A Personalized Intervention Method and System for Reconsolidating Substance Addiction Memories Based on AI Music Generation Technology
By using AI music generation technology combined with multimodal physiological signal analysis, personalized intervention for the reconsolidation of addictive memories was achieved, solving the problems of insufficient memory activation and strong dependence in traditional methods, and improving the effectiveness of the intervention.
Patent Information
- Application Number
- CN202611247064.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-17
- Publication Date
- 2026-11-10
AI Technical Summary
Traditional addiction intervention methods struggle to personalize the activation and modification of addiction-related memories, rely on human experience, lack objective quantitative standards and real-time feedback, resulting in high relapse rates and an inability to achieve precise and lasting intervention effects.
By employing AI music generation technology and collecting and analyzing multimodal physiological signals, personalized memory narrative audio and emotional music are generated. Music parameters are adjusted in real time to form an adaptive intervention closed loop, realizing personalized intervention based on the theory of memory consolidation.
It enables personalized, real-time intervention for addictive memories, reduces relapse rates, improves the accuracy and persistence of interventions, and overcomes the dependence and impersonalization problems of traditional methods.
Smart Images

Figure CN123059243A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of drug addiction technology, specifically to a personalized method and system for reconsolidating substance addiction memories based on AI music generation technology. Background Technology
[0002] Substance addiction is a chronic, relapsing brain disorder characterized by compulsive drug seeking and use, and a high relapse rate. Traditional addiction interventions, such as cognitive behavioral therapy, motivational interviews, and medication, while effective to some extent, generally suffer from high relapse rates. The core challenge lies in the difficulty of persistently and effectively correcting the "pathological memory traces" deeply embedded in the patient's memory and closely linked to addiction cues. These memories often possess high emotional arousal and automaticity, making them easily activated upon exposure to substance-related cues, triggering intense cravings and leading to relapse. Therefore, precisely targeting and correcting these specific addiction-related memories is crucial for improving intervention effectiveness and reducing relapse rates.
[0003] Memory reconsolidation theory offers a theoretical framework for effectively addressing these challenges. This theory posits that when a consolidated memory is reactivated, it is not simply reproduced but temporarily enters an unstable state (the reconsolidation window, typically lasting from minutes to hours). During this period, the memory is susceptible to interference or updating before finally stabilizing and being stored again. Intervention paradigms based on this theory aim to achieve a lasting reduction in the emotional valence and motivational salience of memories by precisely activating specific addictive memories and applying specific interventions within their unstable window.
[0004] However, translating memory reconsolidation theory into standardized, scalable clinical applications faces several bottlenecks: First, the personalization and specificity of memory activation are severely lacking. Current interventions mostly rely on therapist-guided imaginary exposure or the use of standardized pictures and videos. Due to the lack of in-depth matching with the patient's unique addiction experience (such as specific environmental, emotional, and sensory details), memory activation is not personalized enough, the intensity is insufficient, or the target is misaligned. This fails to ensure that addictive memories are effectively activated and enter an unstable reconsolidation state, causing subsequent interventions to lose their target.
[0005] Second, the intervention process relies heavily on human experience and lacks objective quantitative standards. In the traditional model, the judgment of the degree of memory activation and the timing of intervention depend mainly on the therapist's subjective observation and clinical experience. This not only leads to poor intervention consistency and high labor costs, but also makes it difficult to achieve standardized and quantifiable precision medicine.
[0006] Third, there is a lack of adaptive closed-loop control capabilities based on real-time feedback. Because individuals exhibit significant differences in their physiological and psychological responses to interventions, an ideal intervention program should possess dynamic adjustment capabilities. However, existing methods mostly employ an open-loop model of "preset program - linear execution," which cannot dynamically optimize based on the patient's real-time physiological arousal level (such as heart rate variability and skin conductance response) and subjective feelings, resulting in a significant reduction in intervention effectiveness.
[0007] Addressing the aforementioned technical challenges, music intervention, as a non-invasive and highly tolerable neuromodulation method, exhibits unique biological advantages. From a neuroscience perspective, music processing extensively involves key brain regions involved in addiction, such as the prefrontal cortex, nucleus accumbens, amygdala, and hippocampus. Specifically: First, music can precisely regulate emotional valence and arousal index. Music can directly affect the limbic system through the auditory pathway, regulating the release of cortisol and dopamine. Specific musical parameters (such as rhythm, harmony, and spectrum) can significantly alter the activity level of the autonomic nervous system, for example, reducing skin conductance response (SCR) and heart rate, and increasing heart rate variability (HRV), thereby effectively reducing the high physiological arousal index after memory activation and reversing negative emotional valence.
[0008] Secondly, music can reconstruct the sensory context of memories. Compared to static cues, music possesses temporal characteristics, enabling it to combine with narrative content and construct an immersive "situational memory" experience. This multimodal stimulation not only activates addiction memory traces more deeply but also, during reconsolidation interventions, covers the high-reward and high-arousal emotional characteristics of the original memories associated with drug use with new, peaceful, and controllable emotional experiences, achieving a "rewriting" of the emotional aspect of the memory content rather than simply "erasing it." Simultaneously, as a physical sound wave signal, music's parameters (such as BPM, pitch, and reverberation) can be digitally adjusted in real time. This makes adaptive closed-loop intervention based on physiological feedback (such as real-time EEG β / θ ratio and RMSSD) possible, solving the problem that traditional therapies cannot match the patient's state in real time.
