A method for quantitatively evaluating auditory short-term memory based on snare drum rhythm imitation training

Abstract

The application discloses a kind of short-term memory of hearing quantification evaluation method based on snare drum rhythm imitation training, the test system comprising host computer, PPG sensor and electronic drum is first constructed, and the score of the short-term memory of hearing of the subject is obtained as supervision label by DMTS;Snare drum rhythm imitation training is then executed, and the average piece of time-consuming, training times and other behavior data are recorded synchronously, and resting pulse wave signal is collected before and after training;IBI sequence is constructed by band-pass filtering, valley detection and other pretreatments to pulse wave signal, and the difference features before and after training are extracted by extracting pulse rate variability features;Finally, the supervised learning regression model is constructed by fusing physiological features and training behavior features, the non-linear characteristics are fitted by smoothing function, the overfitting is inhibited by regularization constraint, and the objective quantification result of the short-term memory of hearing of new subject is output.The application realizes the integration of measurement and training evaluation, improves the objectivity, stability and repeatability of evaluation, and provides a systematic and deployable technical solution.

Description

Technical Field

[0001] This invention relates to the field of cognitive ability assessment and information technology, specifically to a quantitative assessment method for auditory short-term memory based on snare drum rhythm imitation training. Background Technology

[0002] Auditory short-term memory (ASTM) is an individual's ability to retain, update, and retrieve auditory information within a short period of time. It is a crucial foundation for language comprehension, music processing, attention maintenance, and complex learning. In research and applications, ASTM levels are closely related to classroom learning efficiency, foreign language speech processing, musical instrument learning, and rhythm control. A decline in ASTM may manifest as difficulty retaining auditory information, limited sequence processing, and reduced efficiency in attention allocation. Therefore, establishing quantifiable, repeatable, and easily deployable methods for assessing and training auditory short-term memory has significant theoretical and practical value.

[0003] Currently, the assessment methods for auditory short-term memory mainly fall into two categories: subjective assessment and objective assessment. Subjective assessment often relies on self-report scales or interviews, which has shortcomings such as strong subjectivity, susceptibility to recall bias and social expectations, and difficulty in providing precise quantitative indicators. Objective assessment is usually based on the performance of behavioral tasks, such as digit span, sequence repetition, and delayed sample matching, and uses indicators such as accuracy or reaction time to characterize memory level. However, it also has certain limitations: first, single behavioral indicators are easily affected by differences in subjects' strategies, motivation, and fatigue; second, multiple measurements are prone to the practice effect, leading to decreased comparability across stages and populations; and third, traditional assessment is often separated from the training process, making it difficult to achieve synchronous, continuous, and low-burden dynamic assessment in natural training situations.

[0004] In terms of physiological indicators-assisted assessment, existing research has shown a correlation between autonomic nervous system activity and cognitive load, attentional control, and memory processing states. While traditional techniques such as electroencephalography (EEG) and functional magnetic resonance imaging (fMRI) can provide rich neural information, their expensive equipment, complex procedures, and strong dependence on specific environments make them difficult to widely apply in everyday training settings. In contrast, indicators such as heart rate variability extracted from inter-cardiac interval sequences can reflect sympathetic-parasympathetic regulatory states; photoplethysmography (PPG), as a portable, non-invasive, and low-cost acquisition method, can obtain information related to inter-cardiac intervals in relatively natural training environments, providing a feasible path for objectively characterizing changes in physiological states before and after training. However, current work often only uses physiological indicators for coarse-grained state differentiation, lacking a systematic approach for joint modeling with behavioral training performance.

[0005] In terms of training methods, rhythm processing and auditory-motor coupling training are considered highly correlated with processing processes such as sequence preservation, temporal prediction, and working memory updating. Rhythm imitation training with percussion instruments like the snare drum offers advantages such as controllable task burden, clear feedback, standardized segmentation, and repetitive practice, making it suitable as a training vehicle for auditory short-term memory-related abilities. However, existing training research still has two shortcomings: first, it lacks a unified evaluation framework that integrates training behavioral data with physiological signals, making it difficult to distinguish between the contributions of "mastering the task" and "ability improvement"; second, it lacks a quantitative mapping model using objective memory task scores as supervised labels, resulting in training effect evaluations often remaining at the level of experience-based judgment or single-indicator comparison, making it difficult to provide generalizable individualized assessment results for new subjects. Based on these problems, it is necessary to propose an evaluation method that can simultaneously collect behavioral and pulse wave signals in rhythm imitation training scenarios and achieve quantitative output of auditory short-term memory through supervised learning. Summary of the Invention

[0006] The main objective of this invention is to overcome the shortcomings and deficiencies of existing technologies and provide a quantitative assessment method for auditory short-term memory based on snare drum rhythm imitation training. This invention uses the memory score obtained from an auditory delay sample matching task as a supervisory label. During rhythm imitation training, it synchronously records behavioral performance such as average musical segment duration and training frequency, and collects resting pulse wave signals before and after training. By preprocessing the pulse wave signals and constructing a heart rate interval sequence, it extracts pulse rate variability features and pre- and post-training difference features, further integrating training behavioral features to construct a supervised learning regression model. This achieves an objective quantitative output of the auditory short-term memory ability of new subjects, thus providing a systematic and deployable technical solution for the integrated assessment of auditory short-term memory testing and training in rhythm imitation training scenarios.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] This invention provides a method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training, comprising the following steps:

[0009] S1. Construct an experimental scenario and testing system. The testing system includes a host computer for presenting stimulus materials, a PPG sensor for acquiring photoplethysmography (PPG) signals, and an electronic drum for acquiring performance signals.

[0010] S2. Perform the Auditory Delayed Sample Matching Task (DMTS). The DMTS includes a preset number of trials, a sound stimulus library consisting of a preset number of samples, and new and old stimuli configured in a preset ratio. In each trial, a sequence of several samples is presented first, and after a preset delay, a probe sample is presented. The subject judges whether the probe sample belongs to the aforementioned sequence. The DMTS score is obtained based on the judgment accuracy of all trials and serves as a supervisory label characterizing the subject's auditory short-term memory ability.

[0011] S3. Perform rhythm imitation training and collect physiological signals. Divide the digital score into multiple musical segments that meet homogeneity constraints. Guide the subject to imitate the main melody by hitting the small drum. Use event-level matching and alignment strategies to calculate the correspondence between the target and the actual hitting events. When the subject correctly hits the same musical segment a preset number of times, the musical segment is judged to be passed. Record the training behavior data of the average musical segment time and the number of training sessions. Before and after the rhythm imitation training, collect the subject's resting pulse wave signal for a preset duration as the pre-training baseline and post-training PPG data, respectively.

[0012] S4. Construct physiological signal preprocessing and interval sequences, perform bandpass filtering on the baseline before training and the post-test PPG data after training to eliminate noise, perform valley search on the filtered signal, and suppress false detection based on the minimum interval constraint between adjacent valleys and the adaptive significance threshold to construct the pulse interval IBI sequence.

[0013] S5. Construct physiological feature extraction and difference features, apply physiological range constraints and relative median mutation constraints to the pulse interval sequence, remove abnormal cycles and artifacts, extract pulse rate variability features on the effective cycle set, and form physiological feature vector before training and physiological feature difference vector before and after training respectively.

[0014] S6. Construct an evaluation model and train and infer it. Use the pre-training PPG feature vector, the difference feature vector before and after training, the average number of musical phrase attempts and the number of training sessions as input features, and the DMTS score as the label to construct a supervised learning regression model. Based on the supervised learning regression model, output the quantitative evaluation results of the new subject's auditory short-term memory.

[0015] As a preferred technical solution, step S1 specifically includes:

[0016] S101. Construct experimental scenarios and equipment deployment, and build a test system in the preset experimental environment;

[0017] S102. Establish a multimodal data acquisition link and synchronization marking mechanism, configure the communication and event input link between the host computer and the PPG sensor and electronic drum, set up a unified experimental process control module on the host computer side, perform state management of the experimental stage, and generate corresponding stage numbers and timestamps for each stage, so that PPG continuous sampling data, electronic drum hitting event sequence and task log can be aligned and fused based on timestamps or stage numbers;

[0018] S103. Configure the stimulus presentation and human-computer interaction module, complete the configuration of auditory stimulus playback parameters on the host computer, including stimulus library loading and volume calibration, and configure the subject response input method and recording mechanism.

