40 community-dwelling older adults (51 - 75 years, mean age ± SD, 59.1 ± 5.9) from Guiyang City were recruited and randomly divided into a training group and a control group. All participants were cognitively healthy, with Mini-Mental State Examination (MMSE) scores above 26, and were not taking any medications with central nervous system effects. Before the experiment, all participants were given informed content approved by the Medical Ethics Committee of the Second Affiliated Hospital of Guizhou University of Traditional Chinese Medicine.

All participants conducted the experiment in a dimly lit, electrically shielded, and sound-attenuated room at the Psychology Laboratory of Guizhou University of Traditional Chinese Medicine, China. The presentation of stimuli and the collection of behavioral data were managed using E-Prime 3.0 (Psychology Software Tools, Inc., Pittsburgh, PA, USA) during both the training and assessment tasks.

The audiovisual perceptual training consisted of five separate sessions completed over five consecutive days. On the first day, before training, each participant in both the training and control groups was instructed to perform the SJ task and audiovisual detection task, referred to as the pre-training assessment (Pre-test, Figure 1 ). In the following five days, the training group followed a personalized training schedule, while the control group received no training and remained isolated from the training group. On the final day, after training, participants in both groups performed the same assessment tasks as on the first day, referred to as the post-training assessment (Post-test, Figure 1 ). All experimental tasks were conducted under the same environmental conditions in our psychology laboratory to minimize the influence of unrelated factors.

For the perceptual training task (see Figure 2 ), adapted from previous studies, the visual stimulus (V) was a 10% contrast, 1.5 spatial frequency vertical Gabor with a 5˚ visual angle (52 mm × 52 mm) displayed on a gray 60-cm background monitor (Samsung S19E200). The auditory stimulus (A) was a 1000 Hz, 60 dB sinusoidal tone with a 12 ms duration, presented via a loudspeaker positioned behind the monitor. According to previous studies, the training task was divided into 10 difficulty levels, with stimulus onset asynchronies (SOAs) between the auditory and visual stimuli ranging from 300 ms (level 1) to 30 ms (level 10) in 30 ms intervals. Each level presented 20 synchronous and 20 asynchronous audiovisual stimulus pairs in random order, lasting approximately 2.5 minutes per level. Participants pressed the left mouse button for synchronous stimuli and the right button for asynchronous stimuli. Training began at a medium difficulty level (150 ms SOA), with difficulty adjusted based on participant performance: advancing with three consecutive successful trials (accuracy > 80%), decreasing with a failed trial (accuracy < 60%), or repeating with moderate accuracy (60% ≤ accuracy < 80%). Training lasted one hour per day with self-paced breaks between levels.

The cognitive assessment included the SJ task and audiovisual detection task. For the SJ task, the same experimental materials and paradigm were used, except that audiovisual stimuli with stimulus onset asynchronies (SOAs) ranging from 0 to 300 ms (0, 30, 60, 90, 120, 150, 180, 210, 240, 270, 300 ms) were presented randomly. To balance responses, synchronous and asynchronous stimuli were presented in a 1:1 ratio. In total, there were 600 trials, consisting of 300 synchronous and 300 asynchronous stimuli (15 trials per SOA condition), divided into four blocks with appropriate rest periods between blocks.

For the audiovisual detection task (see Figure 3 ), the visual stimuli were 10% contrast, 1.5 spatial frequency Gabor patches, including both vertical and horizontal orientations. The auditory stimuli were sinusoidal tones at 1000 Hz and 500 Hz. Audiovisual stimuli were presented as combinations of vertical Gabor patches with 1000 Hz tones and horizontal Gabor patches with 500 Hz tones; no other combinations of auditory and visual stimuli were used in this study. All visual stimuli were presented at the center of a gray-background monitor, positioned 60 cm in front of the participant, while auditory stimuli were delivered through speakers located centrally behind the monitor at 60 dB. The experiment began with a fixation cross (“+”) at the center of the screen for 3000 ms as in previous studies. Following fixation, auditory, visual, and audiovisual stimuli were presented randomly for 100 ms with an interstimulus interval (ISI) ranging randomly from 1800 to 3000 ms. Participants were instructed to respond to each stimulus. In total, there were 300 trials, lasting 12 minutes, divided into two blocks with a self-paced break.

For the SJ task, the probability of simultaneity judgments at each SOA was calculated for each participant in both pre- and post-tests. A two-tailed t-test was then conducted for each SOA. Finally, the individual means at each SOA were averaged to generate the grand average plots (see Figure 4 ).

For the audiovisual detection task, hit rates and response times (RTs) were computed separately for each participant and analyzed using a 2 (Group: training, control) × 2 (Test: pre, post) × 3 (Stimulus: A, V, AV) repeated-measures ANOVA, followed by post hoc analysis with Greenhouse-Geisser corrections and adjusted degrees of freedom. Statistical analyses were conducted using IBM SPSS Statistics 22.0 (IBM Corp., Armonk, NY, USA), with statistical significance set at p ≤ 0.05. Additionally, audiovisual integration (AVI) was assessed using a race model based on cumulative distribution functions (CDFs) [14] [15] .

A two-tailed t-test revealed a significant improvement in perceptual ability in the post-test compared to the pre-test across all SOA conditions ( Figure 2 , all ps ≤ 0.014).

Hit rates and response times (RTs) are shown in Table 1 . A three-factor repeated-measures ANOVA for hit rate revealed a significant main effect of the test [F (1, 38) = 3.864, p = 0.047], indicating higher accuracy in the post-test compared to the pre-test, and a significant main effect of stimulus [F(2, 76) = 23.454, p < 0.001], with accuracy for AV stimuli higher than for A and V stimuli (all ps ≤ 0.011).

The three-factor repeated measures ANOVA for RTs revealed a significant main effect of stimulus [F(2, 76) = 85.702, p < 0.001], indicating faster responses to audiovisual (AV) stimuli compared to auditory (A) and visual (V) stimuli (AV > V > A, all ps < 0.001). The interaction between the group and test approached significance [F(1, 38) = 3.018, p = 0.059]. Post hoc analysis indicated that responses to A and V were faster in the pre-test compared to the post-test, whereas responses for the control group were either comparable or reversed. No other significant main effects or interactions were identified.

Additionally, based on responses to auditory and visual stimuli, the race model (RT_{race model}) was computed following previous studies [1] [3] and compared with responses to audiovisual stimuli (RT_AV, see Figure 5 ). In the training group, audiovisual integration (AVI) was significantly higher in the post-test (13.87%) compared to the pre-test (9.95%). However, in the control group, AVI remained relatively stable between the pre-test (10.77%) and post-test (9.61%). These findings suggest that perceptual training enhanced AVI.