[0009] With the development of digital health technology, although tools combining sensors and psychological intervention have emerged, existing technologies still have significant shortcomings: First, they are highly fragmented, often using music as an independent relaxation tool without deep coupling with the "activation-intervention" time sequence logic of memory reconsolidation; second, they are highly generic, using mostly standardized audio materials and lacking algorithms that utilize artificial intelligence to personalize the generation of individual addictive memories; third, they have a low degree of closed-loop integration, although some systems use heart rate variability to guide users to relax, they have not organically combined AI-generated content, multimodal real-time assessment, and dynamic decision-making models to form a complete automated intervention loop.
[0010] In summary, the current field of addiction intervention urgently needs a system and method that can deeply integrate AI-generated personalized content, real-time monitoring of multimodal physiological and psychological states, and memory consolidation theory. Summary of the Invention
[0011] To address the aforementioned technical problems, this invention provides a personalized substance addiction memory reconsolidation intervention method and system based on AI music generation technology. It constructs an intelligent closed loop that integrates personalized content generation, synchronous collection and fusion analysis of multimodal data, and adaptive intervention decision-making to achieve efficient and precise intervention based on memory reconsolidation theory.
[0012] The technical solution adopted is as follows: On the one hand, this invention provides a personalized intervention method for reconsolidating substance addiction memories based on AI music generation, comprising the following steps: In a resting state, users' multimodal physiological signals and baseline subjective scores were collected as baseline data to calculate benchmark relaxation parameters. P base ; By using natural language processing methods to extract key contextual and emotional features from users' descriptions of their addiction experiences, personalized memory narrative audio and corresponding emotional music are generated. Play memory narrative audio and emotional music to activate the user's addictive memories, and simultaneously collect the user's real-time physiological signals and subjective ratings; The physiological arousal index I is calculated based on real-time physiological signals and baseline data. phy and memory activation strength index I act ; Based on memory activation strength index I act Initial music intervention parameters are generated based on the preset wake-up threshold; Combined with the physiological arousal index I phy And the expected target curve I expected (t), adjust the audio playback parameters in real time during the intervention process until the physiological arousal index I is reached. phy Reduced to the expected target curve I expected (t) Below, the audio playback parameters are locked as the baseline relaxation parameter P. base Maintain this until the intervention ends.
[0013] The baseline data collected included: physiological signals, including at least skin conductance, electrocardiogram (ECG) and electroencephalogram (EEG) signals, continuously collected for at least 3 minutes while the user remained at rest; and baseline emotional arousal (A) at rest was collected using a baseline self-assessment questionnaire. base Basic subjective desire C base And basic physical sensation score.
[0014] Specifically, the physiological arousal index I phy The physiological arousal index I was calculated by weighting and fusing the normalized changes in skin conductance, reverse heart rate variability, and changes in the β / θ wave power ratio of electroencephalography. phy The calculation method is as follows: ,
[0015] in, This refers to the change in skin electrical activity. This represents the difference in reverse heart rate variability. This represents the change in the power ratio of β / θ waves in brainwaves. For the preset weighting coefficients, and .
[0016] Specifically, the memory activation intensity index I act Through the physiological arousal index I phy When combined with subjective changes, its calculation method is as follows: , , , Where ΔA represents the change in subjective arousal, and ΔC represents the change in subjective craving. A act To enhance emotional arousal after activating addictive memories; C act To determine the subjective craving level after activating addictive memories; α is the physiological weight, β is the subjective weight, and α+β=1.
[0017] Furthermore, the specific method for adjusting audio playback parameters in real time during the intervention process is as follows: Set the physiological arousal index I phy Expected target curve I expected (t); With a fixed regulation cycle, calculate the real-time physiological arousal index I for each regulation cycle. phy (t) and the expected target curve I expected The deviation δ of (t); If the deviation δ > 0, the physiological arousal index can be adjusted by fine-tuning the attenuation coefficient or introducing a weak negative feedback gain. I phy (t) Accelerate the approach to the expected target curve; If the deviation δ≤0, lock the current audio playback parameters as the baseline relaxation parameters and maintain them until the intervention ends.
[0018] Furthermore, when the real-time physiological arousal index At that time, the attenuation index λ is used as a dynamic adjustment parameter for audio playback; The exponential decay model used:
[0019] The expected target curve used :
[0020] Where P(t) is time... t Intervention parameters, I base The baseline emotional arousal index; P 0 represents the initial music intervention parameter, and λ represents the attenuation coefficient.
[0021] Furthermore, the method also includes a safety boundary control step: setting a safety range for music parameters [P] min ,P max This ensures that the music parameters during the dynamic adjustment process remain within the set safe range. If they exceed the safe range, the circuit breaker mechanism is triggered and the system switches to safe mode.
[0022] On the other hand, the present invention also provides a personalized substance addiction memory reconsolidation intervention system based on AI music generation, the system comprising: a hardware device layer, a software platform layer, and a data / algorithm service layer; The hardware device layer includes: Multimodal physiological data acquisition equipment is used to monitor the user's physiological arousal state during intervention; Interactive and output devices are used to present users with personalized audio and video activation materials and intervention content, and to receive subjective feedback from users; The software platform layer is deployed on the user terminal and is configured to perform real-time data preprocessing, multimodal fusion, and local decision control, including: The interface module guides users through the process of inputting text about their addiction experiences, filling out a self-assessment scale, and controlling the start and stop of the intervention process using wizard-style logic. The AI-generated content management module transforms semi-structured user input into structured, multimodal intervention content. The data fusion and decision-making module is configured to perform low-latency real-time processing; The data / algorithm service layer is deployed on a cloud server cluster and is configured to provide high-performance model inference, complex decision optimization, and full lifecycle data management services to the software platform layer of the user terminal, including: The algorithm model microservice module is configured to host and schedule deep learning models trained on a large number of samples. The intervention execution and optimization module is configured to dynamically generate and optimize intervention instructions based on advanced analytics results from the cloud. The data storage and management service is configured to provide data persistence and collaboration interfaces.
