Adaptive audio adjusting method and system based on earphone noise reduction

Abstract

The invention belongs to the field of earphone adjustment, and particularly relates to a self-adaptive audio adjustment method and system based on earphone noise reduction, and the method comprises the steps: S1, cooperatively collecting multi-source noise according to hardware of a feed-forward microphone and a feedback microphone, analyzing the multi-source noise to obtain noise features, and carrying out the dynamic threshold calculation according to the noise features to generate a noise reduction target curve; s2, acquiring user information through multi-dimensional perception, and fusing the heart rate, skin conductance and accelerometer data of the user; real-time tracking and accurate suppression of a noise spectrum are realized through multi-source noise analysis and dynamic threshold calculation; the physiological signals and the motion data are combined, the user state is recognized, and a noise reduction strategy is driven to adapt; aNC filter parameters and audio equalization parameters are dynamically configured according to the noise characteristics and the user state, and the noise reduction intensity and the sound quality fidelity are balanced; and through signal-to-noise ratio quantitative evaluation and a user feedback driving reinforcement learning model, continuous iterative optimization of a noise reduction strategy is realized.

Description

Technical Field

[0001] The present invention relates to the field of headphone adjustment, and specifically relates to an adaptive audio adjustment method and system based on headphone noise reduction. Background Art

[0002] Noise-canceling headphones refer to headphones that use a certain method to reduce noise. There are two types of noise-canceling headphones: active noise-canceling headphones and passive noise-canceling headphones.

[0003] The active noise-canceling function is to generate an anti-phase sound wave equal to the external noise through the noise-canceling system to neutralize the noise, thereby achieving the noise-canceling effect. Passive noise-canceling headphones mainly form a closed space by surrounding the ears or use sound insulation materials such as silicone earplugs to block external noise.

[0004] The existing headphone noise reduction technologies are mostly based on static noise reduction strategies with fixed thresholds or preset scenarios, and have the following defects:

[0005] It is unable to dynamically track the frequency domain characteristics and energy fluctuations of environmental noise, resulting in noise reduction lag or over-suppression of useful signals; it ignores the impact of the user's physiological state (such as exercise, stress) on the noise reduction demand, resulting in uncomfortable listening or safety hazards; the ANC filter and audio equalization parameters are fixed, making it difficult to balance noise suppression and sound quality optimization in complex scenarios; there is a lack of a dynamic adjustment mechanism based on noise reduction effects and user feedback, and performance degradation is likely to occur after long-term use. Summary of the Invention

[0006] (I) Invention Objectives

[0007] To solve the technical problems in the background art, the present invention proposes an adaptive audio adjustment method based on headphone noise reduction, which has the characteristics of dynamic noise perception, user state fusion, multi-modal control, and closed-loop optimization.

[0008] (II) Technical Solutions

[0009] To solve the above technical problems, the present invention provides an adaptive audio adjustment method based on headphone noise reduction, including the following steps:

[0010] Step S1: The front-feed microphone and the feedback microphone cooperate in hardware to collect multi-source noise, analyze the multi-source noise to obtain noise characteristics, and calculate a dynamic threshold based on the noise characteristics to generate a noise reduction target curve;

[0011] Step S2: Obtain user information through multi-dimensional perception, fuse the user's heart rate, skin conductance, and accelerometer data, and identify the user's state through a pre-trained classification model;

[0012] Step S3: Dynamically configure the noise reduction strategy parameters in the corresponding scenario based on the noise characteristics and user state labels, and perform noise reduction during the audio playback process.

[0013] Step S4: evaluate the noise reduction performance by calculating the signal-to-noise ratio improvement before and after noise reduction in real time, and optimize the strategy parameters by driving the reinforcement learning model based on the user's manual adjustment records.

[0014] Preferably, analyzing the multi-source noise to obtain noise characteristics includes:

[0015] Multi-source noise includes ambient broadband noise picked up by the feedforward microphone and residual ear canal noise captured by the feedback microphone;

[0016] The noise is decomposed into noise energy, frequency domain signal and frequency domain characteristics through fast Fourier transform, and low-frequency, medium-frequency and high-frequency noise types are distinguished.

