The review, led by Jingjing Niu of Herzen State Pedagogical University of Russia together with Yinghao Qi and Xiaobin Li, synthesizes evidence from voice perception, vocal-performance education, music research, and adjacent creative domains. It centers on two psychological constructs. The first is vocal identity threat, defined as the perceived erosion of the bond between a person’s voice and their self-concept when facing an AI voice that can be cloned. The second is occupational self-efficacy, a singer’s subjective judgment of their ability to maintain professional competitiveness, complete occupational tasks, and cope with future work uncertainty. Crucially, the authors acknowledge that no empirical study has yet directly examined singers’ psychological responses to AI voice cloning. Their argument is therefore a conceptual and empirical foundation for future research rather than a set of established findings about singers themselves.

Why would a cloned voice threaten identity at all? Neurocognitive research offers a technical answer. Neuroimaging and EEG studies show that the human voice is not processed as an ordinary acoustic stimulus. Reliable neural differentiation between self and other voices supports specialized self-referential voice processing, and fMRI work on self-face and self-voice recognition places the voice within a broader multisensory system of self-representation. Research on the so-called rubber voice illusion shows that perceived voice ownership can be altered when external vocal signals are closely synchronized with one’s own behavior, echoing the better-known rubber hand illusion in body ownership research. Together, these findings establish that the voice functions as a highly self-relevant identity cue supported by self-prioritized processing, body ownership, and multisensory integration. They do not, by themselves, demonstrate that AI cloning produces identity threat, but they explain why the possibility is psychologically plausible.

The technical baseline of the threat is advancing quickly. NaturalSpeech 2, a system combining a neural audio codec with a diffusion model, achieves zero-shot speech and singing generation trained on 44,000 hours of data. In the controlled track of SVDD 2024, the inaugural singing voice deepfake detection challenge, the best system’s equal error rate dropped to 1.65 percent, meaning automated detectors are becoming highly accurate even as human listeners struggle. A meta-analysis of 56 papers on deepfake detection reported an average human detection accuracy of 62.08 percent for audio deepfakes, while a separate study found listeners classified AI-cloned and authentic voices with approximately 60.8 percent accuracy, significantly above chance but far from reliable. Detection accuracy increased for longer and unscripted clips, and listeners most frequently reported relying on vocal inflection, suggesting that access to richer expressive information facilitates discrimination between human and AI-generated voices.

This point may be particularly relevant to singing, which contains richer melodic, rhythmic, prosodic, and expressive information than ordinary speech. Singing voice identity involves far more than timbral similarity. Vocal timbre, pitch control, vibrato, register, articulation, phrasing, dynamics, and expressive timing all contribute to how a singing voice is produced and perceived. Reproducing a singer’s recognizable timbre does not necessarily entail reproducing the broader expressive characteristics of their performance. The authors argue that three distinctions are critical for interpreting the current evidence. First, timbral similarity should not be equated with artistic equivalence. Second, increasingly humanlike output does not establish perceptual interchangeability, because discrimination varies with stimulus characteristics, listener factors, and the availability of prosodic and expressive cues. Third, the implications of AI may differ depending on whether it substitutes for identity-bearing vocal labor or scaffolds human-controlled performance and creation.

Perception and social evaluation studies add nuance. Familiar listeners were able to distinguish cloned from recorded voices, whereas unfamiliar listeners were unable to distinguish them at the group level and sometimes rated cloned voices as favorably as or more favorably than the originals. Importantly, participants showed no detection advantage for clones of their own voice compared with clones of a friend’s voice, indicating that self-prioritized voice processing does not necessarily translate into superior conscious detection of one’s own voice clone. Prosodic cues can strengthen perceptual boundaries between human speakers and AI voice clones, suggesting that discrimination depends partly on the acoustic and expressive information available to listeners. For singers, whose audiences range from devoted fans who know every nuance to casual listeners who know only a hit single, the familiarity effect suggests that the implications of voice cloning may differ according to how familiar audiences are with the original voice.

Adjacent evidence from vocal-performance education is consistent with the possibility of psychological concern. In a single-time-point mixed-methods study at three Chinese universities, 414 undergraduate students in a Dubbing and Artistic Performance course completed measures of AI anxiety and performance appraisal in creative dubbing tasks. Higher AI anxiety was associated with greater underestimation of expressive performance relative to external ratings and with lower external ratings. However, these external evaluations were provided by peers without professional training, themes concerning artistic identity were derived from interviews with only five students, and career concern was included as an item within the AI anxiety measure. The findings should therefore be interpreted cautiously as adjacent evidence rather than as direct evidence from singers. Broader industry surveys document perceived occupational threat: a PRS for Music survey of over 2,630 members found that 79 percent of respondents worry about AI-generated music competing with human works, and 76 percent believe AI could harm creators’ livelihoods, while the 2025 UK Music creator survey showed that 66 percent of respondents agree that AI poses a threat to their career.

