Music teaching system based on artificial intelligence

Abstract

The invention discloses a music teaching system based on artificial intelligence, and particularly relates to the technical field of artificial intelligence, and the music teaching system comprises a neurocognitive adaptation module, a music DNA map construction module, a cognitive load monitoring module, an anti-AI dependence adjustment module and a multi-source data fusion unit. According to the music teaching system based on artificial intelligence, a dynamically evolved personalized teaching model is constructed through multi-modal data fusion and real-time analysis driven by artificial intelligence. The neurocognitive adaptation module is combined with electroencephalogram feature analysis and physiological signal monitoring to accurately capture cognitive preferences and ability bottlenecks of the learner; the music DNA map is based on the quantum enhancement modeling technology, the skill development trajectory is continuously updated and predicted, the AI teaching strategy can realize millisecond-level dynamic adjustment according to the neural feedback and behavior data of the learner, the skill mastering efficiency and the knowledge retention rate are remarkably improved, and the limitation of staticization and simplification of a traditional teaching system is broken through.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence, and particularly to a music teaching system based on artificial intelligence. Background Art

[0002] In the current music education field, both traditional teaching models and existing intelligent teaching technologies face significant challenges. Traditional music teaching systems have significant static and single characteristics. Under this teaching model, teachers mainly teach according to general teaching syllabuses and their own experiences, and it is difficult to accurately grasp the unique cognitive preferences and ability bottlenecks of each learner. For example, when teaching complex music theory knowledge or difficult performance skills, teachers cannot timely understand their understanding levels and learning difficulties based on students' electroencephalogram (EEG) characteristics and physiological data, and can only adopt a "one-size-fits-all" teaching method. This leads to uneven knowledge mastery among students during the learning process, slow skill improvement, and relatively low knowledge retention rate, unable to meet the personalized learning needs of students. Summary of the Invention

[0003] The main purpose of the present invention is to provide a music teaching system based on artificial intelligence, which can effectively solve the problem of inability to meet the personalized learning needs of students.

[0004] To achieve the above purpose, the technical solution adopted by the present invention is as follows: A music teaching system based on artificial intelligence, the music teaching system based on artificial intelligence is configured as: Neurocognitive Adaptation Module: Real-time collect learners' brain waves and physiological data through multi-modal biosensors, and analyze music cognitive characteristics using a deep neural network; Music DNA Map Construction Module: Generate a dynamic personal music ability gene map based on historical performance data and real-time learning behaviors; Cognitive Load Monitoring Module: Real-time evaluate the attention state using computer vision and micro-expression recognition technologies; Anti-AI Dependence Regulation Module: Dynamically adjust the AI assistance intensity according to the development level of creative thinking; Multi-source Data Fusion Unit: Perform cross-modal feature alignment and joint modeling on EEG signals, performance actions, and audio-visual feedback.

[0005] Preferably, the Neurocognitive Adaptation Module includes an EEG signal spatio-temporal encoder and a dual-channel adversarial neural network. The EEG signal spatio-temporal encoder is used to extract music perception feature vectors in the α, β, and γ bands. The first channel in the dual-channel adversarial neural network is used to process audio signals, and the second channel is used to process synchronous brain wave data. A mapping relationship between music stimuli and neural responses is established through adversarial training.

[0006] Preferably, the neurocognitive adaptation module includes a dynamic deviation calculation unit, which generates a cognitive deviation coefficient by comparing the difference between the measured neural feature vector and the predicted vector and combining the historical data standard deviation.

[0007] Preferably, the music DNA map construction module includes a quantum generative adversarial network Q-GAN, a three-dimensional skill evolution topology map, and a temporal convolutional attention prediction model. The virtual learner database generated by the quantum generative adversarial network is used as a comparison benchmark. The nodes in the three-dimensional skill evolution topology map represent music ability elements, and the edge weights are dynamically updated by a Bayesian inference engine. The temporal convolutional attention prediction model can output a probabilistic index of the skill development trajectory in the next 12 months.

[0008] Preferably, the cognitive load monitoring module includes an eye micro-motion capture unit, a multi-task temporal neural network, and a pre-judgment trigger unit. The eye micro-motion capture unit can calculate the pupil oscillation frequency and eyelid closure rate based on the Facial Action Coding System. The multi-task temporal neural network can synchronously process visual data and performance behavior data and output an attention breakdown probability value. In the pre-judgment trigger unit, when the attention breakdown probability exceeds 0.82, the teaching scenario is switched within 300 milliseconds.