[0017] Preferably, the generating of the noise reduction target curve by performing dynamic threshold calculation according to the noise characteristics includes: the dynamic threshold calculation is to provide a real-time optimized noise elimination benchmark, and the model formula of the dynamic threshold calculation is an exponentially weighted moving average:

[0018]

[0019] According to the current noise energy Dynamically adjust the noise reduction target curve based on the historical threshold. is the smoothing coefficient (0-1),

[0020] Current noise energy >Dynamic Threshold When the ambient noise exceeds the dynamic reference, the noise reduction process needs to be started. At this time, the noise reduction target curve will dynamically adjust the noise reduction intensity according to the excess amplitude.

[0021] Current noise energy ≤ Dynamic Threshold When the noise is judged to have returned to an acceptable range, the noise reduction can be gradually weakened or turned off to avoid excessive suppression of useful signals (such as human voices).

[0022] Preferably, identifying the user status by using a pre-trained classification model includes:

[0023] Calculates physiological stress index based on user's heart rate and skin conductance;

[0024] The exercise intensity index is determined by the three-axis frequency domain entropy value of the accelerometer data;

[0025] The physiological stress index and exercise intensity index are input into the classification model, and the label of the user being in a relaxed / tense / exercise state is output.

[0026] Preferably, the filtering parameters of the ANC filter need to be adjusted in multiple dimensions according to the noise characteristics and user status to ensure accurate noise suppression while retaining key audio information;

[0027] It mainly includes: frequency band targeted noise reduction, dynamic adjustment of noise reduction intensity, and adaptation to user status.

[0028] During the audio playback process, the noise reduction of the headphones adjusts the audio output based on the user status and the noise environment, and optimizes the frequency response curve dynamically based on the audio equalization parameters to optimize the listening experience.

[0029] Preferably, step S3 further includes:

[0030] Continuously monitor the noise reduction effect and user feedback to form a closed-loop optimization of headphone noise reduction and adaptive audio adjustment:

[0031] As an example of biometric signal feedback, if the user's heart rate continues to increase (PSI increases), automatically enhance the noise reduction or switch to a soothing sound effect (such as light music);

[0032] When the accelerometer detects stillness (MII drops suddenly), gradually restore the default noise reduction mode.

[0033] Preferably, in step S4, evaluating the noise reduction efficacy by calculating the SNR improvement value before and after noise reduction in real time includes:

[0034] Measure the noise suppression ability of the noise reduction algorithm through the objective index SNR, avoid the deviation of subjective evaluation, and quantify the noise reduction effect;

[0035] Identify the noise reduction shortboards in different frequency bands, provide a basis for parameter adjustment, and conduct frequency band targeted analysis;

[0036] Detect the noise reduction failure scenario and trigger emergency optimization.

[0037] Preferably, the driving reinforcement learning model adopts the Deep Deterministic Policy Gradient (DDPG) model, which is suitable for continuous action spaces (such as noise reduction depth, frequency band gain);

[0038] Input the current state (noise characteristics, user status, SNR improvement value), and output actions (ANC parameters, audio equalization parameter settings);

[0039] Its calculation formula is: R = w1 ⋅ SNR improvement value + w2 ⋅ user satisfaction - w3 ⋅ power consumption;

[0040] User satisfaction: Quantified according to the user operation direction (such as enhancing noise reduction → +1, weakening → -1);

[0041] Power consumption penalty: Linearly related to the processor load and noise reduction intensity (such as when the noise reduction is -30dB, the power consumption coefficient is 0.8);

[0042] The training process is:

[0043] Offline pre-training: Initialize the model using historical datasets (synthetic noise scenarios + simulated user operations);

[0044] Online fine-tuning: When new data arrives, perform mini-batch gradient descent through the edge computing chip;

[0045] Policy deployment: Push the optimized parameters to the ANC filter module and the audio equalizer.