[0023] The AI generation and content management module includes: The Natural Language Understanding submodule has a built-in pre-trained language model based on the Transformer architecture, which is used to perform semantic parsing on the user's input description of addictive experiences, extract key content and emotional feature vectors, and generate feature maps representing addictive memories. The memory audio synthesis submodule dynamically adjusts the prosodic parameters of the synthesized speech based on the emotion tags in the feature map, generating personalized narrative audio with specific emotional intonation. The music generation and matching submodule configures an emotion-music feature mapping model, and matches or generates emotional music from the music feature library in real time based on the emotional tone of the narrative audio and user preferences. The content caching and management submodule is used to store the raw audio stream, intermediate feature data, and the final synthesized playback file generated during the session.
[0024] The data fusion and decision-making module includes: The signal preprocessing submodule is used to perform bandpass filtering, power frequency notch filtering, and noise reduction on the raw physiological electrical signals to eliminate motion artifacts and environmental interference. The feature extraction and time alignment submodule uses a sliding window mechanism to extract time-frequency domain features of physiological signals and aligns these features with the timestamps of user-reported events to construct a unified multimodal time series dataset. The memory activation state assessment submodule, based on fused multimodal data, calculates one or more memory activation intensity indices or physiological arousal indices in real time through machine learning models or rule-based expert systems. The dynamic intervention decision-making submodule generates intervention parameter adjustment instructions based on the preset intervention logic and memory activation intensity index during the memory consolidation intervention phase.
[0025] The algorithm model microservice module includes: The multimodal state recognition model, built on Transformer or spatiotemporal convolutional network, receives multimodal time-series data and self-assessment scale data uploaded by user terminals, performs end-to-end feature extraction and fusion calculation, and outputs memory activation intensity and addictive memory susceptibility assessment results. The intervention effect prediction model predicts the effect of different music parameter adjustment strategies on reducing the physiological arousal index after memory activation, based on the current distribution of users' physiological characteristics.
[0026] The intervention execution and optimization module includes: The complex decision-making logic unit combines the user's long-term historical data to correct the initial intervention parameter P0 initiated by the user terminal and generate a globally optimal solution that conforms to the current user's personalized characteristics. Real-time audio signal processing service, with its built-in digital signal processor, is used to receive music adjustment parameters and perform real-time resampling, time-frequency transformation and mixing on the original audio stream to generate an audio bitstream that meets the intervention requirements; The safety boundary controller has a built-in clinical safety rule base to verify the compliance of automatically generated intervention parameters, ensuring that the parameter adjustment range is always within the preset safety threshold.
[0027] The data storage and management services include: The full-cycle data warehouse adopts a distributed storage architecture and is configured to encrypt and store users' baseline data, raw physiological signals from each intervention, evaluation index curves, audio logs, and self-assessment results, supporting rapid data retrieval and backtracking.
[0028] The efficacy analysis module performs statistical analysis on population data or long-term individual data, explores the correlation between intervention parameters and efficacy, and uses newly generated labeled data to iteratively train and update the algorithm model. The therapist collaboration interface allows authorized professionals to remotely access the intervention process dashboard, view real-time physiological alerts, or issue manual intervention instructions.
[0029] The technical solution of the present invention has the following advantages: A. The method and system architecture of this invention are specifically optimized for psychological intervention in substance addiction and related mental disorders. Its core application scenarios cover the auxiliary treatment and relapse prevention of nicotine, alcohol and illicit drug dependence. Compared with traditional manual intervention, this invention completes a fully automated process from memory activation to reconsolidation intervention through the orderly data flow and control flow between the hardware device layer, software platform layer and data and algorithm service layer. It realizes digital quantitative assessment of the intervention process, personalized content generation and real-time adaptive adjustment based on biological signals. It provides an innovative technical path to solve the long-standing industry pain points in addiction treatment, such as poor compliance, strong subjective assessment and insufficient standardization of intervention methods.
[0030] B. The system of this invention has the ability to non-invasively, continuously and synchronously collect multiple physiological indicators of users during intervention sessions, providing a data foundation for objectively quantifying users' physiological arousal and emotional state. It can also perform timestamp alignment and fusion analysis of real-time physiological data streams and user-reported psychological self-assessment data (such as drug craving, emotional arousal, and memory vividness), and calculate the "memory activation state index" in real time through a built-in algorithm model, providing a quantitative basis for intervention decisions.
[0031] C. This invention is based on structured addiction experience text input by users. It uses natural language processing and sentiment computing technologies to automatically generate highly personalized narrative audio. Based on the sentiment-music feature mapping model, it matches or generates emotional music that is appropriate to the user, thereby achieving personalized memory activation and overcoming the shortcomings of existing technologies, such as difficulty in personalized activation and strong reliance on manual intervention.