[0009] Preferably, the anti-AI dependence regulation module includes a performance variability analysis unit, a dynamic attenuation controller, and an independent creation protection unit. The performance variability analysis unit can generate an innovation degree index by quantifying the music complexity of the improvised segment. The dynamic attenuation controller reduces the AI intervention intensity according to the innovation degree index according to an exponential function curve. In the independent creation protection unit, when a high innovation index is continuously detected, the output amplitude of the harmony assistant generator is attenuated to 32% ± 5% of the reference value.

[0010] Preferably, the multi-source data fusion unit includes an asynchronous signal alignment encoder, a cross-modal attention weight allocator, and a distributed joint learning framework. The asynchronous signal alignment encoder can solve the sampling rate differences of electroencephalogram signals, motion data, and audio signals. The cross-modal attention weight allocator is used to calculate the contribution of different data sources to the feature representation. The distributed joint learning framework realizes efficient feature sharing and privacy protection of multi-modal data.

[0011] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention constructs a dynamically evolving personalized teaching model through AI-driven multimodal data fusion and real-time analysis. The neurocognitive adaptation module combines electroencephalogram (EEG) feature analysis and physiological signal monitoring to accurately capture the cognitive preferences and ability bottlenecks of learners; the music DNA map, based on quantum-enhanced modeling technology, continuously updates and predicts the skill development trajectory. The AI teaching strategy can achieve millisecond-level dynamic adjustment according to the neural feedback and behavioral data of learners, significantly improving the skill mastery efficiency and knowledge retention rate, breaking through the limitations of static and single traditional teaching systems; 2. The present invention realizes an intelligent balance between assisted training and autonomous creation through an innovatively designed anti-AI dependence regulation algorithm. Based on the dynamic attenuation strategy of deep reinforcement learning, it can autonomously adjust the AI intervention intensity according to the real-time evaluated innovation potential index. This mechanism not only retains the accurate guidance advantage of artificial intelligence in basic skill training but also stimulates the autonomous creation ability of learners through a progressive exit strategy, effectively solving the technical bottleneck of suppressing creativity in existing intelligent teaching systems; 3. Relying on the federated learning framework and quantum encryption technology, the system of the present invention realizes the secure fusion and co-evolution of multi-source heterogeneous data. The cross-modal attention mechanism ensures the spatio-temporal alignment and feature complementarity of data such as EEG, motion, and audio. The privacy protection design under the distributed architecture enables the desensitization processing of biometric data locally. At the same time, the AI model is continuously optimized by absorbing global desensitized data, forming a self-evolution ability of "data closed-loop - algorithm iteration - system upgrade", providing technical guarantee for the long-term improvement of teaching effectiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a flowchart of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0013] To make the technical means, creative features, achieved purposes, and effects of the present invention easy to understand, the present invention will be further described below in conjunction with specific embodiments.

[0014] As Figure 1 shown, an AI-based music teaching system is configured as follows: Neurocognitive adaptation module: Real-time collect the brain waves and physiological data of learners through multimodal biosensors, and analyze music cognitive features using deep neural networks; During implementation, the neurocognitive adaptation module collects raw EEG signals through a 128-channel EEG headset. After wavelet denoising, the signals are input into a spatio-temporal convolutional network (ST-CNN). In the spatial dimension, graph convolution is used to extract the co-activation patterns in the parietal lobe region (musical imagination) and the temporal lobe region (auditory processing). In the temporal dimension, sliding window analysis is adopted to analyze the rhythm synchronization characteristics in the γ band (30 - 50 Hz). The output layer fuses the heart rate variability (HRV) data to generate a cognitive feature vector updated every 250 ms, providing real-time neural feedback for downstream DNA map construction.

[0015] Music DNA Map Construction Module: Generates a dynamic personal music ability gene map based on historical performance data and real-time learning behavior; The above module receives the feature vector from the neurocognitive adaptation module and the performance MIDI data, and expands the training set through a quantum generative adversarial network (Q-GAN). The specific process is as follows: Data Augmentation: Uses the quantum annealing algorithm to generate virtual performer data on the order of 10^6, covering the skill span from the C major scale to Chopin's études; Map Modeling: Constructs a three-dimensional graph network, where the nodes represent 128 skill indicators (such as left-hand span accuracy, grace note fluency), and the edge weights calculate the skill correlation strength through conditional mutual information; Dynamic Prediction: Adopts a temporal convolutional attention mechanism to deduce the probability cloud map of skill development in the next 12 months based on the current DNA state, providing a warning baseline for the cognitive load module.