[0046] The present invention also provides an adaptive audio adjustment system based on headphone noise reduction, including:

[0047] A noise feature module, a front-feed microphone, and a feedback microphone cooperate to collect multi-source noise hardware, analyze the multi-source noise to obtain noise features, and calculate a dynamic threshold according to the noise features to generate a noise reduction target curve;

[0048] A multi-dimensional perception module, multi-dimensionally perceives user information, fuses user heart rate, skin conductance, and accelerometer data, and identifies the user's state through a pre-trained classification model;

[0049] A dynamic noise reduction module, based on noise features and user state labels, dynamically configures noise reduction strategy parameters in the corresponding scenario for noise reduction during the audio playback process;

[0050] A closed-loop optimization module, evaluates the noise reduction efficiency by calculating the real-time improvement value of the signal-to-noise ratio before and after noise reduction, and combines the user's manual adjustment record to drive the reinforcement learning model for policy parameter optimization.

[0051] The above technical solution of the present invention has the following beneficial technical effects:

[0052] 1. Through multi-source noise analysis and dynamic threshold calculation, real-time tracking and precise suppression of the noise spectrum are achieved;

[0053] 2. Combining physiological signals and motion data to identify the user's state and drive the adaptation of the noise reduction strategy;

[0054] 3. Dynamically configure ANC filter parameters and audio equalization parameters according to noise features and user states to balance noise reduction intensity and sound quality fidelity;

[0055] 4. Through signal-to-noise ratio quantization evaluation and user feedback-driven reinforcement learning model, continuous iterative optimization of the noise reduction strategy is achieved. Description of the Drawings

[0056] Figure 1 It is a schematic diagram of the method of the present invention;

[0057] Figure 2 It is a schematic diagram of the noise feature module of the present invention;

[0058] Figure 3Schematic diagram of the multi-dimensional perception module of the present invention. Detailed implementation manners

[0059] To make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the specific implementation manners and with reference to the accompanying drawings. It should be understood that these descriptions are exemplary and are not intended to limit the scope of the present invention. In addition, in the following descriptions, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present invention.

[0060] As Figures 1-3 shown, an adaptive audio adjustment method based on headphone noise reduction proposed by the present invention includes the following steps:

[0061] Step S1: The front-feed microphone and the feedback microphone cooperate in hardware to collect multi-source noise, analyze the noise spectrum composed of the multi-source noise to obtain noise characteristics, and generate a noise reduction target curve according to the noise characteristics through dynamic threshold calculation;

[0062] Among them, analyzing the multi-source noise to obtain noise characteristics includes:

[0063] The multi-source noise includes the environmental broadband noise collected by the front-feed microphone and the ear canal residual noise captured by the feedback microphone;

[0064] The noise is decomposed into noise energy , frequency domain signals and frequency domain characteristics through fast Fourier transform for noise characteristic extraction to distinguish between low-frequency and medium-high frequency noise types.

[0065] The generating of the noise reduction target curve according to the noise characteristics through dynamic threshold calculation includes: The dynamic threshold calculation is to provide a real-time optimized noise cancellation benchmark, and the model formula of the dynamic threshold calculation is the exponentially weighted moving average:

[0066]

[0067] According to the current noise energy and the historical threshold, the noise reduction target curve is dynamically adjusted, is the smoothing coefficient (0 - 1), The larger the value (close to 1), the more sensitive the threshold is to the noise change (quickly tracking the current noise);

[0068] The smaller the value (close to 0), the more the threshold depends on the historical value, with strong smoothing effect but possible delayed response;

[0069] The condition for starting noise reduction (threshold upper limit);

[0070] The current noise energy > the dynamic threshold When it is determined that the ambient noise exceeds the dynamic baseline, noise reduction processing needs to be started. At this time, the noise reduction target curve will dynamically adjust the noise reduction intensity according to the exceeded amplitude;

[0071] The condition for stopping noise reduction (lower threshold);

[0072] Current noise energy ≤Dynamic threshold When it is determined that the noise has returned to the acceptable range, the noise reduction can be gradually weakened or turned off to avoid over-suppressing useful signals (such as human voices).

[0073] The threshold continuously adapts to noise changes through exponential weighted average, which not only avoids mis-triggering by instantaneous fluctuations but also can track the long-term noise trend. The generation of the noise reduction target curve is related to and The difference (for example: the greater the difference, the stronger the noise reduction intensity).

