Three authors (SP, SM, JJ) independently conducted comprehensive searches of five electronic databases—PubMed, CINAHL, Cochrane Library, Web of Science, and Embase—covering literature published between June and August 2025. Search strategies were developed using combinations of terms such as “auditory rehabilitation,” “aural rehabilitation,” “audiologic rehabilitation,” “hearing therapy,” “listening therapy,” and “listening rehabilitation” combined with “speech recognition,” “voice recognition,” and “spoken recognition.” The full list of search terms is provided in Supplementary Table 1 (in the online-only Data Supplement).

Studies were eligible for inclusion if they met all of the following criteria: 1) included children, adults, or older adults with at least mild hearing loss (≥2 years of age); 2) employed self-directed auditory rehabilitation interventions, excluding non-speech-based auditory approaches; 3) reported post-training outcomes compared with pre-training outcomes (repeated-measures design) or with a control group (active or inactive); 4) reported at least one outcome measure based on behavioral or objective auditory tests, or on patient-reported functional or quality-of-life outcomes; and 5) used randomized controlled trials (RCTs), non-randomized controlled trials, cohort studies with control groups, or repeated-measures designs. Review articles, editorials, and grey literature were excluded.

Data from eligible studies were independently extracted by three authors (SP, SM, JJ) using a standardized, pilot-tested form. Extracted information included: 1) study identification details (first author, year of publication, journal, and title); 2) study characteristics (design, duration, location, original eligibility criteria); 3) participant characteristics (sample size, sex distribution, age); 4) intervention details (nature, duration, frequency); and 5) outcomes (types of outcome measures, effect estimates, follow-up duration, attrition/dropout rates, and any adjusted confounders). For studies with control groups, data from both intervention and control groups were extracted where available; for single-group studies, only intervention data were included. Discrepancies in data extraction were resolved by consensus with another author (WH). This process was intended to enhance reliability and minimize bias.

The risk of bias for RCTs was assessed using the Cochrane Risk of Bias 2 (RoB 2) tool [10], which evaluates five domains: bias arising from randomization, deviations from intended interventions, missing outcome data, outcome measurement, and selective reporting. Each domain was rated as low risk, some concerns, or high risk of bias. For non-randomized studies, including quasi-experimental trials, single-group pre-post designs, and natural experiments, we employed the ROBINS-I tool (Risk Of Bias In Non-randomized Studies of Interventions) [11], which evaluates seven domains: confounding, selection of participants, classification of interventions, deviations from intended interventions, missing data, measurement of outcomes, and selection of the reported result, with the overall risk categorized as low, moderate, serious, or critical. Two authors (SM, WH) independently conducted the assessments, and disagreements were resolved through discussion until consensus was reached.

Both qualitative synthesis and quantitative meta-analysis were performed. Statistical heterogeneity across studies was assessed using the I² statistic and Cochran’s Q test. For subgroup analyses, studies were stratified into two categories: children and adults/older adults. When substantial heterogeneity was observed (I²≥30%), pooled effect sizes were estimated using a DerSimonian-Laird random-effects model. Publication bias was evaluated by funnel plots and Egger’s regression test.

A total of 1,851 records were identified through database searches. After removal of 47 duplicates, 1,804 unique articles remained for title and abstract screening. Of these, 126 articles underwent full-text review, with 104 excluded for not meeting PICOS criteria (Table 1). Ultimately, 22 studies were included for qualitative and quantitative synthesis (Fig. 1).

The risk of bias assessments for the 22 included studies [12–33] are summarized in Table 2 (RCTs) and Table 3 (non-randomized studies). Among the 10 RCTs evaluated using the RoB 2 tool (Table 2), three studies were judged to have a low risk of bias, whereas seven were rated as having some concerns, primarily due to issues with the randomization process (e.g., quasi-randomization based on recruitment order) or the absence of blinding for self-reported outcomes.

For the 12 non-randomized studies evaluated with the ROBINS-I tool (Table 3), the overall risk of bias was generally higher. Three studies were classified as having moderate risk of bias, indicating acceptable methodological rigor for non-randomized designs [13,14,23]. Eight studies were rated as having a serious risk, largely due to confounding inherent in single-arm designs or self-selection bias (e.g., [19]). One case study [22] was judged to be at critical risk of bias; this was taken into account in sensitivity analyses and may justify excluding the study from meta-analytic pooling under certain conditions.