[0019] S104. Conduct trial runs and quality checks. Before formal data collection, guide the subjects to complete a short trial run to verify whether the auditory stimulus is presented normally, whether the subject's response is fully recorded, whether the electronic drum hitting event is triggered stably, and whether the PPG signal has disconnection, saturation, or obvious motion artifacts. When the signal quality or link status is found to be unsatisfactory, readjust the sensor wearing status or device connection parameters, and repeat the trial run until the quality check is passed.

[0020] As a preferred technical solution, step S2 specifically includes:

[0021] S201. Establish an auditory stimulus library and configure trial parameters. Establish an auditory stimulus library on the host computer side. The auditory stimulus library includes a preset number of distinguishable sound samples. Configure the trial scale of the DMTS task to a preset number of trials and set the trial ratio of "old stimulus / new stimulus" to a preset ratio. Wherein, "old stimulus" refers to the detection sample belonging to the memory sequence of the current trial, and "new stimulus" refers to the detection sample not belonging to the memory sequence of the current trial.

[0022] S202. Generate a trial sequence and execute the stimulation program. For each trial, generate a memory sequence consisting of a preset number of sound samples based on the auditory stimulus library and present them sequentially. After the memory sequence is presented, set a preset delay interval. After the delay interval, present the probe samples. The probe samples are generated according to the ratio of "old stimulus / new stimulus" and sampling rules.

[0023] S203. Collect the subject's judgment response and record the trial results. After the probe sample is presented, guide the subject to complete the two-choice judgment operation within the preset response window to indicate whether the probe sample belongs to the aforementioned memory sequence. The host computer records the subject's response results.

[0024] S204. Calculate the DMTS score and generate a supervision label. Based on the overall accuracy of the judgments in all trials, the DMTS score is obtained as a quantitative indicator of auditory short-term memory ability and a supervision learning label.

[0025] As a preferred technical solution, step S3 specifically includes:

[0026] S301. Construct a candidate rhythm material library. Select several digital scores that can be used for rhythm imitation training on the host computer side, extract their rhythm and note event sequence information, and form a candidate training material set. The candidate materials are represented by "event time point sequence", with each note / hit corresponding to a target occurrence time point.

[0027] S302. The score is segmented and standardized according to homogenization constraints to generate a set of musical segments; the candidate score is segmented into multiple training segments, and homogenization constraints are applied to each segment to control the consistency of difficulty. The homogenization constraints include at least the beat structure constraint, tempo constraint, segment length constraint, minimum rhythm unit constraint, event quantity constraint, and starting accent position constraint; based on satisfying the above constraints, each segment is converted into a standardized target rhythm template, and a corresponding target hit event timestamp sequence is generated.

[0028] S303. Configure the training process and music segment presentation strategy. In one training session, select a preset number of music segments from the standardized music segment set as training units. Set a closed-loop process for each music segment: "loop playback while imitating the answer - real-time evaluation - pass or fail". The host computer repeatedly plays the current music segment and opens the acquisition window for the subject's imitation input at the beginning of each playback to obtain the actual hitting event sequence of the attempt.

[0029] S304. Collect electronic drum hitting events and construct an actual event sequence. The subject hits the small drum according to the main melody rhythm of the music segment. The electronic drum converts the hitting trigger signal into an event timestamp and transmits it to the host computer. The host computer sorts the hitting events of each attempt by time to form an actual event sequence, and records it in association with the current music segment number, the number of attempts, and the presentation start time to support subsequent matching and scoring.

[0030] S305. Based on the matching strategy, establish the correspondence between the target and the actual events and calculate the passing criteria. For a single attempt, input the target event sequence and the actual event sequence into the matching module. Use the event matching strategy on the time axis to select the actual event with the smallest time difference and satisfying the monotonic order constraint for each target event on the time axis, thereby establishing a one-to-one correspondence. And calculate the correctness index of the attempt based on the correspondence.

[0031] S306. According to the "passing a preset number of times consecutively" rule, determine the mastery of the musical segment and record the training behavior data. For the same musical segment, if the subject continuously reaches the preset number of times and meets the passing criterion, the musical segment is determined to be passed and the subject switches to the next musical segment. If the continuous passing condition is not met, the subject continues to repeat the playback and imitation of the musical segment until the subject passes. Record the number of times the musical segment is repeated during the training process and write the number of repetitions into the training summary record for subsequent modeling.

[0032] S307. Pre- and post-training resting PPG data acquisition and archiving: Before and after the rhythm imitation training, the subjects are guided to perform resting measurements and PPG signals of a preset duration are collected. The PPG signals are continuously sampled and recorded through a preset sensor channel at a preset sampling frequency, and the original time series data of the pre-training baseline and the post-training test are written into the storage medium.

[0033] As a preferred technical solution, in step S305, the criterion includes at least:

[0034] (1) Event consistency constraints are used to suppress excessive knocking, insufficient knocking, or missed matching;

[0035] (2) Time error constraint, used to limit the time deviation between the target event and the corresponding actual event to not exceed a preset threshold;

[0036] If the above criteria are met, the attempt is deemed successful; otherwise, it is deemed unsuccessful and proceeds to the next round of repeated playback and imitation.

[0037] As a preferred technical solution, step S4 specifically includes:

[0038] S401. Obtain PPG data and confirm parameter consistency: Obtain the original PPG time series data archived before and after training as preprocessing input, and confirm that the collection parameters are consistent;

[0039] S402. Bandpass filtering preprocessing to suppress noise and baseline drift: Preprocessing is performed on the acquired raw PPG signal by using a bandpass filter to filter the signal. The bandpass frequency band is set to a preset frequency band range to suppress low-frequency drift caused by body movement and respiration, as well as high-frequency noise caused by environmental electromagnetic and electromyographic factors, thereby obtaining a smooth target PPG signal that can be used for feature point detection.

[0040] S403. Preparation for candidate valley detection: Perform candidate extreme value scanning on the filtered signal to obtain the set of candidate valley locations.

[0041] S404. Based on the set constraints, the candidate valley location set is filtered for validity to obtain the final valley sequence, which is used as the boundary point set for pulse cycle segmentation. The preset constraints include at least: (1) using the minimum interval constraint between adjacent valleys to suppress dense false detections caused by noise or local oscillations; (2) constructing an adaptive significance threshold based on the amplitude distribution of the target PPG signal, and retaining only the candidate valleys that meet the significance requirements to reduce the detection of false valleys caused by small amplitude fluctuations.

[0042] S405. Based on the final trough sequence, calculate the time difference between adjacent troughs to obtain the pulse interval sequence; at the same time, output intermediate result information that matches the IBI sequence, and use the IBI sequence as input for physiological feature extraction; the intermediate structure information includes trough timestamp, effective cycle number and basic quality marker.

[0043] As a preferred technical solution, step S5 specifically includes:

[0044] S501. Clean the IBI sequences and remove abnormal periods and artifacts in the IBI sequences by means of physiological range constraints and relative median mutation constraints to obtain the effective IBI set and the corresponding effective period set;

[0045] S502. Construct a time-domain analysis effective sequence, form a continuous effective IBI sequence on the effective period set, calculate the difference between adjacent effective IBIs to form a difference sequence, and mark the quality of records with insufficient effective sample quantity or too many discontinuities.

[0046] S503. Calculate the time-domain PRV features on the effective IBI sequences, wherein the time-domain PRV features include at least the average heart rate, standard deviation, root mean square difference, and the proportion of adjacent effective IBI sequences whose absolute values ​​exceed 50 ms.

[0047] S504. Extract frequency domain pulse rate variability features and calculate the frequency band ratio. Based on the effective IBI sequence, construct an equal time interval sequence for frequency domain analysis and perform power spectrum estimation to obtain low-frequency power, high-frequency power and low-high frequency power ratio.

[0048] S505. Construct the pre-training feature vector and the difference feature vector before and after training. Extract physiological feature vectors from the pre-training resting segment and the post-training resting segment respectively to obtain the pre-training physiological feature vector and the post-training physiological feature vector.

[0049] As a preferred technical solution, in step S505, the difference between the feature vector after training and the feature vector before training is calculated for the same subject to form a physiological feature difference vector before and after training; finally, the feature vector before training and the difference vector are organized and stored in a unified feature order.