[0074] As an example, if is set, the current threshold: Suddenly there is noise :

[0075] New threshold , since 70 > 54, the system starts noise reduction and calculates the noise reduction intensity according to the difference (16). This design balances real-time performance and stability and is applicable to headphone noise reduction, voice enhancement, and audio processing.

[0076] Step S2: Obtain user information through multi-dimensional perception, fuse user heart rate, skin conductance, and accelerometer data, and identify the user's state through a pre-trained classification model;

[0077] Among them, identifying the user's state through a pre-trained classification model includes:

[0078] Calculate the physiological stress index based on the user's heart rate and skin conductance;

[0079] Judge the exercise intensity index through the three-axis frequency domain entropy value of the accelerometer data;

[0080] Input the physiological stress index and the exercise intensity index into the classification model, and output the label of the user's relaxed / tense / exercising state.

[0081] As an example: The rule of the classification model is the hard threshold method, and the input features are:

[0082] Physiological stress index ( ): Range [0, 1], the larger the value, the higher the stress;

[0083] Exercise intensity index ( ): Range [0, 1], the larger the value, the more intense the exercise;

[0084] Output features:

[0085] ;

[0086] ;

[0087] ;

[0088] Wherein: = Exercise intensity threshold (default 0.5), = Pressure threshold (default 0.7);

[0089] As a classification example: First, determine whether the user is exercising. If MII ≥ 0.5 (such as running, brisk walking), directly classify it as "exercise" and ignore the pressure value;

[0090] If < 0.5 (such as sitting still, standing), proceed to the next judgment;

[0091] Then, determine whether the user is tense. If ≥ 0.7 (such as a rapid heartbeat, profuse sweating), classify it as: "tense";

[0092] If < 0.7, classify it as "relaxed".

[0093] Step S3: Dynamically configure the ANC filter parameters and audio equalization parameters based on the noise characteristics and user status labels, and perform headphone noise reduction during the audio playback process;

[0094] It should be noted that: The filtering parameters of the ANC filter need to be adjusted in multiple dimensions according to the noise characteristics and user status to ensure accurate noise suppression while retaining key audio information.

[0095] The parameter adjustment of the ANC filter is based on the noise characteristics and user status, mainly including: frequency band targeted noise reduction, dynamic adjustment of noise reduction intensity, and user status adaptation;

[0096] As an example of frequency band targeted noise reduction:

[0097] According to the noise spectrum analysis results (low-frequency / mid-high frequency energy distribution), adjust the noise reduction depth of each frequency band in the filter bank;

[0098] Low-frequency noise (such as engine noise, air conditioner noise): Enable a high-order IIR filter to enhance low-frequency attenuation (such as below -30dB);

[0099] Mid-high frequency noise (such as human voice, keyboard noise): Use an adaptive FIR filter and combine notch design to avoid over-suppressing the voice frequency band (such as 1kHz~4kHz).

[0100] As an example of dynamic adjustment of noise reduction intensity:

[0101] According to the difference between the noise energy and the dynamic threshold ( ), the noise reduction intensity is adjusted in real time:

[0102] : Increase the noise reduction depth proportionally (for example, when Δ increases by 10, the noise reduction gain increases by 2 dB);

[0103] : Gradually reduce the noise reduction intensity and switch to the transparent mode (retaining ambient sound).

[0104] As an example of a noise reduction strategy driven by user status:

[0105] 1. Exercise state: Prioritize suppressing wind noise: Detect the movement speed through an accelerometer. If the speed > 5 km / h, enable high-frequency wind noise suppression (> 2 kHz, attenuation -15 dB);

[0106] Retain the ambient sound channel: Open the transparent mode in the mid-frequency range (500 Hz - 1 kHz) to ensure that the user can hear safety signals such as traffic prompts.

[0107] 2. Tense state: Full-band deep noise reduction: Enhance the ANC intensity (for example, -40 dB in the low frequency, -20 dB in the mid-high frequency), and superimpose white noise masking (such as pink noise) to relieve psychological pressure; Sudden noise suppression: Quickly respond to sudden noises (such as the sound of closing a door) through a transient detection algorithm (such as short-time energy analysis), and instantaneously increase the noise reduction depth.

