This study was conducted at Tianjin First Central Hospital, Tianjin, and included 29 patients with tinnitus (15 males and 14 females), aged between 25 and 75 years. All participants underwent audiometry and otoscopy at enrolment by an otolaryngologist, to ensure data accuracy and reliability. Pure-tone audiometry was performed for all participants across the full clinical frequency range of 125 Hz to 8 kHz. For descriptive purposes, the pure-tone average (PTA) was calculated using thresholds at 0.5, 1, 2, and 4 kHz, following standard clinical convention. Tinnitus pitch matching was conducted independently up to 8.5 kHz, ensuring accurate determination of individualized notch-center frequencies even for patients with high-frequency tinnitus. Normal hearing was defined as a PTA below 20 dB HL, while mild hearing loss was defined as a PTA between 21 dB HL and 40 dB HL. Patients with severe hearing loss were excluded to ensure adequate perception of TMNMT. Among the 29 participants, 12 had normal hearing and 17 had mild hearing loss. Exclusion criteria were as follows: (1) history of substance abuse, including psychiatric medications or alcohol; (2) history of psychiatric or neurological disorders; (3) history of significant head trauma; (4) pulsatile tinnitus; (5) diagnosis of Meniere’s disease, external and middle ear lesions, and posterior cochlear lesions; (6) severe hearing loss. Additionally, Tinnitus Handicap Inventory (THI) and Visual Analogue Scale (VAS, including Loudness and Annoyance) were employed to evaluate the impact of tinnitus on participants’ daily lives. The study protocol was approved by the Tianjin First Central Hospital’s ethics committee (2020N114KY). Patients provided written permission for the use of their anonymized clinical data for research purposes, and no study-specific procedures or prospective enrollment were performed.

Tinnitus pitch and loudness matching were performed using a portable tinnitus treatment device. For pitch matching, 188 pure-tone stimuli covering the frequency range of 125 Hz to 8.5 kHz were presented in 1/30-octave steps, as described previously (Zhu and Gong, 2023). Participants listened to a series of tones and compared each with their perceived tinnitus until a tone that best matched their tinnitus pitch was identified. Once the tinnitus pitch was determined, loudness matching was conducted using the same device. The sound level was adjusted in 1 dB sound pressure level (SPL) increments until participants indicated that the loudness matched the perceived intensity of their tinnitus (Kim et al., 2016)^. This stepwise procedure allowed for fine adjustment to closely match each participant’s subjective tinnitus loudness. For bilateral and central tinnitus, tinnitus pitch and loudness were estimated separately for each ear.

This study was based on clinical data obtained during routine tinnitus assessment. As part of the standardized evaluation protocol, patients underwent EEG recording in a single session that consisted of three sequential phases. First, a 3-min resting-state EEG was recorded with eyes closed, followed by a 2-min eyes-open rest period. Subsequently, a 10-min tailor-made notched music stimulation (TMNMT) was presented at an individually adjusted level approximately 10 dB above the matched tinnitus loudness. For each patient, the audio stimulus was digitally notched at the tinnitus center frequency to enhance lateral inhibition at the notch edges. To improve stimulation efficiency in higher frequency bands, the spectral envelope was flattened by reallocating energy from lower to higher frequencies as described by Teismann et al. (2011). The TMNMT material consisted of naturalistic environmental sounds (e.g., rain, streams, waves), which are typically perceived as relaxing (Strauss et al., 2017). Audio was delivered binaurally via insert earphones. During TMNMT, EEG was recorded continuously with eyes closed (duration: 10 min).

After TMNMT, a further 5-min eyes-closed EEG was recorded to capture residual inhibition (RI). Immediately after TMNMT stimulation and EEG acquisition, patients rated tinnitus loudness and annoyance using a visual analog scale (VAS), and clinicians documented the depth and duration of RI based on patient reports. Residual inhibition (RI) was assessed immediately after the 10-min TMNMT stimulation. RI duration was obtained by asking patients to report the moment at which their tinnitus returned to its pre-stimulation baseline. Clinicians recorded the elapsed time from the end of stimulation to the subjective recovery of tinnitus perception. Both RI depth and RI duration were documented as part of the routine clinical evaluation. The effectiveness of TMNMT was further evaluated using a Likert scale ranging from −5 to 5, where −5 indicated complete suppression of tinnitus, 0 indicated no change, and 5 indicated a doubling of perceived loudness.