[0032] D. During memory reconsolidation intervention, the system can dynamically and automatically adjust the acoustic parameters of music (such as rhythm, tempo, melody, and spectral characteristics) based on real-time assessment results, thereby achieving targeted and gentle modification of the emotional attributes of addictive memories and forming a real-time intervention closed loop of "assessment-adjustment-reassessment". This overcomes the technical defects of existing technologies, such as the difficulty of closed-loop adjustment, and realizes efficient and precise intervention based on the theory of memory reconsolidation. Attached Figure Description
[0033] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0034] Figure 1 This is a flowchart of the personalized substance addiction memory consolidation intervention method provided by the present invention; Figure 2 This is a block diagram of the personalized substance addiction memory consolidation intervention system provided by the present invention; Figure 3 This is a flowchart illustrating the specific execution process of the system of this invention. Detailed Implementation
[0035] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0036] like Figure 1As shown, this invention provides a personalized intervention method for reconsolidating substance addiction memories based on AI-generated music, comprising the following steps: [S01] Under resting conditions, collect the user's multimodal physiological signals and baseline subjective scores as baseline data to calculate benchmark relaxation parameters. P base The system automatically verifies device connectivity and monitors signal quality when the user wears a physiological signal acquisition device, ensuring effective acquisition of skin conductance, heart rate, and electroencephalogram (EEG) signals.
[0037] Once the user is seated and remains quiet, the multimodal data acquisition module is activated to continuously collect physiological signals at rest for at least 3 minutes, including at least electrical skin conductance (EDA), electrocardiogram (ECG), and electroencephalogram (EEG). Simultaneously, a baseline self-assessment questionnaire is presented on the interactive and output devices to collect the user's baseline emotional arousal level (A) at rest. base Basic subjective desire C base And basic bodily sensation scores, and calculate the baseline relaxation parameter P based on resting-state data. base .
[0038]
S02
[0039]
S03
[0040]
S04
[0041] To eliminate individual differences, a Min-Max normalization method based on baseline data is used to map each feature value to the [0,1] interval:
[0042] Where x is the eigenvalue, x min and x max These are the minimum and maximum values of the corresponding features in the baseline phase, respectively.
[0043] (2) Calculation of physiological arousal index Since RMSSD is negatively correlated with the arousal index, a polarity reversal is necessary. Define the physiological arousal index I. phy The calculation is as follows:
[0044] in, This refers to the change in skin electrical activity. This represents the difference in reverse heart rate variability. This represents the change in the power ratio of β / θ waves in brainwaves. For the preset weighting coefficients, and .
[0045] (3) Subjective and objective integrated assessment Subjective arousal change and subjective desire change Final memory activation strength index I act It is obtained by fusing physiological arousal index with subjective variability:
[0046] Where: is A act To enhance emotional arousal after activating addictive memories; C act To determine the subjective craving level after activating addictive memories; α is the physiological weight, β is the subjective weight, and α+β=1.
[0047] After the assessment is completed, the system enters a preset 10-minute memory reconsolidation window. During this period, the intervention content is stopped, only low-power physiological monitoring is maintained, and the user is prompted to remain seated to avoid external interference and promote the memory to enter an unstable reconsolidation state.
[0048]
S05
[0049] After the window period ends, the system enters a 30-minute memory consolidation intervention period. Based on the output I... act The indicators are used to generate an initial intervention parameter set P0, and the following dynamic decay strategy is executed: (1) Initial parameter mapping According to I act The size of the value is used to determine the initial music adjustment scheme through a preset mapping relationship:
[0050] The initial intervention parameter set P0 includes the music rhythm coefficient K. t Rhythm complexity coefficient K r and the melody smoothness coefficient K m ;T high and T low This is the preset wake-up threshold.
[0051]
S06
[0052] During the intervention, a control cycle is defined as 120 seconds. The following closed-loop control logic is executed: (1) Target trajectory planning The system is set to a physiological arousal index I. phy Expected target curve I expected (t), the curve decays exponentially over time, guiding the user to smoothly transition from the active state to the relaxed state.
[0053] (2) Real-time deviation calculation At each cycle node, the system calculates the real-time physiological arousal index I. phy The deviation δ between (t) and the target value:
[0054] (3) Dynamic adjustment of parameters Real-time physiological arousal index At that time, the exponential decay model is executed:
[0055] in, Based on real-time deviation Make dynamic adjustments.
[0056] The system of this invention presets a target curve for physiological arousal index during the intervention phase. In each regulation cycle, the system calculates the real-time physiological arousal index I. phy (t) Deviation from the expected target curve .like (Wake-up index too high), the system fine-tunes the attenuation coefficient. Alternatively, a weak negative feedback gain can be introduced to accelerate the approximation of the expected target curve; if (If the target curve is reached or slightly below the target curve, the system locks the current audio playback parameters or switches to maintenance mode to keep the physiological arousal index stable near the target curve until the intervention ends.)
[0057]
S07
[0058] like Figure 2 As shown, the present invention also provides a personalized substance addiction memory reconsolidation intervention system based on AI music generation. The system generally includes a hardware device layer, a software platform layer, and a data / algorithm service layer. Through the coordinated work of orderly data flow and control flow between each layer, a fully automated process from memory activation to reconsolidation intervention is completed.
[0059] I. Hardware Device Layer The system's hardware includes multimodal physiological data acquisition devices and interaction and output devices, used to collect users' physiological and psychological data, enable human-computer interaction, and present content.
[0060] 1. Multimodal physiological data acquisition equipment: used for real-time and continuous monitoring of the user's physiological arousal state during intervention. It includes at least: (1) Electrocardiogram (ECG) or photoplethysmography (PPG) sensor for collecting heart rate (HR) and heart rate variability (HRV) time-domain and frequency-domain indicators; (2) Electrodermal activity (EDA / GSR) sensor, used to measure the skin's electrical conductivity level as an indicator of sympathetic nerve excitability. This device is usually implemented in wearable form (such as wristband or chest strap) and connected to a computer terminal via a wireless communication module (such as Bluetooth or Wi-Fi).