[0108] 3. Relaxed state: Selective noise reduction: Only suppress continuous low-frequency noises (such as air conditioner sounds), and retain the natural sound field (such as ambient human voices, bird sounds). Noise reduction delay off: If , for 10 seconds, gradually reduce the noise reduction intensity to the sleep mode to save power.

[0109] In one embodiment, the headphone noise reduction during audio playback is a frequency response curve dynamically optimized based on audio equalization parameters that adjusts the audio output according to the user status and the noise environment, which can optimize the listening experience;

[0110] As an example of audio equalization parameter adaptation in the exercise state

[0111] Exercise state: Boost the mid-low frequency (80 Hz - 300 Hz) to enhance the rhythm, and attenuate the high frequency (> 8 kHz) to reduce harshness; Dynamic compression prevents drastic volume fluctuations.

[0112] Tense state: Weaken the low frequency (< 200 Hz) to reduce the sense of oppression, enhance the mid frequency (2 kHz - 5 kHz) to improve the clarity of human voices; Enable a limiter to avoid sudden high-volume stimuli.

[0113] Relaxed state: Preserve the original dynamic range of the audio; Optionally add natural sound field expansion (such as virtual surround sound);

[0114] The dynamic audio equalization parameter adjustment optimizes the audio playback effect in real time through the dual inputs of noise energy distribution and user status tags, achieving the following goals:

[0115] Noise masking: Enhance the intelligibility of key frequency bands in a noisy environment;

[0116] Psychological adaptation: Adjust the timbre tendency according to the user's stress / exercise state;

[0117] Hearing protection: Avoid hearing fatigue caused by energy accumulation in extreme frequency bands.

[0118] As an example supplement for the multi-modal feedback closed-loop control of adaptive audio adjustment, continuously monitor the noise reduction effect and user feedback to form a closed-loop optimization of headphone noise reduction and adaptive audio adjustment: Analyze the residual noise after noise reduction in real time through the feedback microphone, perform ear canal residual noise detection, and calculate its residual energy If it exceeds the expected value (such as > 10% of), then iteratively adjust the ANC parameters. If continuously below > 5% of, reduce the ANC power consumption to the energy-saving mode.

[0119] As an example of biosignal feedback, if the user's heart rate continues to rise ( increase), automatically enhance the noise reduction or switch to soothing sound effects (such as light music); when the accelerometer detects rest ( sudden drop), gradually restore the default noise reduction mode.

[0120] As an example of practical application:

[0121] Taking subway waiting and riding as an example (high-frequency noise + tense state):

[0122] Noise characteristics: = 75 (announcement sound, crowd sound), = 60.

[0123] User status: = 0.85 (tense);

[0124] Execute actions:

[0125] ANC full-band deep noise reduction (low frequency -35dB, mid-high frequency -25dB), superimposed with white noise masking;

[0126] The audio equalization parameters improve the clarity of human voices and suppress environmental noise.

[0127] Taking night study as an example (low-frequency noise + relaxed state):

[0128] Noise characteristics: = 40 (low-frequency sound of air conditioner), = 45 (noise reduction not triggered);

[0129] User state: = 0.3 (relaxed);

[0130] Execution action:

[0131] ANC only enables low-frequency lightweight noise reduction;

[0132] The audio equalization parameters adopt the high-fidelity mode, and virtual surround sound is enabled to enhance the immersion.

[0133] Through the deep integration of noise characteristic analysis and user state, the leap from "single noise reduction" to the collaborative optimization of "scene, user, and environment" has been achieved, significantly improving the usability of the headphones in complex environments.

[0134] Step S4: Evaluate the noise reduction efficiency by calculating the SNR improvement value before and after noise reduction in real time, and optimize the policy parameters by combining the user's manual adjustment records to drive the reinforcement learning model.