[0050] As a preferred technical solution, step S6 specifically includes:

[0051] S601. Construct a training sample and feature input set. For each subject, summarize and form a modeling sample that corresponds one-to-one with their DMTS score. The input features include at least the pre-training PPG physiological feature vector, the pre- and post-training difference feature vector, the average number of musical phrase attempts, and the number of training sessions.

[0052] S602. Determine the modeling objectives and labels, using DMTS scores as supervised learning labels and target variables, and establish a mapping relationship between input features and DMTS scores;

[0053] S603. Constructing a regression model and smoothing fitting: Construct a supervised learning regression model, and use spline functions, kernel smoothing or other differentiable smoothing basis functions to fit nonlinear features, and use linear terms to model linear features.

[0054] S604, Regularization Constraints and Model Complexity Control: Introducing regularization constraints to suppress overfitting, adjusting the regularization strength or smoothing the degrees of freedom to balance the model's fitting ability and generalization performance.

[0055] S605. New Subject Inference Output: Input relevant features of the new subject, output the predicted DMTS score as a quantitative evaluation result through the model, and store the relevant data.

[0056] As a preferred technical solution, in step S604, the regularization includes at least a penalty term for the curvature of the smooth function or a penalty term for the magnitude of the parameter, so as to limit function fluctuations and improve generalization ability.

[0057] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0058] (1) This invention proposes a quantitative assessment method for auditory short-term memory that integrates training behavioral data and pulse wave physiological data. This method overcomes the problems of existing technologies that rely solely on subjective scales or single behavioral accuracy rates, resulting in limited dimensions, high susceptibility to strategy and motivation, and difficulty in achieving objective and continuous assessment in training scenarios. By introducing the characteristics of resting pulse waves before and after training and their difference characteristics, and combining them with rhythm imitation training performance indicators, the assessment results have interpretable physiological basis and stronger stability, thereby improving the objectivity and comparability of auditory short-term memory assessment.

[0059] (2) This invention employs a heartbeat interval sequence construction strategy that combines bandpass filtering and valley detection in the pulse wave processing link. It introduces a minimum interval constraint between adjacent valleys and an adaptive significance threshold to suppress false detections. Simultaneously, it combines physiological range constraints and relative median mutation constraints to eliminate abnormal cycles and artifacts, thereby reducing the interference of motion artifacts, transient noise, and false detections on the interval sequence. This method improves the effectiveness and robustness of the heartbeat interval sequence and enhances the accuracy and repeatability of extracting pulse rate variability features such as mean heart rate, SDNN, RMSSD, pNN50, LF, HF, and LF / HF.

[0060] (3) This invention uses the snare drum rhythm imitation training as both a training vehicle and an evaluation scenario. It employs a regression modeling framework with auditory delay sample matching task scores as supervisory labels, integrating features before the first training, differences before and after training, and training behavior features to achieve direct quantitative output of auditory short-term memory ability for new subjects. Compared to the traditional fragmented "training-evaluation" process, this invention achieves integrated testing and training with a generalizable model-based evaluation. It can also be used for quantitative comparison and individualized feedback of training effects, providing a complete and easily deployable technical solution for objective evaluation of auditory short-term memory and assessment of training effects. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 The flowchart is a method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training disclosed in Embodiment 1 of the present invention.

[0063] Figure 2 This is a flowchart of pulse wave signal processing in Embodiment 1 of the present invention;

[0064] Figure 3 This is a flowchart of the pulse interval sequence processing in Embodiment 1 of the present invention;

[0065] Figure 4 This is a pulse wave signal diagram after bandpass filtering from 0.5Hz to 8Hz in Embodiment 2 of the present invention;

[0066] Figure 5 This is a diagram showing the effect of valley bottom detection in Embodiment 2 of the present invention. Detailed Implementation

[0067] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative effort are within the scope of protection of the present application.

[0068] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0069] Example 1

[0070] like Figure 1 As shown in this embodiment, a quantitative assessment method for auditory short-term memory based on snare drum rhythm imitation training includes the following steps:

[0071] S1. Constructing the experimental scenario and hardware / software system. The testing system includes a host computer for presenting stimulus materials, sensors for acquiring photoplethysmography (PPG) signals, and an electronic drum for acquiring performance signals. Subjects are guided through trial runs and equipment quality checks within the experimental scenario.

[0072] Furthermore, step S1 is performed as follows:

[0073] S101. Constructing the experimental scenario and deploying equipment. A testing system is built in a pre-set experimental environment. The testing system includes at least: a host computer for presenting auditory stimuli and performing task control; a PPG sensor for acquiring photoplethysmography (PPG) signals; and an electronic drum input device for acquiring subject-specific striking events.

[0074] S102. Establish a multimodal data acquisition link and synchronization marking mechanism. Configure the data communication link between the host computer and the PPG sensor, and the event input link between the host computer and the electronic drum. Set up a unified experimental process control module on the host computer side to manage the status of the experimental stages and generate corresponding stage numbers and timestamps for each stage, so that PPG continuous sampling data, electronic drum impact event sequences and task logs can be aligned and fused based on timestamps or stage numbers.

[0075] S103. Configure the stimulus presentation and human-computer interaction module. On the host computer side, configure the parameters of the auditory stimulus playback module, including loading the stimulus library and calibrating the playback volume. Configure the subject response input method and recording mechanism to enable subjects to complete judgment responses when performing the DMTS task, and ensure that the response results can be stored in association with the trial number and presentation time.

[0076] S104. Conduct trial runs and quality checks. Before formal data collection, guide subjects to complete a short trial run to verify whether auditory stimuli are presented normally, whether subject responses are fully recorded, whether electronic drum impact events are triggered stably, and whether there are any disconnections, saturation, or obvious motion artifacts in the PPG signal. If the signal quality or link status is found to be unsatisfactory, readjust the sensor wearing status or device connection parameters, and repeat the trial run until the quality check is passed.

[0077] S2. Perform the Delayed Match-to-Sample (DMTS) auditory task. The DMTS task includes a preset number of trials. The sound stimulus library consists of a preset number of samples, with new and old stimuli configured in a preset ratio. In each trial, a sequence consisting of several samples is presented first, followed by a probe sample after a preset delay. The subject judges whether the probe sample belongs to the aforementioned sequence. The DMTS score is obtained based on the accuracy rate of judgments across all trials, serving as a supervisory label characterizing the subject's auditory short-term memory ability.

[0078] Furthermore, step S2 is as follows:

[0079] S201. Establish an auditory stimulus library and complete the trial parameter configuration. An auditory stimulus library is established on the host computer side, containing a preset number of distinguishable sound samples. The trial size of the DMTS task is configured to a preset number of trials, and the trial ratio of "old stimulus / new stimulus" is set to a preset ratio. Here, "old stimulus" refers to the probe sample belonging to the memory sequence of the current trial, and "new stimulus" refers to the probe sample not belonging to the memory sequence of the current trial.

[0080] S202. Generate a trial sequence and perform stimulus presentation. For each trial, a memory sequence consisting of a preset number of sound samples is generated based on the stimulus library and presented sequentially. After the memory sequence is presented, a preset delay interval is set. After the delay, probe samples are presented. The probe samples are generated according to the "old stimulus / new stimulus" ratio and sampling rules to ensure that the conditions of each trial meet the preset distribution constraints.

[0081] S203. Collect the subject's judgment response and record the trial results. After the probe sample is presented, guide the subject to complete a two-choice judgment operation within a preset response window to indicate whether the probe sample belongs to the aforementioned memory sequence. The host computer records the subject's response results.

[0082] S204. Calculate the DMTS score and generate supervision labels. Based on the correctness of judgments across all trials, calculate the overall judgment accuracy to obtain the DMTS score.

[0083] Formula (1)

[0084] in, Let N be the indicator function, and N be the number of trials. For the subjects to judge, The result is a true value (old / new). The DMTS score is used as a quantitative indicator of the subject's auditory short-term memory ability and is written into the task summary record as a supervised learning label for subsequent evaluation model construction.

[0085] S3. Execute rhythm imitation training and collect physiological signals. Divide the digital score into multiple musical segments that conform to homogeneity constraints. Guide the subject to imitate the main melody by striking a small drum, and use event-level matching and alignment strategies to calculate the correspondence between the target and the actual striking events. The subject is considered to have passed when they continuously strike the same musical segment a preset number of times, meeting the passing condition. Record training behavior data such as average musical segment duration and number of training sessions. Before and after the rhythm imitation training begins, collect the subject's resting pulse wave signals for a preset duration as baseline data before training and post-test data after training.