[0016] Cognitive Load Monitoring Module: Uses computer vision and micro-expression recognition technologies to evaluate the attention state in real time; This module forms a closed-loop control with the aforementioned modules: Input Layer: Receives the θ wave energy of the EEG (an indicator of distracted attention), the recent skill growth rate of the DNA map, and the PERCLOS value of eye tracking (the proportion of pupil closure duration); Processing Layer: Uses a double-layer LSTM network to model multi-source time-series data. When the predicted probability of attention collapse > 0.82, a three-level response protocol is triggered: First-level Response (probability 0.82 - 0.85): Reduces the music score scrolling speed to 70% and injects α wave-induced audio; Second-level Response (0.85 - 0.90): Switches to the breakdown training mode, breaking down the music piece into 8-bar segments according to skill requirements; Third-level Response (> 0.90): Activates the holographic guidance interface, highlighting the key fingering areas through photon crystal projection.

[0017] Anti-AI Dependence Adjustment Module: Dynamically adjusts the AI assistance intensity according to the level of creative thinking development; The above module is deeply coupled with the DNA map and cognitive monitoring module: Initial stage: When the skill accuracy shown in the DNA map is <65%, the AI provides real-time fingering projection and harmony filling (intervention intensity K = 0.9); Advanced stage: If the innovation coefficient D_v > 0.7 in three consecutive practices (calculated by comparing the KL divergence between the improvised fragment and the Q-GAN generation mode), start the exponential decay strategy: reduce the volume of the rhythm auxiliary track by 3 dB every 5 minutes and change the harmony prompt from continuous display to strong beat point flashing; Expert stage: When the skill entropy value of the DNA map enters the top 10% range, only retain the error warning function (K = 0.2) and activate the master performance comparison mode.

[0018] Multi-source data fusion unit: Align cross-modal features and jointly model electroencephalogram signals, performance actions, and audiovisual feedback.

[0019] The multi-source data fusion unit is the core hub of the system: Hardware synchronization: Adopt the IEEE1588 precision clock protocol to align the timestamps of electroencephalogram (1000Hz), motion capture (120Hz), and audio (44.1kHz) devices, with an error <0.2ms; Feature alignment: Design a cross-modal Transformer architecture, use electroencephalogram features as Query, performance actions as Key, and audio signals as Value, and generate a joint embedding vector through multi-head attention (8 heads); Federated learning: While locally encrypting and storing the original biological data, upload the desensitized features to the cloud for model update to ensure the coordination of privacy and model iteration.

[0020] The neurocognitive adaptation module includes an electroencephalogram signal spatio-temporal encoder and a dual-channel adversarial neural network. The electroencephalogram signal spatio-temporal encoder is used to extract music perception feature vectors in the α, β, and γ bands. The first channel in the dual-channel adversarial neural network is used to process audio signals, and the second channel is used to process synchronous brain wave data. A mapping relationship between music stimuli and neural responses is established through adversarial training.

[0021] The neurocognitive adaptation module includes a dynamic deviation calculation unit. The dynamic deviation calculation unit generates a cognitive deviation coefficient by comparing the difference between the measured neural feature vector and the predicted vector and combining the standard deviation of historical data.

[0022] Furthermore, the above electroencephalogram (EEG) signal spatio-temporal encoder adopts a dual-channel processing architecture: the spatial path uses a graph attention network (GAT) to analyze the fronto-parietal functional connectivity strength, and the temporal path applies dilated causal convolution (dilation = 4) to extract a neural response delay of up to 1.5 seconds. After the outputs of both are fused by cross-attention, a 64-dimensional spatio-temporal feature vector is generated as the input benchmark for the dual-channel adversarial network.

[0023] The adversarial training process realizes the neural-music closed-loop optimization: the generator G receives the current teaching strategy parameters and outputs the predicted EEG response pattern; the discriminator D compares the real EEG data with the generated data and optimizes the mapping relationship through the Wasserstein distance loss function. When the accuracy of D exceeds 82%, the cognitive bias coefficient recalibration process is triggered.

[0024] The cosine similarity between the measured feature vector and the predicted vector is calculated in real time, and combined with the standard deviation of the historical data within a sliding time window (the most recent 10 minutes), a dynamic cognitive bias coefficient C_d is generated. When C_d > 0.75, an emergency re-evaluation request is sent to the DNA map module, triggering a room-second-level adjustment of the teaching strategy.

[0025] The music DNA map construction module includes a quantum generative adversarial network Q-GAN, a three-dimensional skill evolution topology graph, and a temporal convolutional attention prediction model. The virtual learner database generated by the quantum generative adversarial network is used as a comparison benchmark. The nodes in the three-dimensional skill evolution topology graph represent music ability elements, and the edge weights are dynamically updated by a Bayesian inference engine. The temporal convolutional attention prediction model can output a probabilistic index of the skill development trajectory in the next 12 months.