Yet the evidence does not point in a single direction. A quasi-experimental study of 40 violin performance students found that the group using AI-assisted practice applications scored significantly higher on the performance self-efficacy post-test, with an effect size reaching d = 1.18. AI use was positively associated with musical creativity in another study, with musical self-efficacy statistically mediating the association. In a poetry writing experiment, participants co-creating with AI had significantly higher creative self-efficacy than those editing AI drafts, and their expert creativity scores were no longer significantly lower than the pure human creation condition. Observed differences in self-efficacy appear to vary according to the degree of human control, choice, and process ownership. Record producers and recording engineers often use AI for audio restoration, demo production, and mastering, whereas songwriters, composers, and performers whose professional value depends more directly on original attribution may perceive greater replacement pressure. A mixed-method experimental study of 480 participants from the United Kingdom and Mexico found differences in receptivity to AI integration across cultural groups, while effects on creative self-efficacy were not uniformly positive.

The authors propose a substitution-scaffolding framework to make sense of these apparent inconsistencies. When AI reproduces identity-bearing properties of a singer’s voice, it may be experienced as a threat. When AI supports practice while preserving human control, it may support perceived competence and self-efficacy. Accessibility research presents yet another relationship between synthetic voice and identity: for users of augmentative and alternative communication, a well-matched synthetic voice may support self-representation and agency rather than threaten them, showing that voice identity can remain psychologically meaningful even when the audible voice is technologically mediated. Singing sits between these cases. Like other creative work, it raises questions of authorship and replacement, but the singer’s voice is also embodied, individually recognizable, and directly tied to professional performance, making findings from other domains informative about mechanisms such as agency and identity alignment without assuming they transfer directly.

The review is emphatic that current evidence cannot support the stronger claim that singers will fall into irreversible existential fear long-term. Broader deepfake psychological impact research remains early, methodologically heterogeneous, and lacks long-term tracking, with a critical scoping review of 22 empirical studies highlighting these limitations. Claims about permanent existential fear are currently speculative, not conclusions proven by longitudinal empirical evidence. The most urgent task for the field, the authors argue, is not to produce more purely theoretical stance papers but to conduct rigorous longitudinal quantitative research, ecologically valid voice-cloning scenarios, comparison conditions, and specialized tracking studies targeting professional singers, with direct measures of identity threat and occupational self-efficacy. Future studies could examine whether cloning vocal timbre produces the same response as reproducing more expressive characteristics such as phrasing, vibrato, dynamics, articulation, and prosodic variation, and whether professional singers are more sensitive than non-singers to subtle differences between human and AI-generated singing. Such research would clarify whether generative AI produces primarily short-term unease or contributes to more persistent changes in singers’ vocal identity, professional confidence, and occupational self-efficacy.

Subject of Research: Psychological and professional implications of generative AI voice cloning for singers

Article Title: When voices can be cloned: vocal identity threat and occupational self-efficacy among singers in the era of generative AI audio

Article References: Niu, J., Qi, Y., & Li, X. (2026). When voices can be cloned: vocal identity threat and occupational self-efficacy among singers in the era of generative AI audio. Frontiers in Psychology, 17, Article 1954386. https://doi.org/10.3389/fpsyg.2026.1954386

Image Credits: AI Generated

DOI: 10.3389/fpsyg.2026.1954386

Keywords: voice cloning, generative AI audio, singers, vocal identity threat, occupational self-efficacy, audio deepfakes, voice perception, music industry, AI anxiety, Frontiers in Psychology, voices, cloned

The motivation for the study stems from a methodological bottleneck. As voice assistants, synthetic narrators, and conversational agents become embedded in daily life, researchers increasingly need experimental stimuli in which speaker identity and prosodic style can be manipulated independently. If a scientist wants to test how listeners react to a confident versus a doubtful statement, the comparison is confounded when different human speakers deliver the confident and doubtful versions, because listeners may respond to the person rather than the prosody. Voice cloning promises a solution: train a model on one speaker’s voice, then generate new sentences carrying a target emotional or attitudinal tone. But this only works if the cloning system genuinely transfers prosody rather than flattening it into a generic synthetic delivery.

To test this, the researchers recruited ten native Mandarin speakers who each produced thirty sentences, fifteen drawn from geography statements and fifteen from trivia statements, in three distinct intonations: confident, doubtful, and neutral. This design yielded a rich corpus of human recordings spanning different sentence contents and prosodic intentions. For each speaker and each prosody, the team trained separate AI voice clones using either the geography recordings or the trivia recordings as training material. Each clone was then asked to generate all thirty sentences, producing a total of 2,700 utterances for analysis: 900 human recordings, 900 AI-generated utterances from geography-trained clones, and 900 from trivia-trained clones. Crucially, the training and generation sets were crossed, so a clone trained only on confident geography sentences still had to produce doubtful trivia sentences, forcing the system to generalize prosodic style to entirely novel content.