[0086] Furthermore, step S3 is as follows:

[0087] S301. Select the original score and construct a candidate rhythm material library. On the host computer side, select several digital scores or main melody fragments that can be used for rhythm imitation training, extract their rhythm and note event sequence information, and form a candidate training material set; wherein, the candidate material is represented by an "event time point sequence", that is, each note / hit corresponds to a target occurrence time point.

[0088] S302. Segment and standardize the musical score according to homogenization constraints to generate a set of musical segments. The candidate musical score is segmented into multiple training segments, and homogenization constraints are applied to each segment to control the consistency of difficulty. The homogenization constraints include at least rhythmic structure constraints, tempo constraints, segment length constraints, minimum rhythmic unit constraints, event quantity constraints, and starting accent position constraints. Based on satisfying the above constraints, each segment is converted into a standardized target rhythm template, and a corresponding target hit event timestamp sequence is generated.

[0089] S303. Configure the training process and music segment presentation strategy. In one training session, a preset number of music segments are selected from the standardized music segment set as training units; a closed-loop process of "loop playback - imitation response - real-time evaluation - pass or fail" is set for each music segment, wherein the host computer repeatedly plays the current music segment and opens the acquisition window for the subject's imitation input after each playback to obtain the actual hitting event sequence of that attempt.

[0090] S304. Collect electronic drum hitting events and construct an actual event sequence. The subject hits the snare drum according to the rhythm of the main melody played in the musical segment; the electronic drum converts the hitting trigger signal into an event timestamp and transmits it to the host computer; the host computer sorts the hitting events of each attempt by time to form an actual event sequence, and records it in association with the current musical segment number, the number of attempts, and the presentation start time to support subsequent matching and scoring.

[0091] S305. Establish a correspondence between the target and actual events based on a matching strategy and calculate the pass criterion. For a single attempt, the target event sequence and the actual event sequence are input into the matching module. An event matching strategy on the time axis is used to select the actual event with the smallest time difference and satisfying the monotonic order constraint for each target event on the time axis, thereby establishing a one-to-one correspondence; wherein, the matching mapping... Defined as satisfying Under the constraints, Minimal actual event index:

[0092]

[0093] Formula (2)

[0094] Where M is the number of actual event time points. For the j-th actual event time point, For the first A target event time point. The correctness index of this attempt is calculated based on the correspondence, and the criteria include at least: (1) event consistency constraints, used to suppress multiple knocks, fewer knocks, or missed matches; (2) time error constraints, calculating the time deviation between the target event and the corresponding actual event.

[0095] Formula (3)

[0096] Calculate the standard deviation of this time deviation.

[0097] Formula (4)

[0098] The standard deviation must be less than a preset threshold. If the event consistency is met and the standard deviation of the time deviation is less than the preset threshold, the attempt is considered successful; otherwise, it is considered unsuccessful and proceeds to the next round of repeated playback and imitation.

[0099] S306. Determine the mastery of a musical segment and record the training behavior data according to the "two consecutive passes" rule. For the same musical segment, if the subject meets the passing criteria for a preset number of consecutive times, the segment is considered passed and the subject moves to the next segment. If the consecutive passing condition is not met, the subject continues to repeat the playback and imitation of the segment until the subject passes. The number of times the musical segment is repeated during the training process is recorded and written into the training summary record for subsequent modeling.

[0100] S307. Pre- and Post-Training Resting PPG Acquisition and Data Archiving. Before and after the rhythm imitation training, subjects are guided to perform resting measurements and PPG signals of a preset duration are collected. The PPG signals are continuously sampled and recorded through a preset sensor channel at a preset sampling frequency, and the original time-series data of the pre-training baseline and the post-training test are written into a storage medium.

[0101] S4. Constructing Physiological Signal Preprocessing and Interval Sequences. Bandpass filtering is applied to the baseline before training and the post-training PPG data to eliminate noise. Valley search is performed on the filtered signal, and false detections are suppressed based on the minimum interval constraint between adjacent valleys and the adaptive significance threshold to construct the Inter-Beat Interval (IBI) sequence.

[0102] Furthermore, step S4 is as follows:

[0103] S401. Acquisition of Resting PPG Data Before and After Training and Confirmation of Parameter Consistency. Acquire the raw PPG time series data archived in step S3, including the baseline before training and the post-test data after training, and use the raw data as input for subsequent preprocessing.

[0104] S402. Bandpass filtering preprocessing to suppress noise and baseline drift. The raw PPG signal obtained in step S401 is preprocessed by using a bandpass filter to filter the signal. The bandpass frequency is set to a preset frequency range to suppress low-frequency drift caused by body movement and respiration, as well as high-frequency noise caused by environmental electromagnetic and electromyographic factors, thereby obtaining a smooth target PPG signal that can be used for feature point detection.

[0105] S403. Preparation for candidate valley detection. Perform a candidate extreme value scan on the signal to obtain a set of candidate valley locations, which provides input for subsequent false detection suppression and effective valley selection.

[0106] S404. Valley location and false detection suppression based on interval constraints and adaptive saliency threshold. The candidate valley location set is screened for effectiveness, including at least the following constraints: (1) Employing a minimum interval constraint between adjacent valleys to suppress dense false detections caused by noise or local oscillations. (2) Constructing an adaptive saliency threshold based on the amplitude distribution of the target PPG signal.

[0107] Formula (5)

[0108] in , The 95th and 5th percentiles of the target PPG amplitude are used, respectively, and only candidate valleys that meet the significance requirement are retained to reduce the detection of false valleys caused by small fluctuations. The final valley sequence is obtained through the above constraints and is used as the set of boundary points for pulse cycle segmentation.

[0109] S405. Construct the pulse interval sequence and output intermediate results for subsequent feature extraction. Based on the final valley sequence, calculate the time difference between adjacent valleys to obtain the pulse interval sequence.

[0110] Formula (6)

[0111] in For the first The system outputs the final trough timestamp. Simultaneously, it outputs intermediate result information associated with the IBI sequence, including the trough timestamp, the number of effective cycles, and baseline quality markers, and uses the IBI sequence as input for physiological feature extraction in step S5.

[0112] S5. Constructing Physiological Feature Extraction and Difference Features. Based on the IBI sequence, physiological range constraints and relative median mutation constraints are applied to eliminate abnormal cycles and artifacts. Pulse rate variability features are extracted from the effective cycle set, including mean heart rate, standard deviation of NN intervals (SDNN), root mean square difference (RMSSD), percentage of NN intervals differing by more than 50 ms (pNN50), low-frequency power (LF), high-frequency power (HF), and the low-to-high-frequency power ratio (LF / HF). These are used to form a pre-training physiological feature vector and a pre-training physiological feature difference vector, respectively.

[0113] Furthermore, step S5 is as follows:

[0114] S501. Cleaning the IBI sequences based on physiological range constraints and relative median mutation constraints. The IBI sequences obtained in step S4 are subjected to anomaly removal to reduce artifact effects. The cleaning includes at least: (1) Physiological range constraints: limiting the IBI sequences to a preset physiologically reasonable range, and removing intervals outside this range as abnormal; (2) Relative median mutation constraints: using the median of the IBI sequences as a robust reference, marking and removing samples where the relative median of adjacent intervals or local intervals shows mutations, to suppress octet or half-octet errors caused by false negative / false positive troughs. After the above constraint processing, an effective IBI set and its corresponding effective period set are obtained.

[0115] S502. Construct an effective interval sequence for calculating temporal pulse rate variability. Form a continuous effective IBI sequence on the effective period set and calculate the difference between adjacent effective IBIs to form a difference sequence. Quality-label records with insufficient effective sample size or excessive discontinuities to avoid statistical instability caused by scarce effective samples.

[0116] S503. Extract time-domain pulse rate variability (PRV) features. Calculate time-domain PRV features on the effective IBI sequence, including at least: (1) Mean heart rate: Obtain the mean pulse rate / heart rate level by converting the inverse of the effective IBI; (2) Standard deviation: Calculate the standard deviation of the effective IBI sequence to characterize the overall variability; (3) Root mean square difference: Calculate the root mean square of adjacent effective IBI difference sequences to characterize short-term variability components; (4) Proportion of absolute values ​​of adjacent effective IBI differences exceeding 50 ms: Statistically count the proportion of absolute values ​​of adjacent effective IBI differences exceeding 50 ms to characterize the significant fluctuation proportion of adjacent intervals.