[0135] Among them, evaluating the noise reduction efficiency by calculating the SNR improvement value before and after noise reduction in real time includes:

[0136] Measure the noise suppression ability of the noise reduction algorithm through objective indicators (SNR), avoid the deviation of subjective evaluation, and quantify the noise reduction effect;

[0137] Identify the noise reduction shortboards in different frequency bands (low frequency, medium and high frequency), provide a basis for parameter adjustment, and conduct frequency band targeted analysis;

[0138] Detect noise reduction failure scenarios (such as sudden noise not suppressed), trigger emergency optimization, detect noise reduction failure scenarios (such as sudden noise not suppressed), trigger emergency optimization;

[0139] The SNR calculation formula is:

[0140]

[0142] Among them, the signal power comes from the original audio (such as music playback content);

[0143] The noise power comes from the mixed signal collected by the feedforward / feedback microphone.

[0144] Calculate the moving average with a 5-second window to smooth the instantaneous fluctuations.

[0145] Frequency band division:

[0146] Low frequency (<1kHz): Evaluate the suppression effect of engine noise and air conditioner noise;

[0147] Medium and high frequency (1 - 4kHz): Detect residual noises such as human voices and keyboard sounds.

[0148] The purpose of the user's manual adjustment record is: to learn personalized needs through the user's active operations (such as enhancing noise reduction and switching audio equalization parameters), record the user's operation instructions through physical buttons, touch gestures or APPs (such as "noise reduction + / -", "bass enhancement"), and associate the context information during the operation: noise characteristics ( and frequency band distribution), user status tags (exercise / nervous / relaxed).

[0149] In one embodiment, the deep deterministic policy gradient (DDPG) model is used to drive the reinforcement learning model for policy parameter optimization, which is applicable to continuous action spaces (such as noise reduction depth and frequency band gain);

[0150] Input the current state (noise characteristics, user status, SNR improvement value), and output actions (ANC parameters and audio equalization parameter settings);

[0151] Its calculation formula is: ;

[0152] User satisfaction: Quantified according to the user's operation direction (such as enhancing noise reduction → +1, weakening → -1);

[0153] Power consumption penalty: Linearly related to the processor load and noise reduction intensity (such as when the noise reduction is -30dB, the power consumption coefficient is 0.8);

[0154] The training process is as follows:

[0155] Offline pre-training: Initialize the model using historical data sets (synthetic noise scenarios + simulated user operations);

[0156] Online fine-tuning: When new data arrives, perform mini-batch gradient descent through the edge computing chip;

[0157] Policy deployment: Push the optimized parameters to the ANC filter module and the audio equalizer.

[0158] Step S4 upgrades the "static rules" of traditional noise reduction to "dynamic learning" through the combination of quantitative evaluation and user feedback, realizes intelligent noise reduction with "environment and user" collaboration, and the closed-loop optimization process can be continuously iteratively improved to form a closed loop of "monitoring → evaluation → optimization → execution" to adapt to environmental and user changes.

[0159] An adaptive audio adjustment system based on headphone noise reduction, comprising:

[0160] A noise feature module, a feedforward microphone and a feedback microphone cooperate to collect multi-source noise, analyze the multi-source noise to obtain noise features, and calculate a dynamic threshold according to the noise features to generate a noise reduction target curve;

[0161] A multi-dimensional perception module, which obtains user information through multi-dimensional perception, fuses user heart rate, skin conductance and accelerometer data, and identifies the user's state through a pre-trained classification model;

[0162] A dynamic noise reduction module, which dynamically configures noise reduction strategy parameters in the corresponding scenario based on noise features and user state labels, and performs noise reduction during the audio playback process;

[0163] A closed-loop optimization module, which evaluates the noise reduction efficiency by calculating the improvement value of the signal-to-noise ratio before and after noise reduction in real time, and optimizes the strategy parameters by combining the user's manual adjustment records to drive the reinforcement learning model.

[0164] It should be understood that the above specific embodiments of the present invention are only used for exemplary illustration or explanation of the principles of the present invention, and do not constitute a limitation to the present invention. Therefore, any modifications, equivalent replacements, improvements, etc. made without departing from the spirit and scope of the present invention shall be included within the protection scope of the present invention. In addition, the appended claims of the present invention are intended to cover all changes and modification examples that fall within the scope and boundaries of the appended claims, or equivalent forms of such scope and boundaries.