The analytical strategy combined machine learning classification with traditional statistical modeling. The researchers extracted 88 acoustic features using the extended Geneva Minimalistic Acoustic Parameter Set, or eGeMAPS, a standardized toolkit widely used in voice research and affective computing. They then trained gradient-boosted tree classifiers, a machine learning approach implemented through the XGBoost framework, to distinguish confident from doubtful prosody within each source of speech. Within a single source, whether human or AI, the classifiers achieved high accuracy, averaging 0.85, indicating that the acoustic signatures of confidence and doubt are robust and detectable in both natural and synthetic voices. More importantly, when classifiers trained on human speech were tested on AI speech, and vice versa, accuracy remained substantially above chance, averaging 0.65. This cross-source generalization demonstrates that the prosodic cues of confidence and doubt are encoded in a shared acoustic currency that transcends the boundary between human and machine voices.

Among the 88 features, one emerged as a consistently dominant contributor: spectral flux, a measure of how rapidly the frequency content of the speech signal changes over time. Spectral flux captures the crispness and dynamism of articulation, and confident speech, in both humans and their AI clones, exhibited higher spectral flux than doubtful speech. This makes intuitive sense. A confident speaker articulates with decisive energy, producing sharper spectral transitions, whereas a hesitant speaker tends toward softer, less defined acoustic edges. Notably, spectral flux showed no effect of speaker sex, which the authors interpret as evidence that it indexes prosodic style rather than stable anatomical differences between speakers. In other words, it is a marker of how something is said, not of who is saying it, which is precisely the kind of feature a stimulus designer would want to manipulate independently of speaker identity.

Linear mixed-effects models, a statistical framework that accounts for the nested structure of the data with multiple speakers, sentences, and prosodies, confirmed a coherent acoustic profile of vocal confidence. Across human and AI speakers alike, confident prosody was associated with higher spectral flux, a longer estimated vocal tract length, and lower fundamental frequency, the acoustic correlate of pitch, compared with doubtful prosody. The estimated vocal tract length, derived from formant frequencies, reflects how speakers shape their resonating cavities; a longer apparent vocal tract projects a larger, more authoritative body, echoing a well-documented literature on vocal size exaggeration in humans. That AI clones reproduced this constellation of cues, including the anatomical-sounding shift in apparent vocal tract size, suggests that modern cloning systems capture not only surface acoustics but the embodied gestalt of an attitudinal vocal stance.

The study also probed how prosodic information unfolds in time. Using time-resolved decoding of fundamental frequency, the researchers found that the distinction between confident and doubtful prosody was most reliably recovered in the late windows of each utterance, across human and AI sources alike. This late-emerging pattern aligns with the intuition that speakers often commit to an attitude as a sentence progresses, with final intonational contours sealing the impression of certainty or hesitation. However, a subtle asymmetry appeared: human speakers distributed their prosodic markers more broadly across the utterance, weaving cues of confidence and doubt throughout the signal, whereas the AI clones concentrated their prosodic differentiation more narrowly toward the end. This temporal signature may represent one of the remaining acoustic fingerprints distinguishing synthetic from natural expressive speech.

Indeed, the authors are careful to note that despite the impressive prosodic fidelity, AI-generated speech still forms a distinguishable acoustic population relative to human speech. The cross-source classification accuracy of 0.65, while well above chance, falls short of the 0.85 achieved within sources, indicating that machine and human voices are similar but not acoustically identical. This finding resonates with recent work showing that voice clones sound realistic but not yet hyperrealistic, and that human listeners and neural systems can, under some conditions, separate deepfake from genuine speaker identity. For the immediate purposes of psychological research, however, the residual human-machine gap may even be an advantage, since it allows researchers to verify that their synthetic stimuli behave acoustically as intended while retaining a measurable distinction from natural recordings.

The broader significance of the study lies in its methodological contribution. By demonstrating that voice cloning preserves prosodic style alongside speaker identity, Chen and Jiang provide empirical license for a new generation of controlled experiments on human-AI voice interaction. Researchers can now, with appropriate validation, generate stimulus sets in which the same voice delivers the same content with systematically varied confidence, doubt, or neutrality, isolating the causal impact of prosody on listener trust, memory, persuasion, and social perception. As AI voices become conversation partners, teachers, and companions, understanding how their vocal expressions of certainty are produced, perceived, and potentially distinguished from human ones becomes a scientific priority. This study shows that the tools for doing that science rigorously are already within reach, and that the line between human and machine vocal expression, while still detectable, is growing finer by the year.

Subject of Research: Whether AI voice cloning systems can reproduce human prosodic expressions of confidence and doubt

Article Title: Voice-cloning artificial-intelligence speakers can also mimic human-specific vocal expression

Article References: Chen, W., & Jiang, X. (2026). Voice-cloning artificial-intelligence speakers can also mimic human-specific vocal expression. Current Psychology, 45(17), Article 1490. https://doi.org/10.1007/s12144-026-09991-w

Image Credits: AI Generated

DOI: 10.1007/s12144-026-09991-w

Keywords: voice cloning, artificial intelligence, prosody, speech synthesis, vocal confidence, spectral flux, machine learning, human-AI interaction, acoustic analysis, eGeMAPS, fundamental frequency, Current Psychology