[0117] S504. Extract frequency domain pulse rate variability features and calculate the frequency band ratio. Based on the effective IBI sequence, construct an equal time interval sequence for frequency domain analysis and perform power spectrum estimation to obtain: (1) Low frequency power (LF): obtained by integrating the power spectrum within the preset low frequency band; (2) High frequency power (HF): obtained by integrating the power spectrum within the preset high frequency band; (3) Low-high frequency power ratio (LF / HF): calculate the ratio of LF to HF to characterize the relative proportion of power components in different frequency bands.

[0118] S505. Construct the pre-training feature vector and the pre- and post-training difference feature vector. Perform steps S501–S504 for the pre-training resting segment and the post-training resting segment respectively to obtain the pre-training physiological feature vector and the post-training physiological feature vector. Further, calculate the difference between the post-training feature vector and the pre-training feature vector for the same subject, forming the pre- and post-training physiological feature difference vector. Finally, organize and store the pre-training feature vector and the difference vector in a unified feature order for use in the evaluation model training and inference in step S6.

[0119] S6. Evaluation Model Construction, Training, and Inference. Using the pre-training PPG feature vector, the difference feature vector before and after training, the average number of musical phrase attempts, and the number of training sessions as input features, and the DMTS score as the label, a supervised learning regression model is constructed for regression modeling. A smoothing function is used to fit some non-linear features, and overfitting is suppressed through regularization constraints. Finally, the model outputs a quantitative evaluation result of the new subject's auditory short-term memory.

[0120] Furthermore, step S6 is as follows:

[0121] S601. Construct training samples and feature input sets. For each subject, compile a modeling sample that corresponds one-to-one with their DMTS score. The input features include at least: pre-training PPG physiological feature vector, pre- and post-training physiological difference feature vector, average number of musical phrase attempts, and number of training sessions.

[0122] S602. Determine the supervised learning regression modeling objective and label definition. Use the DMTS score calculated in step S2 as the supervised learning label, and use it as the target variable characterizing the subject's auditory short-term memory ability. Establish a mapping relationship from input features to DMTS scores for regression prediction and individualized assessment.

[0123] S603. Construct a regression model structure and introduce a smoothing fitting mechanism. Construct a supervised learning regression model to fit the mapping relationship. For features that may exhibit nonlinear effects, a smoothing function is introduced for nonlinear fitting. This smoothing function can be a spline function, kernel smoothing, or other differentiable smoothing basis functions to characterize nonlinear trends while maintaining interpretability. For features with approximately linear effects, linear terms are used for direct modeling.

[0124] S604. Suppress overfitting and determine model complexity through regularization constraints. To avoid overfitting caused by smoothing terms or high-dimensional features, regularization constraints are introduced on the model parameters. The regularization includes at least a penalty term for the curvature of the smoothing function or a penalty term for the magnitude of the parameters, to limit function fluctuations and improve generalization ability. Simultaneously, by adjusting the regularization strength or smoothing degrees of freedom, model complexity is controlled, achieving a balance between the model's fitting ability on the training set and its generalization performance on the validation set.

[0125] S605. New Subject Inference Output. For a new subject, input their pre-training PPG feature vector, pre- and post-training difference vector, and training behavior features. Use the final evaluation model to output the predicted value of the DMTS score as a quantitative evaluation result of the new subject's auditory short-term memory ability. Record and store the prediction result and corresponding features to support subsequent follow-up analysis.

[0126] Example 2

[0127] In a more specific embodiment, the present invention provides a method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training, specifically including the following steps:

[0128] S1. Constructing the experimental scenario and hardware / software system. The testing system includes a host computer for presenting stimulus materials, sensors for acquiring photoplethysmography (PPG) signals, and an electronic drum for acquiring performance signals. Guide the subjects through trial runs and equipment quality checks within the experimental scenario. Specific steps are as follows:

[0129] S101. Constructing the experimental scenario and deploying equipment. A testing system is set up in a quiet laboratory. The testing system includes at least: a Windows host computer for presenting auditory stimulus materials and performing task control; a PPG sensor based on an ESP32 controller and a MAX30102; and an electronic drum input device based on an N32G452 and a piezoelectric sensor.

[0130] S102. Establish a multimodal data acquisition link and synchronization marking mechanism. Configure a USB serial port data communication link between the host computer and the PPG sensor, and a USB serial port event input link between the host computer and the electronic drum. On the host computer side, use Python to set up a unified experimental process control module to manage the experimental stages and generate corresponding stage numbers and timestamps for each stage, so that PPG continuous sampling data, electronic drum impact event sequences, and task logs can be aligned and fused based on timestamps or stage numbers.

[0131] S103. Configure the stimulus presentation and human-computer interaction module. On the host computer side, configure the parameters of the auditory stimulus playback module, including loading the stimulus library and calibrating the playback volume. Configure the subject response input method and recording mechanism to enable subjects to complete judgment responses when performing the DMTS task, and ensure that the response results can be stored in association with the trial number and presentation time.

[0132] S104. Conduct trial runs and quality checks. Before formal data collection, guide subjects to complete a short trial run to verify whether auditory stimuli are presented normally, whether subject responses are fully recorded, whether electronic drum impact events are triggered stably, and whether there are any disconnections, saturation, or obvious motion artifacts in the PPG signal. If the signal quality or link status is found to be unsatisfactory, readjust the sensor wearing status or device connection parameters, and repeat the trial run until the quality check is passed.

[0133] S2. Perform the auditory DMTS task. The DMTS task consists of 100 trials. The sound stimulus library consists of 25 samples, with each sound sample appearing in 4 trials. New and old stimuli are configured in a 1:1 ratio. In each trial, a sequence of 6 samples is presented first, followed by a probe sample after a 1.2-second delay. The subject judges whether the probe sample belongs to the aforementioned sequence. The DMTS score is calculated based on the accuracy of judgments across all trials and serves as a supervisory label characterizing the subject's auditory short-term memory ability. The specific steps are as follows:

[0134] S201. Establish an auditory stimulus library and configure trial parameters. An auditory stimulus library is established on the host computer side, containing 25 distinguishable sound samples. The trial size for the DMTS task is configured to 100 trials per session, ensuring each sound sample appears in 4 trials, and the trial ratio of "old stimulus / new stimulus" is set to 1:1. Here, "old stimulus" refers to the probe sample belonging to the memory sequence of the current trial, and "new stimulus" refers to the probe sample not belonging to the memory sequence of the current trial.

[0135] S202. Generate a trial sequence and perform stimulus presentation. For each trial, a memory sequence consisting of 6 sound samples is generated based on the stimulus library and presented sequentially. A 1.2-second delay interval is set after the memory sequence is presented. After the delay, a probe sample is presented. The probe sample is generated according to the "old stimulus / new stimulus" ratio and sampling rules to ensure that the conditions of each trial meet the preset distribution constraints.

[0136] S203. Collect the subject's judgment response and record the trial results. After the probe sample is presented, guide the subject to complete a two-choice judgment operation within a preset response window to indicate whether the probe sample belongs to the aforementioned memory sequence. The host computer records the subject's response results.

[0137] S204. Calculate the DMTS score and generate supervision labels. Based on the correctness of judgments across all trials, calculate the overall judgment accuracy to obtain the DMTS score. Use the DMTS score as a quantitative indicator representing the subject's auditory short-term memory ability, and write it into the task summary record as a supervision learning label for subsequent evaluation model construction.

[0138] S3. Perform rhythm imitation training and collect physiological signals. Divide the digital score into multiple segments that meet homogeneity constraints. Guide the subject to imitate the main melody by hitting a small drum, and use a greedy matching strategy to calculate the correspondence between the target and the actual hitting events. The subject is considered to have passed when they meet the passing condition twice consecutively for the same segment. Record training behavior data such as average segment duration and number of training sessions. Before and after the rhythm imitation training, collect the subject's resting pulse wave signal for 120 seconds as the pre-training baseline and post-training post-test data, respectively. The specific steps are as follows:

[0139] S301. Select the original score and construct a candidate rhythm material library. On the host computer, select several digital scores or melody fragments that can be used for rhythm imitation training, extract their rhythm and note event sequence information, and form a candidate training material set. The candidate materials are represented by an "event time point sequence," meaning each note / hit corresponds to a target occurrence time point.

