survey:
VOSviewer, a computer progr am for bibliometric mapping.
Scientometrics 84, 523–538
69. van der Linden, S. (2023) Foolproof: Why We Fall for Misinfor-
mation and How to Build Immunity, W.W. Norton & Company
70. Fiske, S.T. (2018) Stereotype content: warmth and compe-
tence endure. Curr. Dir. Psychol. Sci. 27, 67–73
71. Todorov, A. et al. (2008) Understanding evaluation of faces on
social dimensions. Trends Cogn. Sci. 12, 455–460
72. Sutherland, C.A.M. et al. (2013) Social inferences from faces:
ambient images generate a three-dimensional model. Cognition
127, 105–118
73. Sutherland, C.A.M. et al. (2016) Integrating social and facial
models of person perception: converging and diverging dimen-
sions. Cognition 157, 257–267
74. Nussbaum, C. et al. (2023) Perceived naturalness of emotional
voice morphs. Cognit. Emot. 1–17
75. Mori, M. et al. (2012) The uncanny valley. IEEE Robot. Automat.
Mag. 19, 98–100
76. Romportl, J. (2014) Speech synthesis and uncanny valley. In Text,
Speech and Dialogue (Horák, A. et al. , eds), pp. 595–602,
Springer International Publishing
77. Diel, A. and Lewis, M. (2024) Deviation from typical organic
voices best explains a vocal uncanny valley. Comput. Hum.
Behav. Rep. 14, 100430
78. van Prooije, T. et al. (2024) Perceptual and acoustic analysis of
speech in spinocerebellar ataxia type 1. Cerebellum 112–120
79. Moore, B.C.J. and Tan, C.-T. (2003) Perceived naturalness of
spectrally distorted speech and music. J. Acoust. Soc. Am. 114,
408–419
80. Rao, M.V. et al. (2018) Effect of source filter interaction on
isolated vowel-consonant-vowel perception. J. Acoust. Soc.
Am. 144, EL95
81. Ra tcliff, A. et al. (2002 ) Fact or s influencing ratings of speech
naturalness in augmentative and alternative communication.
Augment. Altern. Commun. 18, 11–19
82. Meltzner, G.S. and Hillman, R.E. (2005) Impact of aberrant
acoustic properties on the perception of sound quality in
electrolarynx speech. J. Speech Lang. Hear. Res. 48, 766–779
83. Andics, A. et al. (2010) Neural mechanisms for voice recognition.
Neuroimage 52, 1528–1540
84. Valentine, T. et al. (2016) Face-space: a unifying concept in face
recognition research. Q. J. Exp. Psychol. (Hove) 69, 1996–2019
85. Lima, C.F. et al. (2021) Authentic and posed emotional vocaliza-
tions trigger distinct facial responses. Cortex 141, 280–292
86. Sarzedas, J. et al. (2024) Blindness influences emotional
authenticity perception in voices: behavioral and ERP evidence.
Cortex 172, 254–270
87. Anikin, A. and Lima, C.F. (2 017) Perceptual and acoustic
differences between authentic and acted nonverbal emotional
vocalizations. Q. J. Exp. Psychol. (Hove) 71, 622–641
88. Kachel, S. et al. (2020) Gender (conformity) matters: cross-
dimensional and cross-modal associations in sexual orientation
perception. J. Lang. Soc. Psychol. 39, 40–66
89. Mills, M. et al. (2017) Expanding the evidenc e: develop ments
and innovati ons in clinical practice, training and competency
within voice and communication therapy for trans and gend er
diverse people. Int. J. Transgend. 18, 328–342
90. von Eiff, C.I. et al. (2022) Crossmodal benefits to vocal
emotion perception in cochlear implant users. iScience 25,
105711
91. Schweinberger, S.R. and von Eiff, C.I. (2022) Enhancing socio-
emotional communication and quality of life in young cochlear
implant reci pients: perspectives from parameter-specific
morphing and caricaturing. Front. Neurosci. 16, 956917
92. Yamagishi, J. et al. (2012) Speech synthesis technologies for indi-
viduals with vocal disabilities: voice banking and reconstruction.