Classification of RI outcomes followed the criteria described by Hu et al. (2021). Patients were categorized as “RI positive” if they exhibited an averaged maximum RI depth of −4 or −5, corresponding to almost complete or complete tinnitus suppression. Those who did not meet this criterion—including individuals with minimal suppression (RI depth > − 1) or partial suppression (−4 < RI depth ≤ − 1)—were classified as “RI negative.”

Pre-recording controls: Participants were instructed to obtain adequate sleep and to abstain from smoking, caffeine, and strenuous physical or mental activity for at least 24 h prior to EEG recording.

Setting and equipment: EEG was recorded in a sound-attenuated and electromagnetically shielded room using an EGI GES 400 system (Electrical Geodesics, Inc., USA) with a 64-channel HydroCel sponge cap pre-soaked in potassium chloride (KCl) solution to improve electrode–skin conductance. Electrode impedances were maintained below 50 kΩ throughout data collection.

Acquisition parameters: Signals were acquired with Net Station Acquisition 5.4.3-R, using Cz as the online reference, at a sampling rate of 1 kHz, and with a 0.1–30 Hz band-pass. Electrodes were positioned according to the international 10–20 system.

Participant instructions during recording. During EEG acquisition, participants were asked to minimize facial, ocular, and neck movements. In eyes-open segments, a central white fixation cross was presented on the display, and participants were instructed to maintain fixation; eyes-closed segments were recorded while participants remained relaxed and still.

All electroencephalogram (EEG) data were processed and visualized using the EEGLAB v2021.0 toolbox in MATLAB R2023b (Delorme and Makeig, 2004). The data preprocessing pipeline was conducted as follows:

Filtering implementation: A 1 Hz high-pass filter, a 30 Hz low-pass filter, and a 50 Hz notch filter were applied to mitigate electrode drift, muscle noise, and power line interference, respectively;

This preprocessing pipeline effectively attenuated non-neuronal artifacts and enhanced data quality prior to subsequent analytical procedures.

Power spectral analysis were computed for each electrode and then averaged across all electrodes to obtain a global spectral estimate. The Spectral Power Ratio (SPR) was derived from this global average, as the goal of the present study was to assess overall oscillatory modulation rather than spatially localized effects. No electrode-specific or topographical analyses were performed. Power spectrum analysis was then performed using the Welch method to obtain the power spectrum within the frequency range of 1–30 Hz, where the vertical axis represents power (in μV²/Hz) and the horizontal axis represents frequency (in Hz). Four classic EEG frequency bands were defined as follows: delta (1–4 Hz), theta (4–8 Hz), alpha (8–13 Hz), and beta (13–30 Hz). The power of each frequency band was calculated by integrating the power spectrum curve over the corresponding frequency range (i.e., calculating the area under the curve within the specific band). The total power was determined as the sum of the powers of the four bands (equivalent to integrating the power spectrum over the entire 1–30 Hz range). Finally, the ratio of each band’s power to the total power was computed using the formula: (Power of target band / Total power) × 100%, expressed as a percentage. We can analyze changes in the activity intensity of a specific frequency band by means of the spectral power ratio (SPR).