[0140] S302. Segment and standardize the musical score according to homogeneity constraints to generate a set of musical segments. The candidate musical score is segmented into multiple training segments, and homogeneity constraints are applied to each segment to control the consistency of difficulty. The homogeneity constraints include at least 4 / 4 time signature, 60 beats per second, 2 measures, the shortest note being an eighth note, 6-12 notes, and the first note being on an accented beat. Based on the above constraints, each musical segment is converted into a standardized target rhythm template, and a corresponding target hit event timestamp sequence is generated.

[0141] S303. Configure the training process and music segment presentation strategy. In one training session, select 6 music segments from the standardized music segment set as training units. Set up a closed-loop process for each music segment: "loop playback while imitating the response - real-time evaluation - pass or fail". The host computer repeatedly plays the current music segment and opens the acquisition window for the subject's imitation input at the beginning of each playback to obtain the actual hitting event sequence of that attempt.

[0142] S304. Collect electronic drum striking events and construct an actual event sequence. The subject strikes the snare drum according to the main melody rhythm played in the musical segment. The electronic drum converts the striking trigger signal into an event timestamp and transmits it to the host computer. The host computer sorts the striking events of each attempt by time to form an actual event sequence, and associates and records it with the current musical segment number, the number of attempts, and the presentation start time to support subsequent matching and scoring.

[0143] S305. Establish the correspondence between target and actual events based on the matching strategy and calculate the pass criterion. For an imitation attempt of a training musical segment, the host computer obtains the time sequence of the target events for that musical segment and the time sequence of the actual events generated by the subject through the electronic drum / keyboard. The matching module then establishes the event correspondence and determines the pass criterion. The matching uses a greedy strategy: it processes the target events sequentially along the timeline. For each target event, it searches for available actual events within a 300-millisecond time window before and after its target time point. Actual events already occupied by preceding target events are excluded. From the remaining candidates, the actual event with the smallest time difference from the target event is selected as the corresponding event and marked as occupied, thus forming a one-to-one correspondence between target events and actual events. After matching, the system first checks event consistency, requiring no over-hitting, under-hitting, or missed matching. Then, based on the established correspondence, it calculates the time deviation between each target event and its corresponding actual event and calculates the standard deviation of this time deviation, requiring it to be less than 80 milliseconds. If event consistency is achieved and the standard deviation is less than 80 milliseconds, the attempt is considered successful. Otherwise, it will be judged as a failure and will proceed to the next round of repeated playback and imitation process.

[0144] S306. Determine mastery of a musical segment and record training behavior data according to the "two consecutive passes" rule. For the same musical segment, if the subject meets the passing criterion for a preset number of consecutive times, the segment is considered passed and the subject moves on to the next segment. If the consecutive passing condition is not met, the subject continues to repeat playback and imitation of the segment until it is passed. Record the number of repetitions of the musical segment during training and write it into the training summary record for subsequent modeling.

[0145] S307. Pre- and Post-Training Resting PPG Acquisition and Data Archiving. Before and after the rhythm imitation training, subjects are guided to perform resting measurements and 120-second PPG signals are acquired. The PPG signals are continuously sampled and recorded via the sensor's infrared channel at a sampling frequency of 800Hz, and the raw time-series data of the pre-training baseline and the post-training test are written to a storage medium.

[0146] S4. Physiological signal preprocessing and interval sequence construction. In this embodiment, the baseline and post-training photoplethysmography (PPG) wave data of the subjects are preprocessed to be consistent, and the pulse cycle is segmented by valley location, thereby constructing a pulse interval sequence as input for subsequent physiological feature extraction. Figure 2 The flowchart for the above pulse wave signal processing is as follows, and the specific implementation steps are as follows:

[0147] S401. Acquisition of Resting PPG Data Before and After Training and Confirmation of Parameter Consistency. The host computer reads the baseline data before training and the archived raw PPG time series data after training, and verifies that the sampling rate, channel type, and other acquisition parameters of the two data segments are consistent. If there are parameter inconsistencies, they are corrected or marked according to the unified system preset parameters to ensure the comparability of the pre- and post-test data, and the two raw signals are used as inputs for subsequent preprocessing.

[0148] S402. Bandpass filtering preprocessing to suppress noise and baseline drift. The raw PPG signal obtained in step S401 is subjected to bandpass filtering. The filter uses a third-order Butterworth bandpass structure with a passband range of 0.5–8 Hz to suppress low-frequency drift caused by body movement and respiration, while also reducing environmental electromagnetic interference and high-frequency noise such as electromyography, thereby obtaining a smooth target PPG signal suitable for feature point detection. Figure 4 As shown, the low-frequency drift and high-frequency noise of the pulse wave signal after bandpass filtering are suppressed. The signal does not show a continuous increasing or decreasing trend, nor does it have short-term random jitter. It is locally uniform and smooth.

[0149] S403. Candidate Valley Detection Preparation. After bandpass filtering, the target PPG signal is scanned along the time axis to extract possible valley locations and form a candidate valley set. The time index and amplitude information of each local depression point are retained in the candidate set to provide input for subsequent effective valley screening and false detection suppression.

[0150] S404. Valley Location and False Detection Suppression Based on Interval Constraints and Adaptive Significance Thresholds. The candidate valley set obtained in step S403 is screened one by one: First, a minimum interval constraint is applied between adjacent valleys, requiring the time interval between two adjacent valleys to be no less than 0.5 seconds, to eliminate dense false detections caused by noise jitter or local oscillations. Second, the amplitude range corresponding to the 5%–95th percentile is calculated based on the amplitude distribution of the target PPG signal, and an adaptive significance threshold is constructed using 10% of this amplitude range. Only candidate points whose valley concavity significance reaches this threshold are retained to reduce false valleys caused by small amplitude fluctuations. After the above constraints, the final valley sequence is obtained and used as the boundary point set for pulse cycle segmentation. Figure 5As shown, the final valley sequence corresponds one-to-one with the actual valley, with no false detections or missed detections.

[0151] S405. Construct a pulse interval sequence and output intermediate results for subsequent feature extraction. Based on the final trough sequence, calculate the time difference between adjacent troughs to obtain the pulse interval sequence. Simultaneously output the corresponding intermediate result information, including a list of trough timestamps, the number of valid cycles, and basic quality markers. Use the pulse interval sequence as input data for physiological feature extraction in step S5.

[0152] S5. Physiological feature extraction and difference feature construction. In this embodiment, based on the pulse interval (IBI) sequence constructed in step S4, abnormal cycle cleaning is first performed to reduce the impact of octave / half-frequency errors caused by body motion artifacts and false trough detection. Then, time-domain and frequency-domain pulse rate variability features are extracted from the effective cycle set to form the pre-training physiological feature vector and the pre-training difference feature vector, respectively. Figure 3 The flowchart for processing the above pulse interval sequence is as follows:

[0153] S501. Cleaning the IBI sequences based on physiological range constraints and relative median mutation constraints. Anomaly removal is performed on the IBI sequences obtained in step S4, including at least: (1) Physiological range constraints: IBIs are limited to a physiologically reasonable range of 0.33–1.50 seconds (corresponding to approximately 40–180 times / minute). Intervals exceeding this range are considered abnormal and removed; (2) Relative median mutation constraints: The median of the IBI sequence is used as a robust reference. If the relative deviation of an IBI from the median exceeds 30%, or the relative change between two adjacent IBIs exceeds 30%, the abnormal interval is marked and removed to suppress frequency doubling or half-frequency errors caused by the trough of missed / false detections. After the above constraint processing, a set of effective IBIs and its corresponding set of effective periods are obtained.

[0154] S502. Construct an effective interval sequence for calculating temporal pulse rate variability. Form a continuous effective IBI sequence on the effective period set, and construct a difference sequence of adjacent effective IBIs for short-term fluctuation statistics.

[0155] S503. Extract temporal pulse rate variability features. Extract temporal features from the effective IBI sequence, including: (1) mean heart rate: used to characterize the overall pulse rate / heart rate level; (2) standard deviation (SDNN): used to characterize the overall variability and overall fluctuation level; (3) root mean square difference (RMSSD): used to characterize the short-term change intensity of adjacent intervals; (4) pNN50: statistically analyzes the proportion of adjacent effective IBI differences with an absolute value exceeding 50 milliseconds, used to characterize the frequency of significant fluctuations in adjacent intervals. The above features are calculated independently in the resting period before training and the resting period after training.

