Music modulates emotional and physiological states by activating multiple brain regions. Research demonstrates its ability to regulate the limbic system, particularly the amygdala and hippocampus—key structures in emotional processing (Koelsch, 2014). Additionally, uplifting music triggers dopamine release, engaging the brain’s reward circuitry (Chanda and Levitin, 2013). From a physiological perspective, soothing music has been shown to reduce heart rate, blood pressure, and cortisol levels, thereby attenuating stress responses (Thoma et al., 2013; Jiang et al., 2016).

A critical psychoacoustic mechanism is rhythmic entrainment, the synchronization of internal neural oscillations with external acoustic rhythms (Thaut and Hoemberg, 2014). According to this theory, when individuals listen to music with specific tempi (e.g., 60–80 beats per minute), neural oscillations associated with relaxation (e.g., alpha waves) align with the musical rhythm (Ding et al., 2025). This synchronization shifts brain activity from a tense to a calmer state, serving as a core pathway for music’s regulatory effects on emotion and physiological arousal. The AI system in this study leverages this principle by dynamically adjusting musical parameters (e.g., tempo) via biofeedback to enhance entrainment effects.

Rooted in ancient Chinese music psychology, five-tone theory correlates the five tones (Gong, Shang, Jue, Zhi, and Yu) with five elements (earth, metal, wood, fire, and water), five visceral organs (spleen, liver, lung, heart, and kidney), and five emotional states (pensiveness, grief, anger, joy, and fear; see Table 1). It posits that specific musical modes can target corresponding organ systems and modulate associated emotions (Yang et al., 2021).

It is critical to bridge the conceptual gap between TCM theory and modern psychophysiological measurement. In TCM, the “Jue” tone is linked to “liver qi” and the emotion of “anger/anxiety”. From a modern psychophysiological perspective, anxiety states are operationalized as a dysregulation of the autonomic nervous system (ANS), characterized by heightened sympathetic dominance (the “fight-or-flight” response) and reduced parasympathetic vagal tone (the “rest-and-digest” system). Therefore, this study hypothesizes that the TCM concept of “soothing the liver” (疏肝) to alleviate anxiety should manifest physiologically as a rebalancing of the ANS. We selected Heart Rate Variability (HF-HRV) as a robust marker of parasympathetic activity and Electrodermal Activity (SCL) as a marker of sympathetic arousal to serve as objective, quantifiable proxies for the targeted anxiolytic effect.

Despite its systematic framework, the theory’s foundational hypothesis—that specific sound frequencies “resonate” with target organs or their neural representations—remains philosophical. Recent studies have begun examining its mechanisms through modern medical and acoustic lenses (Liu et al., 2014). This study builds on that premise, using AI to generate music within theoretically specified frequency ranges and employing psychophysiological metrics to empirically test the ancient theory.

Emerging evidence suggests that combining AI with biofeedback enables an effective approach to personalized music therapy (Fedotchev et al., 2014; Li and Xiong, 2016; Rashmi and Shantala, 2024). AI algorithms can generate music based on users’ emotional labels or physiological signals, while biofeedback establishes a closed-loop system that continuously monitors intervention efficacy and dynamically adjusts musical parameters. For instance, when the system detects increased sympathetic nervous system activity—indicating heightened tension—it may automatically slow the musical tempo or simplify melodic complexity to guide the user back to a relaxed state (Şahin, 2025).

Current research has explored using AI to replicate established relaxation music styles (Kant and Sharma, 2023) or to develop more precise emotion-recognition models, such as those based on electroencephalogram (EEG) signals, laying the groundwork for truly responsive systems (Zhong et al., 2022; Jiao, 2025). However, few studies have employed AI as a “translation” and “validation” tool to systematically examine comprehensive traditional medical frameworks like the five-tone theory. The innovation of this study lies in leveraging AI as a bridge to transform this traditional qualitative theory into a quantifiable, testable, and personalized intervention, while empirically evaluating its efficacy.

We recruited 40 university students (20 males, 20 females; mean age = 21.5 ± 2.2 years) through campus announcements and social media platforms.

Inclusion criteria: (1) Aged 18–28 years; (2) Hamilton Anxiety Rating Scale (HAMA) score ≥14, indicating at least mild anxiety (Ferguson et al., 2023; Hamilton, 1959); (3) No self-reported hearing impairments; (4) No systematic psychological or pharmacological treatments for anxiety in the past month.

Exclusion criteria: (1) History of severe mental disorders (e.g., schizophrenia, bipolar disorder); (2) Strong music aversion or amusia.

We implemented a single-group pre-test/post-test quasi-experimental design to evaluate the immediate effects of AI music intervention (see Figure 1 for study flowchart). The experimental protocol comprised the following steps:

All statistical analyses were performed using Python (v3.13) with the statsmodels library. To account for intra-individual variability from repeated measurements, as recommended, we employed linear mixed-effects models (LMM). For each of the three models, the outcome measure (HAMA score, HF-HRV, or SCL) was entered as the dependent variable. Time (pre-test vs. post-test) was entered as the fixed effect, and Participant was entered as a random effect (random intercept). Effect sizes were calculated using Cohen’s d. The significance threshold (α) was set at 0.05 for all statistical tests.

The study protocol was approved by the Institutional Review Board of the School of Law at Xuchang University. All participants provided written informed consent after receiving complete disclosures regarding study procedures, potential risks, and benefits. To ensure confidentiality, all collected data were anonymized and de-identified prior to analysis.