Statistical analysis was conducted using SPSS 25.0 software. The quantitative data that conforms to a normal distribution is represented as x ± SD. Due to the small number of experimental samples, non-parametric tests were used in the SPR analysis (Mann and Whitney, 1947). Band-limited SPR (delta, theta, alpha, beta) across time (pre, during, post) was tested using the Friedman test, with post-hoc Wilcoxon signed-rank tests (pre vs. during, pre vs. post, during vs. post), and correction effects were applied to multiple comparisons of false discovery rate (FDR). Linear regression and correlation analysis were performed using GraphPad Prism 9.0 software, with statistical significance set at p < 0.05. Continuous variables were analyzed using t-test or Mann–Whitney U test, while categorical variables were analyzed using chi square test. One way analysis of variance was used to compare the clinical differences between RI positive and RI negative groups. Using binary logistic regression analysis to study the factors affecting RI positivity rate. RI outcomes were summarized as depth (ordinal scale) and duration (seconds).

Table 1 summarizes the baseline characteristics of the study cohort, including demographic information, tinnitus-related clinical measures (such as tinnitus duration, laterality, loudness, and matched frequency), audiometric thresholds, and other routine baseline assessments collected prior to TMNMT.

A paired-sample t-test of VAS scores before and after TMNMT stimulation showed VAS-L (t = 6.648, p < 0.001) was significantly reduced after treatment and VAS-A (t = 5.446, p < 0.001) (Figure 1).

Due to the deviation of SPR value distribution from normality, Friedman test analysis was performed on the results of the subjects before, during, and after TMNMT stimulation (Figure 2). Figure 2A displays the grand-average power spectrum ratio (0–30 Hz) across all scalp electrodes for the pre-, during-, and post-stimulation conditions, providing an overview of the frequency-dependent SPR profile before examining band-specific differences. The results showed a significant increase in SPR in the δ and θ frequency bands, and a significant decrease in SPR in the α frequency band. Wilcoxon sign rank test was then used for pairwise comparison between groups. After FDR correction, the SPR of delta band (during-stim vs. pre-stim, p = 0.011; post-stim vs. pre-stim, p = 0.011; Figure 2B) and theta band (post-stim vs. pre-stim, p = 0.033; Figure 2C) were significantly increased, alpha band (during-stim vs. pre-stim, p = 0.029;post-stim vs. pre-stim, p < 0.001; Figure 2 D) was significantly decreased. No significant changes were observed in the beta band (during-stim vs. pre-stim, p = 0.606; post-stim vs. pre-stim, p = 0.854; Figure 2E). Besides, natural logarithmic transformation (ln transformation) was performed on the SPR value to improve its normality and used for two factor analysis of variance. The converted data was verified to follow a normal distribution (Shapiro Wilk test, p > 0.05). The frequency band (delta, theta, alpha, beta) was regarded as a within-subject factor and group (pre-stim, during-stim, post-stim) as a between-subject factor. From an analysis of the waveforms, Figure 2A showed the spectral power ratio (SPR) of whole brain. The results of SPR showed that the main effect of frequency band factor (F = 54.96; p < 0.0001) and frequency band Group interaction effect (F = 7.999; p < 0.0001) were significant. The main effect of group factor was not significant (F = 2.001; p = 0.1482) (Table 3).

This study examined the correlation between the changes in VAS scores (ΔVAS) before and after TMNMT stimulation and THI scores (before treatment), It was observed that there was a statistical significance in the increase of ΔVAS as THI score increased (Figure 3A). Further correlation analyses were conducted on the RI duration and ΔVAS, It was observed that there was a statistical significance in the increase of ΔVAS as the RI duration increased (Figure 3B).

Previous studies have shown that tinnitus is often associated with abnormal activity in central brain regions, and resting state spectral features can serve as a biological indicator of subjective tinnitus (Moazami-Goudarzi et al., 2010). Most prior work has quantified spectral changes using power spectral density (PSD) (Zhu and Gong, 2023); or absolute band power (Vanneste and De Ridder, 2016; Kyong et al., 2024; Huang et al., 2021). Because PSD/absolute power is sensitive to inter-individual factors (e.g., skull conductivity, electrode impedance), absolute band powers vary widely across participants, which can obscure group-level patterns. The Relative Power, which indicates the ratio of the power of a frequency band to the total band power, could be used to reduce this problem (Ahn et al., 2013; Dohrmann et al., 2007; Güntensperger et al., 2019; Jensen et al., 2023).