[0156] S504. Extract frequency domain pulse rate variability features and calculate the frequency band ratio. To perform frequency domain analysis, the unequal interval IBI sequence is first converted into an equal time interval sequence. The implementation method uses cubic spline interpolation and resampling to 4 Hz. Then, the equal interval sequence is subjected to mean removal and windowing, and then power spectrum estimation is performed. The Welch method is used, with a window length of 64 seconds and 50% overlap between windows. The following are calculated on the power spectrum according to the preset frequency bands: (1) Low frequency power (LF): the frequency band is set to 0.04–0.15 Hz; (2) High frequency power (HF): the frequency band is set to 0.15–0.40 Hz; (3) Low-high frequency power ratio (LF / HF): used to characterize the relative proportion of low-frequency and high-frequency power components.

[0157] S505. Construct the pre-training feature vector and the pre-training difference feature vector. Perform steps S501–S504 for the pre-training resting segment and the post-training resting segment respectively to obtain the pre-training physiological feature vector and the post-training physiological feature vector. Further, for the same subject, subtract corresponding features according to their feature dimensions to construct the post-training difference feature vector relative to the pre-training vector. Finally, organize and store the pre-training feature vector and the difference feature vector in a unified feature order, and archive them together with the subject identifier and experimental condition identifier for use in the evaluation model training and inference in step S6.

[0158] S6. Evaluation Model Construction, Training, and Inference. In this embodiment, physiological characteristics and training behavioral characteristics obtained from the host computer are used as input, and the Auditory Delay Matching to Sample (DMTS) task score is used as the supervision label to construct a regression model to achieve a quantitative assessment of the subject's auditory short-term memory ability. The model input includes at least the pre-training PPG physiological feature vector, the pre- and post-training difference feature vector, the average number of musical phrase attempts, and the average number of musical phrase attempts and training times, among other behavioral characteristics. By introducing smooth fitting for features that may have nonlinear effects during the modeling stage, and combining regularization constraints to control model complexity, the generalization performance is improved. The specific implementation steps are as follows:

[0159] S601. Construct training samples and feature input sets. For each subject, generate modeling samples based on one complete experimental procedure, and establish a one-to-one correspondence between the samples and the corresponding DMTS scores. The input features include at least: (1) Pre-training PPG physiological feature vector: take the subject's resting physiological features before entering training, to reflect the individual's baseline physiological state; (2) Pre- and post-training physiological difference feature vector: take the change in physiological features of the same subject after training relative to before training, to reflect the state changes caused by training / task load; (3) Rhythm imitation training behavior features: including the average number of musical phrase attempts and the number of training sessions, to characterize training performance and learning efficiency.

[0160] S602. Determine the supervised learning regression modeling objective and label definition. Use the DMTS score obtained in step S2 as the supervised learning label, serving as the target variable characterizing the subject's auditory short-term memory ability. Establish a mapping relationship from "PPG baseline features + PPG difference features + training behavior features" to the DMTS score for regression prediction and individualized assessment.

[0161] S603. Construct a regression model structure and introduce a smoothing fitting mechanism. Construct an interpretable supervised learning regression model to fit the above mapping relationship. For each training iteration, a smoothing function is introduced into the model for fitting. The smoothing function uses cubic splines, enabling the model to depict the decreasing trend of returns as the training amount increases. For the remaining features, linear terms are used for direct modeling.

[0162] S604. Suppress overfitting and determine model complexity through regularization constraints. To avoid overfitting caused by excessive freedom of the smoothing term or a large number of feature dimensions, regularization constraints are introduced during model training: on the one hand, a curvature penalty is applied to the smoothing term to limit function fluctuations; on the other hand, a penalty is applied to the parameter amplitude to suppress overfitting to noise. Simultaneously, cross-validation is used to select the regularization strength and smoothing complexity parameters, achieving a balance between the model's fitting ability on the training set and its generalization performance on the validation set.

[0163] S605, Inference Output for New Subjects. For new subjects, after completing one training session and collecting PPG data before and after training, the system extracts their pre-training PPG physiological feature vector, pre-training difference vector, and training behavioral features, and processes them according to the same feature order and scaling parameters determined during the training phase. This input is fed into the final evaluation model, outputting a predicted DMTS score as a quantitative assessment result of the new subject's auditory short-term memory ability. The system stores the inference output along with the corresponding input features, timestamp, and subject identifier for subsequent tracking analysis and model iteration updates.

[0164] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0165] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A quantitative assessment method for auditory short-term memory based on snare drum rhythm imitation training, characterized in that, Includes the following steps: S1. Construct an experimental scenario and testing system. The testing system includes a host computer for presenting stimulus materials, a PPG sensor for acquiring photoplethysmography (PPG) signals, and an electronic drum for acquiring performance signals. S2. Perform the Auditory Delayed Sample Matching Task (DMTS). The DMTS includes a preset number of trials, a sound stimulus library consisting of a preset number of samples, and new and old stimuli configured in a preset ratio. In each trial, a sequence of several samples is presented first, and after a preset delay, a probe sample is presented. The subject judges whether the probe sample belongs to the aforementioned sequence. The DMTS score is obtained based on the judgment accuracy of all trials and serves as a supervisory label characterizing the subject's auditory short-term memory ability. S3. Perform rhythm imitation training and collect physiological signals. Divide the digital score into multiple musical segments that meet homogeneity constraints. Guide the subject to imitate the main melody by hitting the small drum. Use event-level matching and alignment strategies to calculate the correspondence between the target and the actual hitting events. When the subject correctly hits the same musical segment a preset number of times, the musical segment is judged to be passed. Record the training behavior data of the average musical segment time and the number of training sessions. Before and after the rhythm imitation training, collect the subject's resting pulse wave signal for a preset duration as the pre-training baseline and post-training PPG data, respectively. S4. Construct physiological signal preprocessing and interval sequences, perform bandpass filtering on the baseline before training and the post-test PPG data after training to eliminate noise, perform valley search on the filtered signal, and suppress false detection based on the minimum interval constraint between adjacent valleys and the adaptive significance threshold to construct the pulse interval IBI sequence. S5. Construct physiological feature extraction and difference features, apply physiological range constraints and relative median mutation constraints to the pulse interval sequence, remove abnormal cycles and artifacts, extract pulse rate variability features on the effective cycle set, and form physiological feature vector before training and physiological feature difference vector before and after training respectively. S6. Construct an evaluation model and train and infer it. Use the pre-training PPG feature vector, the difference feature vector before and after training, the average number of musical phrase attempts and the number of training sessions as input features, and the DMTS score as the label to construct a supervised learning regression model. Based on the supervised learning regression model, output the quantitative evaluation results of the new subject's auditory short-term memory.

2. The method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training according to claim 1, characterized in that, Step S1 specifically includes: S101. Construct experimental scenarios and equipment deployment, and build a test system in the preset experimental environment; S102. Establish a multimodal data acquisition link and synchronization marking mechanism, configure the communication and event input link between the host computer and the PPG sensor and electronic drum, set up a unified experimental process control module on the host computer side, perform state management of the experimental stage, and generate corresponding stage numbers and timestamps for each stage, so that PPG continuous sampling data, electronic drum hitting event sequence and task log can be aligned and fused based on timestamps or stage numbers; S103. Configure the stimulus presentation and human-computer interaction module, complete the configuration of auditory stimulus playback parameters on the host computer, including stimulus library loading and volume calibration, and configure the subject response input method and recording mechanism. S104. Conduct trial runs and quality checks. Before formal data collection, guide the subjects to complete a short trial run to verify whether the auditory stimulus is presented normally, whether the subject's response is fully recorded, whether the electronic drum hitting event is triggered stably, and whether the PPG signal has disconnection, saturation, or obvious motion artifacts. When the signal quality or link status is found to be unsatisfactory, readjust the sensor wearing status or device connection parameters, and repeat the trial run until the quality check is passed.