The included studies encompassed a broad range of participant ages (from preschool children to older adults), intervention modalities (various digital and computer-based auditory training programs), control conditions (active, passive, or none), outcome measures (behavioral, objective, and patient-reported), and study designs (RCTs, crossover trials, cohort studies, and pilot studies). Across studies, all interventions focused on self-directed auditory rehabilitation, typically delivered through home-based or gamified platforms, with substantial variations in both duration and intensity. Classification of the 22 included interventions identified three principal categories: 1) standard auditory training (standard AT), the most common approach, typically involving repetitive phoneme- or word-level identification tasks; 2) gamified AT, primarily observed in pediatric studies, in which digital toys such as Tiptoi [20] or serious-game platforms [21] were employed to enhance engagement; and 3) AT combined with counseling/education, mainly represented by Listening and Communication Enhancement (LACE)-based programs that integrated auditory training with communication strategy instruction. Further details are provided in Table 4.

A random-effects meta-analysis was conducted on four studies [12,20,24,29] of pediatric participants with hearing loss who completed self-directed auditory training. The primary outcomes were the pre-post mean differences or intervention-control differences on standardized speech recognition tasks. Pooled analysis yielded a statistically significant effect (Cohen’s d=1.32, standard error [SE]=0.12, 95% confidence interval [CI]: 1.08–1.56; p<0.05), indicating that auditory training conferred measurable improvements in speech recognition for children. Between-study heterogeneity was moderate, reflecting differences in study protocols and participant characteristics (Supplementary Table 2 in the online-only Data Supplement and Fig. 2).

A funnel plot showed no obvious asymmetry (Fig. 3), and Egger’s regression test yielded a non-significant regression intercept (−0.92, p=0.23). However, because only four studies were included (k=4), the statistical power of Egger’s test is limited, so these quantitative findings are reported for transparency but should be interpreted with caution; in this context, evaluation of publication bias relies primarily on visual inspection of the funnel plot, which did not suggest marked asymmetry.

Intervention durations ranged from several days to four weeks, with total training times from 5 to 16 hours. The meta-analysis indicates that even relatively brief, self-directed programs can yield clinically meaningful gains in speech recognition. Follow-up assessments available in three studies [18,24,28] confirmed that post-training gains were generally maintained for 1 to 3 months after the intervention, with no evidence of regression to baseline.

Eighteen studies [13–19,21–23,25–28,31–33] were pooled in a separate random-effects meta-analysis for adults and older adults. The aggregated effect size (Cohen’s d) was 0.55 (SE=0.05, 95% CI: 0.45–0.65), representing a moderate, statistically significant improvement in speech recognition following self-directed auditory training (Supplementary Table 3 in the online-only Data Supplement and Fig. 4). Substantial between-study heterogeneity was observed, consistent with the variation in program intensity, modality, and participant characteristics.

Visual inspection of the funnel plot, together with Egger’s regression test (intercept=2.62, p<0.001), revealed evidence of publication bias (Fig. 5), suggesting that published effect sizes may be overestimated. To assess the impact of this bias, a trim-and-fill sensitivity analysis was performed, imputing six potentially missing studies to the left of the mean to correct for funnel-plot asymmetry; the adjusted pooled effect size decreased to 0.42 (95% CI: 0.31–0.53), compared with the original estimate of 0.55 (95% CI: 0.45–0.65), yet remained statistically significant, indicating that the beneficial effect of self-directed auditory rehabilitation persists even after accounting for small-study effects.

Most interventions were delivered as 1–2 hour sessions, 3–5 times per week, over 4–12 weeks. Studies with the largest gains typically provided a cumulative training duration of at least 20 hours. Longer total duration and consistent engagement were associated with enhanced improvements, although there was some evidence of a plateau effect for programs extending beyond 8–12 weeks [32]. Follow-up assessments at 1–3 months post-training showed that both objective (speech and phoneme recognition) and patient-reported outcomes were maintained, confirming the lasting benefits of self-directed auditory rehabilitation.

This systematic review and meta-analysis confirm that self-directed auditory rehabilitation leads to significant and sustained improvements in speech recognition across the lifespan, benefiting both pediatric and older adult populations with hearing loss. In pediatric cohorts, the observed large effect size of Cohen’s d=1.32 (SE=0.12) in standardized speech recognition scores aligns with recent studies [12,24], which have also documented the effectiveness of digital home-based training for children. These findings bolster the emerging consensus that early, self-directed auditory intervention outside the traditional clinic setting can yield clinically meaningful gains in listening skills [8].