Participants demonstrated significant reductions in HAMA scores from pre-intervention (M = 18.2, SD = 2.5) to post-intervention (M = 11.5, SD = 2.1). The linear mixed-effects model confirmed this decrease was statistically significant (t(78) = −21.84, p < 0.001), with a large effect size (Cohen’s d = 1.95) (Boucsein, 2012), indicating substantial clinical improvement in subjective anxiety experiences following the intervention.

Physiological data aligned with psychological reports (Figure 2). Post-intervention, participants exhibited a pronounced shift toward relaxation in autonomic nervous system (ANS) activity:

For parasympathetic nervous system activity (measured by HF-HRV), the mean values significantly increased from 450.5 ms² (SD = 150.2) at baseline to 675.8 ms² (SD = 180.4) post-intervention (t(78) = 10.99, p < 0.001).

For sympathetic arousal (measured by SCL), mean values decreased from 4.1 μS (SD = 1.1) at baseline to 2.5 μS (SD = 0.8) post-intervention (t(78) = −11.71, p < 0.001).

This study examined the efficacy of an AI music intervention system integrating Traditional Chinese Five-Tone Theory with modern biofeedback technology. The results strongly support our primary hypothesis that this novel intervention can simultaneously alleviate anxiety both psychologically and physiologically in undergraduates with moderate anxiety. The significant reduction in HAMA scores (effect size d = 1.95) demonstrates both statistical and clinical significance, with improvement levels comparable to brief psychotherapy or pharmacological interventions (Cohen, 1988).

More importantly, the subjective improvements were corroborated by objective physiological data: substantial increases in HF-HRV and decreases in SCL indicate a clear physiological mechanism involving a beneficial shift in autonomic nervous system balance—from sympathetic (“fight-or-flight”) dominance to parasympathetic (“rest-and-digest”) predominance (Lu et al., 2021). This psychophysiological concordance provides preliminary evidence for the intervention’s potential efficacy beyond subjective reports, although placebo effects cannot be excluded by the current design. This finding represents the study’s primary contribution.

It is important to interpret the large effect size observed for HAMA scores (Cohen’s d = 1.95) with extreme caution. While encouraging, this value is likely inflated due to the single-group pre-post design, which does not control for factors such as expectancy, attention, or placebo effects. Such large effects are not uncommon in preliminary, uncontrolled intervention studies.

From the perspectives of music psychology and interdisciplinary research, this study carries multiple significant implications. First, it provides a preliminary quantitative exploration of the ancient five-tone theory’s modern applicability and potential scientific basis. The observed anxiolytic effects of Jue-tone music offer preliminary quantitative evidence for the traditional “liver-anger/anxiety-Jue” correspondence supporting the “acoustic resonance hypothesis” that specific musical structures or acoustic parameters may preferentially interact with particular physiological systems or their neural representations—a promising avenue for further scientific exploration.

Furthermore, this study demonstrates AI’s potential as a “computational bridge.” Our designed AI system translates an abstract, qualitative traditional medical theory into executable computer parameters (e.g., pitch patterns, tempo ranges, harmonic progressions) that can be modulated by physiological signals, thereby connecting conceptual traditional wisdom with measurable modern psychophysiology. This methodology opens new possibilities for examining other traditional medical theories (e.g., Indian Raga therapy) or psychoacoustic hypotheses (Thayer et al., 2012).

The practical implications of this research are substantial. This AI-driven, biofeedback-enhanced system represents a highly personalized, scalable, and non-invasive digital mental health intervention tool. Unlike conventional one-size-fits-all relaxation music playlists, it dynamically adjusts auditory experiences based on users’ real-time physiological states, potentially revolutionizing treatment adherence and efficacy. This technological paradigm holds significant application potential in three key areas:

While the results are promising, this study has several significant limitations that must be acknowledged and addressed in future research.

The primary limitation is the single-group pre-post design, which is insufficient to establish causal efficacy. He observed changes cannot be definitively attributed to the AI-driven music intervention itself, as they could be the result of non-specific factors such as expectancy, simple attention from researchers (i.e., Hawthorne effects), or spontaneous fluctuations in anxiety. Therefore, the claims of this paper are exploratory and generative, not confirmatory. Future studies must employ rigorous randomized controlled trial (RCT) designs. As suggested, this should include (1) an active control group (e.g., listening to generic relaxation music or a sham intervention like white noise), and (2) a theory-specific control (e.g., non-adaptive Jue-tone music, or adaptive music based on a different tonal theory) to isolate the specific effects of the Five-Tone framework and the biofeedback loop.

The current study only examined immediate intervention effects, leaving the long-term outcomes unknown. Longitudinal follow-up studies are needed to determine whether repeated, prolonged interventions can produce sustained anxiety reduction and whether effects diminish over time. Additionally, the relatively homogeneous sample of university students limits the generalizability of findings. Future validation should include more diverse populations: clinically diagnosed anxiety patients, adults across different age groups, and individuals from varied cultural backgrounds (Aalbers et al., 2017; Witusik and Pietras, 2019).

In conclusion, this investigation focused exclusively on validating the Jue tone within the five-tone theory. Subsequent research should systematically test other theoretical postulates—such as using Zhi tone for alleviating low mood or Yu tone for fear reduction—while also examining combinatorial effects of different tones. This systematic approach will ultimately yield a more comprehensive and precise AI-powered personalized treatment model based on five-tone theory.