3. The method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training according to claim 1, characterized in that, Step S2 specifically includes: S201. Establish an auditory stimulus library and configure trial parameters. Establish an auditory stimulus library on the host computer side. The auditory stimulus library includes a preset number of distinguishable sound samples. Configure the trial scale of the DMTS task to a preset number of trials and set the trial ratio of "old stimulus / new stimulus" to a preset ratio. Wherein, "old stimulus" refers to the memory sequence of the probe sample belonging to the current trial, and "new stimulus" refers to the memory sequence of the probe sample not belonging to the current trial. S202. Generate a trial sequence and execute the stimulation program. For each trial, generate a memory sequence consisting of a preset number of sound samples based on the auditory stimulus library and present them sequentially. After the memory sequence is presented, set a preset delay interval. After the delay interval, present the probe samples. The probe samples are generated according to the ratio of "old stimulus / new stimulus" and sampling rules. S203. Collect the subject's judgment response and record the trial results. After the probe sample is presented, guide the subject to complete the two-choice judgment operation within the preset response window to indicate whether the probe sample belongs to the aforementioned memory sequence. The host computer records the subject's response results. S204. Calculate the DMTS score and generate a supervision label. Based on the overall accuracy of the judgments in all trials, the DMTS score is obtained as a quantitative indicator of auditory short-term memory ability and a supervision learning label.

4. The method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training according to claim 1, characterized in that, Step S3 specifically includes: S301. Construct a candidate rhythm material library. Select several digital scores that can be used for rhythm imitation training on the host computer side, extract their rhythm and note event sequence information, and form a candidate training material set. The candidate materials are represented by "event time point sequence", with each note / hit corresponding to a target occurrence time point. S302. The score is segmented and standardized according to homogenization constraints to generate a set of musical segments; the candidate score is segmented into multiple training segments, and homogenization constraints are applied to each segment to control the consistency of difficulty. The homogenization constraints include at least the beat structure constraint, tempo constraint, segment length constraint, minimum rhythm unit constraint, event quantity constraint, and starting accent position constraint; based on satisfying the above constraints, each segment is converted into a standardized target rhythm template, and a corresponding target hit event timestamp sequence is generated. S303. Configure the training process and music segment presentation strategy. In one training session, select a preset number of music segments from the standardized music segment set as training units. Set a closed-loop process of "looping playback while imitating the answer - real-time evaluation - whether it passes" for each music segment. The host computer repeatedly plays the current music segment and opens the acquisition window for the subject's imitation input at the beginning of each playback to obtain the actual hitting event sequence of the attempt. S304. Collect electronic drum hitting events and construct an actual event sequence. The subject hits the small drum according to the main melody rhythm of the music segment. The electronic drum converts the hitting trigger signal into an event timestamp and transmits it to the host computer. The host computer sorts the hitting events of each attempt by time to form an actual event sequence, and records it in association with the current music segment number, the number of attempts, and the presentation start time to support subsequent matching and scoring. S305. Based on the matching strategy, establish the correspondence between the target and the actual events and calculate the passing criteria. For a single attempt, input the target event sequence and the actual event sequence into the matching module. Use the event matching strategy on the time axis to select the actual event with the smallest time difference and satisfying the monotonic order constraint for each target event on the time axis, thereby establishing a one-to-one correspondence. And calculate the correctness index of the attempt based on the correspondence. S306. According to the "passing a preset number of times consecutively" rule, determine the mastery of the musical segment and record the training behavior data. For the same musical segment, if the subject continuously reaches the preset number of times and meets the passing criterion, the musical segment is determined to be passed and the subject switches to the next musical segment. If the continuous passing condition is not met, the subject continues to repeat the playback and imitation of the musical segment until the subject passes. During the training process, the number of times the musical segment is repeated is recorded and the number of repetitions is written into the training summary record for subsequent modeling. S307. Pre- and post-training resting PPG data acquisition and archiving: Before and after the rhythm imitation training, the subjects are guided to perform resting measurements and PPG signals of a preset duration are collected. The PPG signals are continuously sampled and recorded through a preset sensor channel at a preset sampling frequency, and the original time series data of the pre-training baseline and the post-training test are written into the storage medium.

5. The method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training according to claim 4, characterized in that, In step S305, the criteria for judgment include at least: (1) Event consistency constraints are used to suppress excessive knocking, insufficient knocking, or missed matching; (2) Time error constraint, used to limit the time deviation between the target event and the corresponding actual event to not exceed a preset threshold; If the above criteria are met, the attempt is deemed successful; otherwise, it is deemed unsuccessful and proceeds to the next round of repeated playback and imitation.

6. The method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training according to claim 1, characterized in that, Step S4 specifically includes: S401. Obtain PPG data and confirm parameter consistency: Obtain the original PPG time series data archived before and after training as preprocessing input, and confirm that the collection parameters are consistent; S402. Bandpass filtering preprocessing to suppress noise and baseline drift: Preprocessing is performed on the acquired raw PPG signal by using a bandpass filter to filter the signal. The bandpass frequency band is set to a preset frequency band range to suppress low-frequency drift caused by body movement and respiration, as well as high-frequency noise caused by environmental electromagnetic and electromyographic factors, thereby obtaining a smooth target PPG signal that can be used for feature point detection. S403. Preparation for candidate valley detection: Perform candidate extreme value scanning on the filtered signal to obtain the set of candidate valley locations. S404. Based on the set constraints, the candidate valley location set is filtered for validity to obtain the final valley sequence, which is used as the boundary point set for pulse cycle segmentation. The preset constraints include at least: (1) using the minimum interval constraint between adjacent valleys to suppress dense false detections caused by noise or local oscillations; (2) constructing an adaptive significance threshold based on the amplitude distribution of the target PPG signal, and retaining only the candidate valleys that meet the significance requirements to reduce the detection of false valleys caused by small amplitude fluctuations. S405. Based on the final trough sequence, calculate the time difference between adjacent troughs to obtain the pulse interval sequence; at the same time, output intermediate result information that matches the IBI sequence, and use the IBI sequence as input for physiological feature extraction; the intermediate structure information includes trough timestamp, effective cycle number and basic quality marker.

7. The method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training according to claim 1, characterized in that, Step S5 specifically includes: S501. Clean the IBI sequences and remove abnormal periods and artifacts in the IBI sequences by means of physiological range constraints and relative median mutation constraints to obtain the effective IBI set and the corresponding effective period set; S502. Construct a time-domain analysis effective sequence, form a continuous effective IBI sequence on the effective period set, calculate the difference between adjacent effective IBIs to form a difference sequence, and mark the quality of records with insufficient effective sample quantity or too many discontinuities. S503. Calculate the time-domain PRV features on the effective IBI sequences, wherein the time-domain PRV features include at least the average heart rate, standard deviation, root mean square difference, and the proportion of adjacent effective IBI sequences whose absolute values ​​exceed 50 ms. S504. Extract frequency domain pulse rate variability features and calculate the frequency band ratio. Based on the effective IBI sequence, construct an equal time interval sequence for frequency domain analysis and perform power spectrum estimation to obtain low-frequency power, high-frequency power and low-high frequency power ratio. S505. Construct the pre-training feature vector and the difference feature vector before and after training. Extract physiological feature vectors from the pre-training resting segment and the post-training resting segment respectively to obtain the pre-training physiological feature vector and the post-training physiological feature vector.

8. The method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training according to claim 1, characterized in that, In step S505, the difference between the feature vector after training and the feature vector before training is calculated for the same subject to form a physiological feature difference vector before and after training; finally, the feature vector before training and the difference vector are organized and stored in a unified feature order.

9. The method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training according to claim 1, characterized in that, Step S6 specifically includes: S601. Construct a training sample and feature input set. For each subject, summarize and form a modeling sample that corresponds one-to-one with their DMTS score. The input features include at least the pre-training PPG physiological feature vector, the pre- and post-training difference feature vector, the average number of musical phrase attempts, and the number of training sessions. S602. Determine the modeling objectives and labels, using DMTS scores as supervised learning labels and target variables, and establish a mapping relationship between input features and DMTS scores; S603. Constructing a regression model and smoothing fitting: Construct a supervised learning regression model, and use spline functions, kernel smoothing or other differentiable smoothing basis functions to fit nonlinear features, and use linear terms to model linear features. S604, Regularization Constraints and Model Complexity Control: Introducing regularization constraints to suppress overfitting, adjusting the regularization strength or smoothing the degrees of freedom to balance the model's fitting ability and generalization performance. S605. New Subject Inference Output: Input relevant features of the new subject, output the predicted DMTS score as a quantitative evaluation result through the model, and store the relevant data.

10. The method for quantitative assessment of auditory short-term memory based on snare drum rhythm imitation training according to claim 1, characterized in that, In step S604, the regularization includes at least a penalty term for the curvature of the smooth function or a penalty term for the magnitude of the parameter, in order to limit function fluctuations and improve generalization ability.

Images