[0061] (3) EEG sensor, used to collect the user's brain activity signals during the intervention process, can further analyze the neural characteristics related to emotional arousal and attention. The device can adopt a portable dry electrode type wearable headband to ensure comfort and ease of deployment.
[0062] 2. Interaction and Output Devices: These devices are used to present personalized audio and video activation materials and intervention content to users, and to receive subjective feedback from users. Serving as the main user interface and local computing core, they are typically smart mobile terminals such as smartphones, tablets, dedicated touchscreen devices, and touch or button input modules for users to input subjective ratings. The interaction and output devices run client software, guide users to input text or voice recordings of their addiction experiences, present real-time physiological data visualizations, present self-assessment scales to users at preset time points during the assessment and intervention process, collect subjective data such as emotion, craving level, and memory vividness, and include high-fidelity audio output devices (such as headphones with some noise isolation) to play high-fidelity personalized audio and music generated by the system, ensuring an immersive experience. All hardware devices support low-latency data transmission to ensure the time synchronization of physiological data collection and intervention content output, meeting the temporal accuracy requirements of memory consolidation interventions.
[0063] II. Software Platform Layer The software platform layer is deployed on the user terminal, serving as the core hub connecting the user interface, underlying hardware devices, and cloud-based data / algorithm services. This layer is configured to perform high-real-time data preprocessing, multimodal fusion, and local decision control, and includes the following core modules: 1. Interface Module This module is configured to provide a graphical user interface (GUI) that guides users through the input of texts about their addiction experiences, the completion of self-assessment scales, and the start and stop of the intervention process using wizard-style logic.
[0064] In terms of hardware communication, this module integrates a standardized Hardware Abstraction Layer (HAL) and connects to physiological sensors and audio playback devices via Bluetooth Low Energy (BLE). The module further includes: (1) Data receiving submodule: used to monitor and receive multi-channel physiological electrical signals from the hardware layer in real time; (2) Command sending submodule: used to encapsulate and send audio parameter configuration commands generated in the cloud or locally to the audio codec to realize playback control.
[0065] 2. AI Generation and Content Management Module This module is configured to transform semi-structured user input into structured multimodal intervention content, specifically including: (1) Natural Language Understanding Submodule: Built-in pre-trained language model based on Transformer architecture, configured to perform semantic parsing on the user's input description of addiction experience, extract key content (such as drug type, usage environment, associated tools, emotions, psychological and physical feelings) and emotional feature vectors (such as emotional valence, arousal index labels), and generate feature maps representing addictive memories.
[0066] (2) Memory audio synthesis submodule: Coupled with the text-to-speech (TTS) module, based on the emotion tags in the feature map, the prosodic parameters of the synthesized speech (such as male and female voices, pitch, speech rate, and pause duration) are dynamically adjusted to generate personalized narrative audio with specific emotional intonation in order to accurately activate the target addictive memory.
[0067] (3) Music generation and matching submodule: For the module of addictive memory activation, an emotion-music feature mapping model is configured to match or generate emotional music in real time from the music feature library based on the emotional tone of the narrative audio and user preferences.
[0068] The music generation and matching submodule structures and marks music parameters (such as rhythm, pitch, harmony, and instruments) for subsequent dynamic adjustments. For the module targeting addiction memory consolidation intervention, a built-in dynamic audio rendering module is configured to retrieve matching base music templates from the intervention material library based on real-time assessed physiological parameters (such as heart rate variability, skin conductance, and electroencephalography) and self-assessment data, and then apply parameters (such as rhythm coefficients) issued by the data fusion and decision-making module. (Spectrum centroid offset), real-time synthesis of intervention music segments adapted to the current physical and mental state.
[0069] The matching logic follows this pattern: the memory activation part uses high arousal and negative emotions to match dissonant intervals or fast rhythms, while the intervention part gradually adjusts towards harmony, stability, and slow rhythms.
[0070] (4) Content caching and management submodule: used to store the original audio stream, intermediate feature data and final synthesized playback file generated during the session, to ensure service continuity in offline environment.
[0071] 3. Data Fusion and Decision-Making Module This module is configured to perform low-latency real-time processing and includes: (1) Signal preprocessing submodule: Performs bandpass filtering, power frequency notch filtering and noise reduction on the raw physiological electrical signals to eliminate motion artifacts and environmental interference; (2) Feature extraction and time alignment submodule: It is configured to extract the time-frequency domain features of physiological signals using a sliding window mechanism, and align the features with the timestamps of the user's self-reported events to construct a unified multimodal time series dataset; (3) Memory Activation State Assessment Submodule: Based on the fused multimodal data, one or more “memory activation intensity indices” or “physiological arousal indices” are calculated in real time using a lightweight machine learning model (such as logistic regression or support vector machine) or a rule-based expert system. These indices comprehensively reflect the probability that addictive memories are successfully activated and enter an unstable state.
[0072] (4) Dynamic intervention decision-making sub-module: During the memory consolidation intervention stage, this sub-module generates intervention parameter adjustment instructions based on the preset intervention logic and the real-time assessed "memory activation intensity index". For example, when the arousal index is too high, the instructions may include "reduce the music speed by 20%", "reduce the density of notes by 20%", and "increase the smoothness of the melody by 10%".