Among adult and older adult participants, the meta-analysis revealed a moderate but statistically robust improvement in speech recognition following self-directed rehabilitation, with a pooled effect size of Cohen’s d=0.55 (SE=0.05, 95% CI: 0.45–0.65). These effects were consistent across intervention platforms and generally persisted for up to 3 months post-training, underscoring that self-directed programs are both effective and durable for older individuals [2–4]. This observation is consistent with prior meta-analyses, reinforcing that adults and older adults can attain comparable auditory outcomes from autonomous digital interventions, and supporting their broader implementation in contemporary hearing healthcare [7].

A diverse range of digital and gamified platforms—including computer-based phoneme training, interactive games, and tablet applications—proved effective in delivering self-directed auditory rehabilitation. This is in line with previous literature indicating that computerized auditory training offers benefits similar to those of conventional clinic-based or clinician-led modalities for adult populations [3,16]. The wide variety of digital solutions has the potential to overcome major barriers such as accessibility and long-term adherence [4], although specific challenges remain for some older adults, particularly in regard to digital literacy. Such issues likely contribute to the between-study heterogeneity observed in effect sizes for older adult cohorts.

For adult and older adult individuals, the moderate effect size observed in this review is supported by earlier meta-analyses [2,15] and reinforced by independent studies demonstrating similar outcomes with computer-based auditory training. Furthermore, the persistence of benefits for at least three months post-intervention, as noted in this and other studies [5], attests to the potential for durable improvement through self-directed rehabilitation. However, some recent trials have failed to observe long-term gains in specific domains [18], such as complex sentence comprehension, which may be attributable to lower engagement levels or less intensive training protocols.

Intervention duration and intensity emerged as key factors in determining rehabilitation success. Programs totaling 20 or more hours of cumulative training consistently achieved the most substantial gains, a pattern consistent with the dose-response relationship reported in earlier reviews [7]. Still, there is evidence that extending the intervention beyond 8–12 weeks may result in a plateau effect [32], indicating that training regimens should be personalized, taking into account age, baseline abilities, and motivational factors.

While publication bias appeared minimal among pediatric samples according to funnel plot and Egger’s regression analyses, it is important to note that in the pediatric group (k=4), Egger’s regression has limited statistical power due to the small number of included studies. Thus, the non-significant result (intercept=−0.92, p=0.23) is reported for transparency but should be interpreted with caution, and the assessment of publication bias in this subgroup relies primarily on the visual inspection of the funnel plot. In contrast, publication bias was evident in the literature on adult and older adult individuals [23,25], suggesting that published effect sizes in older cohorts may be overestimated. Heterogeneity also remained substantial across studies, likely reflecting differences in intervention type, participant demographics, device use, and selected outcome measures. These insights underscore the necessity for transparent reporting of null or modest results and advocate for greater standardization in future research design and reporting.

Several limitations warrant consideration. There was considerable variability among included studies regarding intervention content, intensity, delivery platform, and outcome measures, which may compromise the comparability of results and the reliability of pooled effect sizes, especially for long-term outcomes. Despite rigorous methodology, a subset of studies displayed moderate-to-high risk of bias and evidence of publication bias in adult and older adult groups, which could potentially inflate effect estimates. The scarcity of studies with follow-up durations beyond three months also limits understanding of the long-term sustainability of observed benefits. Moreover, most interventions were tested in controlled research settings rather than real-world clinical environments, potentially restricting the generalizability of the findings. Participant pools were also not always representative of individuals with additional disabilities, severe to profound hearing loss, low digital literacy, or diverse cultural and linguistic backgrounds. Finally, the predominant focus on speech recognition means that important patient-reported outcomes—such as satisfaction, psychosocial well-being, and overall quality of life—remain insufficiently explored. Future research should emphasize methodological consistency, extended follow-up, comprehensive patient outcomes, and greater inclusion of underserved groups to further delineate and strengthen the evidence base.

Taken together, these findings support the integration of self-directed auditory rehabilitation as an adjunct to traditional clinician-led care. Recent international guidelines recommend broadening access to digital and autonomous interventions [3,13], especially in settings where specialist availability is limited [7,24]. The present evidence adds further impetus for such policy developments and highlights the need for ongoing research to refine intervention content, delineate subgroup-specific effects, and leverage emerging technologies—such as artificial intelligence and real-time feedback systems—to optimize therapeutic outcomes.