[0026] Furthermore, Q-GAN runs in a quantum computing environment, uses 8 qubits to encode skill parameters (such as dynamics control, rhythm stability), and generates skill combination states through quantum entanglement (such as the entangled state of high vibrato precision and medium improvisation ability). During classical-quantum hybrid training, the quantum processor optimizes the generator parameters, and the GPU cluster processes the gradient backpropagation of the discriminator.

[0027] Furthermore, for the above topology graph dynamic update mechanism, every 30 minutes of new performance data is added, and hidden skill paths are mined through a random walk algorithm (for example, it is found that the improvement of legato ability can drive the expressiveness of ornaments). The edge weights between skill nodes are updated using Bayesian inference, reflecting the individual-differentiated skill development patterns of learners.

[0028] The prediction engine adopts a multi-scale architecture: Short-term prediction (within 1 month): Use dilated convolution (dilation = 2) to capture recent skill fluctuations; Medium-term prediction (3 - 6 months): Identify periodic growth patterns through an attention mechanism; Long-term prediction (12 months): Incorporate the master growth trajectory data through transfer learning and output the probability distribution map of skill breakthroughs.

[0029] The cognitive load monitoring module includes an eye micro-motion capture unit, a multi-task temporal neural network, and a pre-judgment trigger unit. The eye micro-motion capture unit can calculate the pupil oscillation frequency and eyelid closure rate based on the Facial Action Coding System. The multi-task temporal neural network can synchronously process visual data and performance behavior data and output the probability value of attention breakdown. In the pre-judgment trigger unit, when the probability of attention breakdown exceeds 0.82, the teaching scenario is switched within 300 milliseconds.

[0030] The eye micro-motion capture unit is linked with the neurocognitive module. When the EEG detects an increase in theta wave energy, high-speed eye movement tracking (240fps) is initiated. The attention dispersion features (such as frequent saccades outside the music score area) are identified through a transfer learning model. After the data is analyzed by LSTM time series, the skill decomposition strategy in the DNA map is triggered, and complex passages are automatically disassembled into digestible segments.

[0031] The task temporal neural network adopts a shared-branch architecture. The underlying Bi-LSTM extracts common temporal features, and the upper layer branches out the cognitive load score (normalized to 0-1) and the intervention strategy selection (simplification / decomposition / immersion mode). When the score > 0.82, the intervention plan with the highest historical success rate in the DNA map is preferentially called.

[0032] The three-level response mechanism of the pre-judgment trigger unit collaborates with the anti-AI dependence module: The high-level response (P > 0.9) not only switches the training mode but also reduces the AI intervention intensity by 15%, forcing the learner to call on their own skill reserves. The response delay is controlled within 280ms through FPGA hardware acceleration.

[0033] The anti-AI dependence adjustment module includes a performance variability analysis unit, a dynamic attenuation controller, and an original creation protection unit. The performance variability analysis unit can generate an innovation degree index by quantifying the music complexity of improvised segments. The dynamic attenuation controller reduces the AI intervention intensity according to the innovation degree index according to an exponential function curve. In the original creation protection unit, when high innovation indicators are continuously detected, the output amplitude of the harmony assistant generator is attenuated to 32% ± 5% of the reference value.

[0034] The performance variability analysis unit compares the user's improvised segments with a pattern library of 10^4 magnitudes generated by Q-GAN, calculates the melody contour difference degree through the DTW algorithm, and then generates a comprehensive innovation coefficient D_v in combination with the harmony progression entropy value. When D_v > 0.75 for 3 consecutive times, a creativity mark is written into the DNA map to unlock advanced training content.

[0035] The dynamic attenuation controller adopts a non-linear attenuation function:

[0036] Among them, Dynamically adjust according to the skill stage in the DNA map , and the attenuation rate is inversely correlated with the cognitive load score to ensure the balance of learning pressure.

[0037] The independent creation protection unit adopts the coordination of progressive attenuation and multi-modal feedback: when the harmony assistance volume drops to 32%, the sound field visualization feedback in the holographic interface is enhanced synchronously, and the acoustic wave interference pattern is presented through the photonic crystal projection to assist the learner in independently constructing the harmony.

[0038] The multi-source data fusion unit includes an asynchronous signal alignment encoder, a cross-modal attention weight allocator, and a distributed joint learning framework. The asynchronous signal alignment encoder can solve the sampling rate differences of electroencephalogram signals, motion data, and audio signals. The cross-modal attention weight allocator is used to calculate the contribution degree of different data sources to the feature representation. The distributed joint learning framework realizes the efficient feature sharing and privacy protection of multi-modal data.