Although the SPR definition varies across studies (e.g., (α + β)/(δ + θ), α/δ, θ/β), these ratio-based metrics share the same rationale of normalizing band-limited activity to reduce inter-individual variability and to emphasize relative spectral redistribution across bands. Prior electrophysiological work has demonstrated that the balance between slow-wave activity (δ, θ) and faster rhythms (α, β) reflects large-scale changes in cortical excitability and network dynamics (Steriade, 2006; Nunez, 1981; Benoit et al., 2000; Uchida et al., 1991). In tinnitus research, characteristic patterns such as increased δ–θ activity accompanied by reduced α power have been repeatedly observed (Moazami-Goudarzi et al., 2010; Weisz et al., 2005), suggesting that changes in the relative contribution of slow versus fast oscillations may better capture the underlying pathophysiology than absolute power alone. Spectral ratio metrics have also been shown to provide sensitive indicators of physiological state transitions in cognitive and sensory domains (Ahn et al., 2013; Puma et al., 2018; Gevins, 1997). Accordingly, the SPR formulation adopted here serves as a normalized index representing the proportion of slow-wave activity within the total spectral power, thereby highlighting global shifts in inhibitory–excitatory balance that may occur during and after TMNMT.

In the present study, spectral power ratio (SPR) was selected as the primary EEG metric to characterize oscillatory changes associated with TMNMT-induced residual inhibition. Unlike conventional PSD-based band power, which reflects absolute or regional power changes, SPR provides a normalized measure of relative spectral redistribution across frequency bands (Kopčanová et al., 2024; Zawiślak-Fornagiel et al., 2023). This ratio-based approach is particularly suitable for heterogeneous clinical populations, such as tinnitus patients, in whom inter-individual variability in absolute EEG power can be substantial. By normalizing each frequency band to the total broadband power, SPR emphasizes cross-band modulation and global excitatory–inhibitory balance rather than absolute power magnitude. Accordingly, the present EEG findings should be interpreted as reflecting relative changes in spectral composition rather than absolute increases or decreases in specific frequency bands.

Within this framework, we observed a frequency-specific pattern characterized by increased delta and theta SPR accompanied by reduced alpha SPR during and after TMNMT exposure. This pattern is broadly consistent with models of tinnitus-related thalamocortical dysrhythmia, in which alterations in low-frequency and alpha-band activity are thought to reflect changes in inhibitory–excitatory balance within auditory and non-auditory networks. Importantly, the observed modulation occurred on a short timescale and was temporally aligned with behavioral residual inhibition, suggesting that TMNMT may transiently reshape large-scale oscillatory dynamics rather than inducing sustained changes in absolute band power.

For completeness, we additionally present a conventional PSD-based band-power analysis in the Supplementary Material. As shown in the supplemental figure, absolute band-power measures yield a partially different pattern compared with the SPR-based results. This divergence does not indicate inconsistency, but rather reflects the fact that PSD and SPR capture complementary but non-identical aspects of neural oscillatory activity. Whereas PSD emphasizes absolute power changes within specific frequency bands, SPR highlights relative spectral redistribution across bands. Accordingly, the supplemental PSD analysis is intended to illustrate the metric-dependent nature of spectral interpretation, while all primary conclusions of the present study are based on SPR.