[0073] III. Data / Algorithm Service Layer The data / algorithm service layer is deployed on a cloud server cluster and configured to provide high-performance model inference, complex decision optimization, and full lifecycle data management services to the terminal software platform layer. This layer includes the following core services: 1. Algorithm Model Microservice Module It is configured to host and schedule deep learning models trained on a large number of samples, and is made available to end users via a RESTful API or gRPC interface, specifically including: (1) Multimodal state recognition model: Based on Transformer or spatiotemporal convolutional network (ST-CNN), it is configured to receive multimodal time series data and self-assessment scale data uploaded by the terminal, and perform end-to-end feature extraction and fusion calculation (corresponding to I in the system usage process). act (Computational logic), outputting high-precision assessment results of memory activation intensity and susceptibility to addictive memories; (2) Intervention effect prediction model: It is configured to predict the effect of different music parameter adjustment strategies (such as changes in rhythm decay coefficient λ) on reducing the physiological arousal index after memory activation based on the current physiological characteristic distribution of the user, so as to provide a basis for decision-making.
[0074] 2. Intervention Implementation and Optimization Module This module is configured to dynamically generate and optimize intervention instructions based on advanced analytics results from the cloud, specifically including: (1) Complex decision logic unit: It is configured to combine the user's long-term historical data (such as past intervention response patterns and baseline drift characteristics) to correct the initial intervention parameter P0 initiated by the user terminal and generate a global optimal solution that conforms to the current user's personalized characteristics. (2) Real-time audio signal processing service: Built-in digital signal processor (DSP) configured to receive music adjustment parameters (such as tempo coefficient K). t(Spectrum centroid offset), the original audio stream is resampled in real time, time-frequency transformed and mixed to generate an audio bitstream that meets the intervention requirements; (3) Safety boundary controller: It has a built-in clinical safety rule library and is configured to perform compliance verification on automatically generated intervention parameters to ensure that the parameter adjustment range (such as volume change, frequency range) is always within the preset safety threshold to prevent overstimulation or induce discomfort. If necessary, it will trigger the stop mechanism and switch to safety mode.
[0075] 3. Data storage and management services This service is configured to provide highly reliable data persistence and collaboration interfaces, specifically including: (1) Full-cycle data warehouse: It adopts a distributed storage architecture and is configured to encrypt and store users' baseline data, original physiological signals of each intervention, evaluation index curves, audio logs and self-evaluation results, supporting rapid data retrieval and backtracking.
[0076] (2) Efficacy analysis module: It is configured to perform statistical analysis on group data or individual long-term data, explore the correlation between intervention parameters and efficacy (such as the rate of decline in desire), and use newly generated labeled data to iteratively train and update the algorithm model; (3) Therapist Collaboration Interface: Allows authorized professionals to remotely access the intervention process dashboard, view real-time physiological warnings, or issue manual intervention instructions through this interface to achieve "human-machine collaboration". Example
[0077] like Figure 3 As shown, this system is explained in detail using a specific intervention case of a patient with severe nicotine addiction.
[0078] Step 1: Patient Information and Initial Input The patient is a 45-year-old male, clinically diagnosed with severe nicotine dependence (FTND score ≥7). The patient's chief complaint is: "I must smoke immediately upon waking every morning. I feel chest tightness and anxiety, and only feel relaxed when I'm on the balcony watching the smoke billowing out the window. I used to enjoy listening to fast-paced rock music." The user inputs this experience description through the interactive module and puts on headphones, electrodes, and a wristband-type physiological sensor. The system automatically completes device connection verification and signal quality monitoring, confirming normal acquisition of electrical conductance analysis (EDA), electrocardiogram (ECG), and electroencephalogram (EEG) signals.
[0079] Step 2: Baseline Data Acquisition Once the user is seated and remains quiet, the system activates the multimodal physiological data acquisition module, continuously acquiring physiological signals at rest for 3 minutes. The acquired baseline data is shown in the table below:
[0080] The system calculates the corresponding baseline relaxation parameter P based on resting-state data. base = (0.6, 0.3, 0.2) (representing the slow tempo, low complexity, and high smoothness of the music, respectively).
[0081] The system plays the personalized audio combination generated in step 2 (including the patient's narrative of smoking on the balcony and matched high-arousal rock music). Playback is paused after 3 minutes to collect activation data and calculate the memory activation intensity index I. act .
[0082] (1) Physiological feature extraction and normalization The measured data during the activation phase are as follows: RMSSD=26ms, β / θ=2.1, SCR peak count=8, SCR amplitude=1.8μS.
[0083] The Min-Max normalization method based on baseline data was adopted to set the baseline ranges for each feature as follows: RMSSD [20,60] ms, β / θ [0.5, 3.0], SCR peak value [0, 10], and SCR amplitude [0,5].
[0084] The normalization results are as follows:
[0085] (2) Calculation of physiological arousal index Since RMSSD is negatively correlated with wakefulness, the inverse difference ΔRMSSD is used. inv =0.7 - 0.15 = 0.55. Set the weights. =0.4, =0.3, =0.3, substitute into the following formula: I phy =0.4×(0.6+0.28)+0.3×0.55+0.3×0.48=0.661 (3) Subjective and objective integrated assessment Subjective change: ΔA (subjective arousal change) = 8 − 2 = 6, ΔC (subjective desire change) = 9 − 1 = 8. Set weights α = 0.6, β = 0.4, and substitute into the formula: I act =0.6×0.661+0.4×(6+8) / 2=0.3966+2.8=3.1966 Step 4: Decision-making and Adaptive Intervention Phase After the system enters the 10-minute window, a 30-minute intervention is initiated. Based on I act =3.1966 and greater than the high threshold Thigh =2.5, the data fusion and decision module determines it to be a high-intensity activation, and generates initial parameters P0=(K t1 =0.95,K r1 =0.9,K m1 =0.85), meaning the music maintains a fast tempo and high complexity to match the current high arousal state.