Acoust. Sci. Technol. 33, 1–5
93. Be lin, P . et al. (2004) Thinking the vo ice: ne ural co rrelates of
voice perception. Trends Cogn. Sci. 8, 129–135
94. Belin, P. et al. (201 1) Unders tanding voice perception. Br.
J. Psychol. 102, 711–725
95. Lavan, N. and McGettigan, C. (2023) A model for person per-
ception from familiar and unfamiliar voices. Commun. Psychol.
1, 1–11
96. S taib, M . and Fr ühholz, S. (2023) Distinct func tional levels of
human voice processing in the auditory cortex. Cereb. Cortex
33, 1170–1185
97. Staib, M. and Frühholz, S. (2021) Cortical voice processing is
grounded in elementary sound analyses for vocalization relevant
sound patterns. Prog. Neurobiol. 200, 101982
98. Pinheiro, A.P. et al. (2021) Emotio nal a uthenticity modu lates
affective and social trait inferences from voices. Philos. Trans.
R. Soc. Lond. Ser. B Biol. Sci. 376, 20200402
Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx 13

Trends in Cognitive Sciences
OPEN ACCESS
99. Duville, M.M. et al. (2022) Neuronal and behavioral affective per-
ceptions of human and naturalness-reduced emotional proso-
dies. Front. Comput. Neurosci. 16, 1022787
100. Duville, M.M. et al. (2024) Improved emotion differentiation
under reduced acoustic variability of speech in autism. BMC
Med. 22, 121
101. Nussbaum, C. et al. (2022) Contribut ions of fundame ntal fre-
quency and timbre to vocal emotion perception and their elec-
trophysi ological correlat es. Soc.Cogn.Affect.Neurosci. 17,
1145–1154
102. Kosilo, M. et al. (2021) The neural basis of authenticity recogni-
tion in laughter and crying. Sci. Rep. 11, 23750
103. Conde, T. et al. (2022) The time course of emotional authenticity
detection in nonverbal vocalizations. Cortex;J.Dev.Study
Nerv. Syst. Behav. 151, 116–132
104. Miller, E.J. et al. (2023) How do people respond to computer-
generated versus human faces? A systematic review and
meta-analyses. Comput. Hum. Behav. Rep. 10, 100283
105. Miller, E.J. et al. (2023) AI hyperrealism: why AI faces are per-
ceived as more real than human ones. Psychol. Sci. 34,
1390–1403
106. Cabral, J.P. et al. (2017) The influence of synthetic voice on the
evaluation of a virtual character. In Interspeech 2017, pp.
229–233, ISCA
107. Ehret, J. et al. (2021) Do prosody and embodiment influence
the perceived naturalne ss of conversa tional a gents’ speech?
ACM Trans. Appl. Percept. 18, 1–15
108. Ferstl, Y. et al. (2021) Human or robot? I nvesti gati ng voice,
appearance and gesture motion r ealism of conversational
social agents. In Proceedings of the 21th ACM International
Conference on Intelligent Virtual Agents, pp. 76–83, ACM
109. Go ng, L. and Nass, C. (2007) When a talking-f ace compu ter
agent is half-human and half-humanoid: human identity and
consistency preference. Human Comm. Res. 33, 163–193
110. Higg ins, D. et al. (2022) Sympat hy for the digital: influenc e of
synthetic voice on affinity, social presence and empathy for
photorealistic virtual humans. Comput. Graph. 104, 116–128
111. Li, M. et al. (2023) Effects of robot gaze and voice human-likeness
on users’ subjective perception, visual attention, and cerebral ac-
tivity in voice conversations. Comput. Hum. Behav. 141, 107645
112. Parmar, D. et al. (2 022) Designing empathic virtual agents:
manipulating animation, voice, rendering, and empathy to
create persuasive agents. Auton. Agent. Multi-Agent Syst. 36,
1–24
113. Sarigul, B. and Urgen, B.A. (2023) Audio–visual predictive pro-
cessing in the perception of humans and robots. Int. J. Soc .
Robot. 15, 855–865
114. Lowry, H. et al. (2013) Behavioural responses of wildlife to urban
environments. Biol. Rev. Camb. Philos. Soc. 88, 537–549
115. Kauk, J. et al. (2024) The adaptive community-response (ACR)
method for collecting misinformation on social media. J. Big
Data 11, 1–32
116. Malisz, Z. et al. (2020) Modern speech synthesis for phonetic
sciences: a discussion and an evaluation. pp. 487–491
14 Trends in Cognitive Sciences, Month 2025, Vol. xx, No. xx