This study found that, relative to pre-stimulation, δ-band SPR increased significantly both during TMNMT and in the 10-min post-stimulation period. As part of the slow-wave range, δ oscillations characteristically emerge in cortical regions lacking thalamocortical afferent input (Steriade, 2006). Low-threshold spike (LTS) bursts in thalamic nuclei—triggered when neurons become hyperpolarized because of deafferentation or excessive inhibition—can amplify cortical slow waves (Weisz et al., 2007). Fundamentally, δ activity reflects the oscillatory dynamics of deafferented or deprived neural networks (Llinás et al., 2005), and fluctuations in oscillatory EEG power generally index changes in population-level neuronal synchrony (Nunez, 1981; Bickford et al., 1979). In tinnitus, increases in δ and θ power accompanied by decreases in α power are associated with the presence and severity of the percept; δ enhancement typically co-occurs with α reduction—a pattern consistent with the present findings (Uchida et al., 1991; Benoit et al., 2000; Weisz et al., 2005). In studies evaluating electrophysiological changes associated with improvement in tinnitus perception via auditory coordinated reset (CR) stimulation (Tass et al., 2012), pretreatment pathological delta- and gamma-band activity normalized only in treatment responders (Adamchic et al., 2014), suggesting that thalamocortical dysrhythmia (TCD) may be causally linked to tinnitus. Consistent with this, a TMNMT intervention study reported significant increases in relative δ- and θ-band power in the treatment group, suggesting central cortical remodeling that may alter thalamocortical connectivity and, in turn, tinnitus-related neural networks (Zhu and Gong, 2023)^. δ-band alterations are further linked to functional connectivity within tinnitus-related networks, including interactions between the auditory cortex and limbic structures (Adamchic et al., 2014; Vanneste and De Ridder, 2011). Viewed through the lens of thalamocortical dysrhythmia, TMNMT may suppress aberrant fast activity (e.g., γ range) and shift thalamocortical rhythms from pathological high-frequency dominance toward physiological slow-wave dominance, thereby increasing the proportion of δ-band power.

In this study, we observed a significant increase in theta-band SPR after 10 min of TMNMT stimulation. This pattern aligns with Gong et al. (Zhu and Gong, 2023)^, who likewise reported a significant increase in theta-band PSD in the TMNMT group. Note that SPR (spectral power ratio) indexes relative band power—i.e., theta power normalized to total or reference-band power—whereas PSD reflects absolute power; despite these different scales, both metrics are sensitive to band-limited activity, so an increase in theta SPR is expected to co-vary with increases in theta PSD when the normalization is stable. Patho-physiologically, tinnitus has been attributed to bottom-up deafferentation and/or deficits in top-down noise cancellation. Both mechanisms are thought to alter thalamocortical signaling and give rise to thalamocortical dysrhythmia (TCD). Under deafferentation, TCD is characterized by a resting-state slowing from alpha to theta activity together with enhanced surrounding gamma activity, yielding persistent theta–gamma cross-frequency coupling (De Ridder et al., 2015; Weisz et al., 2007; De Ridder et al., 2011). This framework has been corroborated by EEG, MEG, and intracranial recordings (Weisz et al., 2007; Llinás et al., 2005; De Ridder et al., 2011; van der Loo et al., 2009; Llinás et al., 1999). Information streams linked to tinnitus, pain, movement, and emotion conveyed by high-frequency rhythms (beta/gamma) can be nested within theta via cross-frequency coupling (De Ridder and Vanneste, 2014). Against this background, the post-TMNMT elevation in theta power may index a transient reorganization of neuronal synchrony that could disturb the persistent theta–gamma coupling underlying the tinnitus percept (Weisz et al., 2007). It may also reflect a temporary rebalancing between excitatory and inhibitory neuronal ensembles, in line with observations during residual inhibition (Kahlbrock and Weisz, 2008). Additionally, theta oscillations support memory retrieval and large-scale network synchronization (Eggermont and Tass, 2015); thus, the enhanced theta activity observed here may suggest activation of Para hippocampal–auditory cortical connections involved in retrieving auditory information that is missing due to deafferentation (De Ridder et al., 2015).

In our study, alpha-band SPR decreased significantly both during TMNMT and after the stimulation period. This pattern aligns with Gong et al. (Zhu and Gong, 2023), who likewise reported a significant reduction in alpha-band PSD in the TMNMT group. Functionally, alpha oscillations are widely interpreted as an index of cortical inhibition: elevations in alpha power are observed under fatigue and are thought to downregulate processing of external information to limit cognitive load (Käthner et al., 2014; Puma et al., 2018; Gevins, 1997; Lou et al., 2025; Lorist et al., 2000). Conversely, lower alpha power is associated with active neural processing and better perceptual performance (Lorist et al., 2000; Gevins, 1997)^, whereas higher alpha indicates active suppression in task-irrelevant regions (Lorist et al., 2000; Gevins, 1997). In tinnitus, increased alpha1、alpha2 activity has been linked to greater tinnitus-related distress (Vanneste et al., 2010). Against this background, the TMNMT-induced reduction in alpha power observed here may index a normalization of cortical dynamics—i.e., reduced need for tonic inhibitory control—allowing more efficient sensory–attentional processing and, potentially, less attentional capture and distress associated with the tinnitus percept.