[0086] During the intervention, closed-loop control is executed in 120-second cycles. At the 15-minute mark, time... t =15, set the attenuation coefficient λ=0.05, and execute the exponential decay formula: P(15) = (0.95, 0.9, 0.85)·e -0.75 +(0.6,0.3,0.2)·(1- e -0.75 ) The calculated values are P(15)≈(0.765, 0.584, 0.507). At this point, the real-time physiological arousal index I... phy (t)=0.35, expected target I expected (t)=0.30, deviation δ=0.05>ϵ (error tolerance threshold 0.02). The system determined that the arousal level was still too high, automatically increased λ to 0.07, accelerated the decay of music rhythm and complexity, and simultaneously collected subjective feedback at that moment, showing that the patient's emotional arousal level had dropped to 4 and subjective desire level had dropped to 5, verifying the effectiveness of the intervention.
[0087] Step 5: Steady-state locking and termination At the 28th minute of the intervention, the real-time physiological arousal index I... phy (t) = 0.18, lower than the baseline emotional arousal index I. base =0.2, the system collects subjective data again and displays the emotional arousal level A. end =2, Subjective Desire C end =3. The system triggers a locking mechanism, stops dynamic adjustment, and maintains the intervention parameter at P. base =(0.6,0.3,0.2) (slow tempo, minimalist music), until the 30-minute intervention ends.
[0088] Step 6: Validation of Results Final data showed that the patient's subjective desire level decreased from 9 at activation to 3 (a 67% decrease), and emotional arousal level decreased from 8 to 2 (returning to baseline calm levels). Physiological indicators showed that RMSSD recovered to 42 ms, and β / θ decreased to 1.0. The system generated a visualization report showing I act The results showed an exponential downward trend as music parameters were adjusted, and both subjective scores and physiological indicators decreased synchronously, confirming that the system can effectively reduce physiological hyperarousal and subjective cravings caused by addictive memories through dual guidance of subjective and objective data.
[0089] Any aspects not covered in this invention are applicable to existing technologies.
[0090] Obviously, the above embodiments are merely examples to clearly illustrate the technical solution of the present invention, and are not intended to limit the implementation. Those skilled in the art can make other variations or modifications based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the protection scope of the present invention.
Claims
1. A personalized intervention method for reconsolidating substance addiction memories based on AI-generated music, characterized in that, Includes the following steps: In a resting state, users' multimodal physiological signals and baseline subjective scores were collected as baseline data to calculate benchmark relaxation parameters. P base ; By using natural language processing methods to extract key contextual and emotional features from users' descriptions of their addiction experiences, personalized memory narrative audio and corresponding emotional music are generated. Play memory narrative audio and emotional music to activate the user's addictive memories, and simultaneously collect the user's real-time physiological signals and subjective ratings; The physiological arousal index I is calculated based on real-time physiological signals and baseline data. phy and memory activation strength index I act ; Based on memory activation strength index I act Initial music intervention parameters are generated based on the preset wake-up threshold; Combined with the physiological arousal index I phy And the expected target curve I expected (t), adjust the audio playback parameters in real time during the intervention process until the physiological arousal index I is reached. phy Reduced to the expected target curve I expected (t) Below, the audio playback parameters are locked as the baseline relaxation parameter P. base Maintain this until the intervention ends.
2. The personalized substance addiction memory reconsolidation intervention method based on AI music generation according to claim 1, characterized in that, The baseline data collected included: physiological signals, including at least skin conductance, electrocardiogram (ECG) and electroencephalogram (EEG) signals, continuously collected for at least 3 minutes while the user remained at rest; and baseline emotional arousal (A) at rest was collected using a baseline self-assessment questionnaire. base Basic subjective desire C base And basic physical sensation score.
3. The personalized substance addiction memory reconsolidation intervention method based on AI music generation according to claim 1, characterized in that, The physiological arousal index I phy The physiological arousal index I was calculated by weighting and fusing the normalized changes in skin conductance, the difference in reverse heart rate variability, and the changes in the β / θ wave power ratio of electroencephalography. phy The calculation method is as follows: , in, This refers to the change in skin electrical activity. This represents the difference in reverse heart rate variability. This represents the change in the power ratio of β / θ waves in brainwaves. For the preset weighting coefficients, and .
4. The personalized substance addiction memory reconsolidation intervention method based on AI music generation according to claim 3, characterized in that, The memory activation intensity index I act Through the physiological arousal index I phy When combined with subjective changes, its calculation method is as follows: , , , Where ΔA represents the change in subjective arousal, and ΔC represents the change in subjective craving. A act To enhance emotional arousal after activating addictive memories; C act To determine the subjective craving level after activating addictive memories; α is the physiological weight, β is the subjective weight, and α+β=1.
5. The personalized substance addiction memory reconsolidation intervention method based on AI music generation according to claim 1, characterized in that, The specific method for adjusting audio playback parameters in real time during the intervention process is as follows: Set the physiological arousal index I phy Expected target curve I expected (t); With a fixed regulation cycle, calculate the real-time physiological arousal index I for each regulation cycle. phy (t) and the expected target curve I expected The deviation δ of (t); If the deviation δ > 0, the physiological arousal index I can be adjusted by fine-tuning the attenuation coefficient or introducing a weak negative feedback gain. phy (t) Accelerates the approach to the expected target curve; If the deviation δ≤0, lock the current audio playback parameters as the baseline relaxation parameters and maintain them until the intervention ends.