[0039] The asynchronous signal alignment encoder adopts a two-level alignment strategy: at the hardware level, us-level synchronization is achieved through the PTP protocol, and at the software level, dynamic time warping (DTW) is applied to eliminate the residual deviation. The aligned data stream is input into the cross-modal Transformer to generate a 256-dimensional joint embedding vector.

[0040] The weight calculation in the cross-modal attention weight allocator introduces the metadata of the DNA map: when the map shows weak rhythm ability, the weight of the motion capture data is increased by 40%; during the creativity development stage, the attention to audio features is strengthened, and the weight allocation is dynamically updated every 5 seconds.

[0041] The distributed joint learning framework designs a three-layer federated learning architecture: the device side trains the modality-specific encoder, the edge node aggregates the cross-modal features, and the cloud updates the global model. The biometric data is encrypted homomorphically throughout the process, and only the gradient tensor is transmitted during model updates.

[0042] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence-based music teaching system, characterized in that: the artificial intelligence-based music teaching system is configured to: Neurocognitive adaptation module: Real-time collect learners' brain waves and physiological data through multi-modal biosensors, and analyze music cognitive characteristics using deep neural networks; Music DNA map construction module: Generate a dynamic personal music ability gene map based on historical performance data and real-time learning behaviors; Cognitive load monitoring module: Use computer vision and micro-expression recognition technologies to evaluate the attention state in real time; Anti-AI dependence adjustment module: Dynamically adjust the AI assistance intensity according to the level of creative thinking development; Multi-source data fusion unit: Perform cross-modal feature alignment and joint modeling on electroencephalogram signals, performance actions, and audio-visual feedback.

2. The music teaching system based on artificial intelligence according to claim 1, wherein: The neurocognitive adaptation module includes an electroencephalogram signal spatio-temporal encoder and a dual-channel adversarial neural network. The electroencephalogram signal spatio-temporal encoder is used to extract music perception feature vectors in the α, β, and γ bands. The first channel in the dual-channel adversarial neural network is used to process audio signals, and the second channel is used to process synchronous brain wave data. A mapping relationship between music stimuli and neural responses is established through adversarial training.

3. The music teaching system based on artificial intelligence according to claim 1, characterized in that: The neurocognitive adaptation module includes a dynamic deviation calculation unit. The dynamic deviation calculation unit generates a cognitive deviation coefficient by comparing the difference between the measured neural feature vector and the predicted vector and combining the standard deviation of historical data.

4. An AI-based music teaching system according to claim 1, characterized in that: The music DNA map construction module includes a quantum generative adversarial network Q-GAN, a three-dimensional skill evolution topology map, and a temporal convolutional attention prediction model. The virtual learner database generated by the quantum generative adversarial network is used as a comparison benchmark. The nodes in the three-dimensional skill evolution topology map represent music ability elements, and the edge weights are dynamically updated by a Bayesian inference engine. The temporal convolutional attention prediction model can output a probabilistic index of the skill development trajectory in the next 12 months.

5. The music teaching system based on artificial intelligence according to claim 1, wherein: The cognitive load monitoring module includes an eye movement capture unit, a multi-task temporal neural network, and a pre-judgment trigger unit. The eye movement capture unit can calculate the pupil oscillation frequency and eyelid closure rate based on the facial action coding system. The multi-task temporal neural network can synchronously process visual data and performance behavior data and output an attention breakdown probability value. In the pre-judgment trigger unit, when the attention breakdown probability exceeds 0.82, the teaching scenario is switched within 300 milliseconds.

6. The music teaching system based on artificial intelligence according to claim 1, wherein: The anti-AI dependence adjustment module includes a performance variability analysis unit, a dynamic attenuation controller, and an independent creation protection unit. The performance variability analysis unit can generate an innovation degree index by quantifying the music complexity of improvised segments. The dynamic attenuation controller reduces the AI intervention intensity according to the innovation degree index according to an exponential function curve. In the independent creation protection unit, when a high innovation index is continuously detected, the output amplitude of the harmony assistant generator is attenuated to 32% ± 5% of the reference value.

7. An artificial intelligence-based music teaching system according to claim 1, characterized in that: The multi-source data fusion unit includes an asynchronous signal alignment encoder, a cross-modal attention weight allocator, and a distributed joint learning framework. The asynchronous signal alignment encoder can solve the sampling rate differences of electroencephalogram signals, motion data, and audio signals. The cross-modal attention weight allocator is used to calculate the contribution degrees of different data sources to feature representations. The distributed joint learning framework realizes efficient feature sharing and privacy protection of multi-modal data.

Images