In this study, the β-band spectral power ratio (SPR) showed a downward trend 10 min after TMNMT stimulation, although the change did not reach statistical significance. Rather than claiming full consistency with prior work, we note that Zhu and Gong (2023) reported a non-significant reduction in β-band power spectral density (PSD) 5 min after TMNMT; despite differences in time points (5 vs. 10 min) and metrics (PSD vs. SPR), the effects were directionally similar. Converging evidence links anxiety traits, frontoparietal β activity, and hyperacusis (Inguscio et al., 2024). In the resting state, patients with tinnitus have higher power in the β band than healthy controls, suggesting that this may be related to fatigue and emotional effects produced by tinnitus (Vanneste et al., 2010; Meyer et al., 2014; Meyer et al., 2017). And greater distress has been associated with increased β activity (Jokić-begić and Begić, 2003; Begić et al., 2001). Against this background, the observed post-TMNMT reduction in β-band SPR—albeit nonsignificant—may indicate a transient modulation of neural processes related to distress and anxiety, suggesting that TMNMT could help alleviate the perceived burden of tinnitus. Larger samples and longer intervention windows are needed to confirm the effect size and clinical relevance.

Subjective tinnitus is a common clinical condition frequently accompanied by anxiety, depression, sleep disturbance, and autonomic dysfunction; in severe cases, maladaptation may lead to self-harm or suicidal ideation. Sound-based therapy can alleviate the adverse experience of tinnitus and, to some extent, reduce perceived loudness. In this retrospective study, we observed that a subset of patients exhibited residual inhibition (RI) after tailor-made notched music training (TMNMT), suggesting potential therapeutic benefit. Our aim was to summarize RI outcomes among patients of this project and to conduct a preliminary analysis of factors associated with RI in response to TMNMT, to inform future clinical application.

We included 29 patients in total; 20 patients (69%) showed a positive RI response, whereas 9 (31%) did not. Univariate analyses indicated that the RI-positive group had higher baseline Tinnitus Handicap Inventory (THI) scores and a greater proportion with symptom duration exceeding six months. In binary logistic regression, baseline THI was not a statistically significant predictor; however, the odds ratio for THI was greater than 1, suggesting a trend toward a higher RI-positive rate among patients with moderate–to–severe tinnitus compared with those with mild tinnitus. Prior studies have similarly reported that greater baseline tinnitus severity predicts better response to sound-based interventions. For example, Theodoroff et al. (2014) found that patients who perceived their tinnitus as more severe had approximately 3.8-fold higher odds of treatment effectiveness relative to those with milder symptoms, and Cai et al. (2017) reported that each 1-point increase in baseline THI was associated with a 4% increase in the probability of therapeutic benefit with masking therapy. To mitigate small-sample effects in our cohort, we categorized patients by THI into mild versus moderate–to–severe (often corresponding to compensated vs. decompensated tinnitus, respectively). We observed a higher treatment response in the decompensated (moderate–to–severe) group than in the compensated (mild) group. Collectively, these findings are consistent with the notion that greater tinnitus severity may be associated with better response to TMNMT therapy.

Having shown that TMNMT can elicit RI in a proportion of patients and having preliminarily explored potential correlates of RI, our results should be interpreted with caution given the limited sample size. Future work with larger cohorts and more comprehensive covariate collection will be valuable to derive more reliable estimates and to delineate patient characteristics most likely to benefit from TMNMT, thereby providing practical guidance for clinicians.