6. The personalized substance addiction memory reconsolidation intervention method based on AI music generation according to claim 5, characterized in that, Real-time physiological arousal index At that time, the attenuation coefficient λ is used as a dynamic adjustment parameter for audio playback; The exponential decay model used: The expected target curve used : , Where P(t) is time... t Intervention parameters, I base The baseline emotional arousal index; P 0 represents the initial music intervention parameter, and λ represents the attenuation coefficient.
7. The personalized substance addiction memory reconsolidation intervention method based on AI music generation according to any one of claims 1-6, characterized in that, The method further includes a safety boundary control step: setting a safety range for music parameters [P] min , P max This ensures that the music parameters during the dynamic adjustment process remain within the set safe range. If they exceed the safe range, the circuit breaker mechanism is triggered and the system switches to safe mode.
8. A personalized substance addiction memory reconsolidation intervention system based on AI music generation, characterized in that, The system comprises: a hardware device layer, a software platform layer, and a data / algorithm service layer; The hardware device layer includes: Multimodal physiological data acquisition equipment is used to monitor the user's physiological arousal state during intervention. Interactive and output devices are used to present users with personalized audio and video activation materials and intervention content, and to receive subjective feedback from users; The software platform layer is deployed on the user terminal and is configured to perform real-time data preprocessing, multimodal fusion, and local decision control, including: The interface module guides users through the process of inputting text about their addiction experiences, filling out a self-assessment scale, and controlling the start and stop of the intervention process using wizard-style logic. The AI-generated content management module transforms semi-structured user input into structured, multimodal intervention content. The data fusion and decision-making module is configured to perform low-latency real-time processing; The data / algorithm service layer is deployed on a cloud server cluster and is configured to provide high-performance model inference, complex decision optimization, and full lifecycle data management services to the software platform layer of the user terminal, including: The algorithm model microservice module is configured to host and schedule deep learning models trained on a large number of samples. The intervention execution and optimization module is configured to dynamically generate and optimize intervention instructions based on advanced analytics results from the cloud. The data storage and management service is configured to provide data persistence and collaboration interfaces.
9. The personalized substance addiction memory reconsolidation intervention system based on AI music generation according to claim 8, characterized in that, The AI generation and content management module includes: The Natural Language Understanding submodule has a built-in pre-trained language model based on the Transformer architecture, which is used to perform semantic parsing on the user's input description of addictive experiences, extract key content and emotional feature vectors, and generate feature maps representing addictive memories. The memory audio synthesis submodule dynamically adjusts the prosodic parameters of the synthesized speech based on the emotion tags in the feature map, generating personalized narrative audio with specific emotional intonation. The music generation and matching submodule configures an emotion-music feature mapping model, and matches or generates emotional music from the music feature library in real time based on the emotional tone of the narrative audio and user preferences. The content caching and management submodule is used to store the raw audio stream, intermediate feature data, and the final synthesized playback file generated during the session.
10. The personalized substance addiction memory reconsolidation intervention system based on AI music generation according to claim 8, characterized in that, The data fusion and decision-making module includes: The signal preprocessing submodule is used to perform bandpass filtering, power frequency notch filtering, and noise reduction on the raw physiological electrical signals to eliminate motion artifacts and environmental interference. The feature extraction and time alignment submodule uses a sliding window mechanism to extract time-frequency domain features of physiological signals and aligns these features with the timestamps of user-reported events to construct a unified multimodal time series dataset. The memory activation state assessment submodule, based on fused multimodal data, calculates one or more memory activation intensity indices or physiological arousal indices in real time through machine learning models or rule-based expert systems. The dynamic intervention decision-making submodule generates intervention parameter adjustment instructions based on the preset intervention logic and memory activation intensity index during the memory consolidation intervention phase.
11. The personalized substance addiction memory reconsolidation intervention system based on AI music generation according to claim 8, characterized in that, The algorithm model microservice module includes: The multimodal state recognition model, built on Transformer or spatiotemporal convolutional network, receives multimodal time-series data and self-assessment scale data uploaded by user terminals, performs end-to-end feature extraction and fusion calculation, and outputs memory activation intensity and addictive memory susceptibility assessment results. The intervention effect prediction model predicts the effect of different music parameter adjustment strategies on reducing the physiological arousal index after memory activation, based on the current distribution of users' physiological characteristics.
12. The personalized substance addiction memory reconsolidation intervention system based on AI music generation according to claim 8, characterized in that, The intervention execution and optimization module includes: The complex decision-making logic unit combines the user's long-term historical data to correct the initial intervention parameter P0 initiated by the user terminal and generate a globally optimal solution that conforms to the current user's personalized characteristics. Real-time audio signal processing service, with its built-in digital signal processor, is used to receive music adjustment parameters and perform real-time resampling, time-frequency transformation and mixing on the original audio stream to generate an audio bitstream that meets the intervention requirements; The safety boundary controller has a built-in clinical safety rule base to verify the compliance of automatically generated intervention parameters, ensuring that the parameter adjustment range is always within the preset safety threshold.
13. The personalized substance addiction memory reconsolidation intervention system based on AI music generation according to claim 8, characterized in that, The data storage and management services include: The full-cycle data warehouse, employing a distributed storage architecture, is configured to encrypt and store users' baseline data, raw physiological signals from each intervention, assessment index curves, audio logs, and self-assessment results, supporting rapid data retrieval and backtracking. The efficacy analysis module performs statistical analysis on population data or long-term individual data, explores the correlation between intervention parameters and efficacy, and uses newly generated labeled data to iteratively train and update the algorithm model. The therapist collaboration interface allows authorized professionals to remotely access the intervention process dashboard, view real-time physiological alerts, or issue manual intervention instructions.