Claims

1. An adaptive audio adjustment method based on headphone noise reduction, characterized in that, It includes the following steps: Step S1: Co-collect multi-source noise according to the hardware of the feedforward microphone and the feedback microphone, analyze the multi-source noise to obtain noise characteristics, and calculate a dynamic threshold based on the noise characteristics to generate a noise reduction target curve; Step S2: Obtain user information through multi-dimensional perception, fuse user heart rate, skin conductance, and accelerometer data, and identify the user's state through a pre-trained classification model; Step S3: Dynamically configure the noise reduction strategy parameters corresponding to the scenario based on the noise characteristics and user state labels to perform noise reduction during the audio playback process; Step S4: Evaluate the noise reduction efficiency by calculating the SNR improvement value before and after noise reduction in real time, and optimize the strategy parameters by combining the user's manual adjustment records to drive the reinforcement learning model.

2. The adaptive audio adjustment method based on headphone noise reduction according to claim 1, wherein, The analysis of the multi-source noise to obtain noise characteristics includes: The multi-source noise includes the environmental broadband noise collected by the feedforward microphone and the ear canal residual noise captured by the feedback microphone; Decompose the noise into noise energy, frequency domain signal, and frequency domain characteristics through fast Fourier transform to distinguish between low-frequency and medium-high frequency noise types.

3. An adaptive audio adjustment method based on headphone noise reduction according to claim 1, characterized in that The calculation of the dynamic threshold based on the noise characteristics to generate a noise reduction target curve includes: The dynamic threshold calculation is to provide a real-time optimized noise cancellation benchmark, and the model formula for the dynamic threshold calculation is the exponentially weighted moving average: According to the current noise energy dynamically adjust the noise reduction target curve based on the historical threshold, where is the smoothing coefficient (0 - 1), Current noise energy > Dynamic threshold When it is determined that the ambient noise exceeds the dynamic reference and noise reduction processing needs to be started, the noise reduction target curve will dynamically adjust the noise reduction intensity according to the exceeded amplitude; Current noise energy ≤Dynamic threshold When this is the case, it is determined that the noise has returned to the acceptable range, and noise reduction can be gradually weakened or turned off to avoid over-suppressing useful signals.

4. An adaptive audio adjustment method based on headphone noise reduction according to claim 1, characterized in that The identification of the user's state through a pre-trained classification model includes: Calculate the physiological stress index based on the user's heart rate and skin conductance; Judge the exercise intensity index through the three-axis frequency domain entropy value of the accelerometer data; Input the physiological stress index and the exercise intensity index into the classification model to output the user's relaxation / tension / exercise state label.

5. The adaptive audio adjustment method based on headphone noise reduction according to claim 1, wherein, The filtering parameters of the ANC filter need to be adjusted in multiple dimensions according to the noise characteristics and the user's state to ensure accurate noise suppression while retaining key audio information. It mainly includes: band-targeted noise reduction, dynamic adjustment of noise reduction intensity, and user state adaptation.

6. The adaptive audio adjustment method based on headphone noise reduction according to claim 1, wherein, The headphone noise reduction during the audio playback process is to optimize the listening experience by adjusting the frequency response curve dynamically optimized based on the audio equalization parameters of the audio output according to the user's state and the noise environment.

7. An adaptive audio adjustment method based on headphone noise reduction according to claim 1, characterized in that Step S3 also includes: Continuously monitor the noise reduction effect and user feedback to form a closed-loop optimization of headphone noise reduction and adaptive audio adjustment: As an example of bio-signal feedback, if the user's heart rate continues to rise, automatically enhance the noise reduction or switch to a soothing sound effect; When the accelerometer detects static, gradually restore the default noise reduction mode.

8. An adaptive audio adjustment method based on headphone noise reduction according to claim 1, characterized in that, In step S4, the evaluation of the noise reduction efficiency by calculating the SNR improvement value before and after noise reduction in real time includes: Measure the noise suppression ability of the noise reduction algorithm through the objective index SNR, avoid the deviation of subjective evaluation, and quantify the noise reduction effect; Identify the noise reduction shortboards in different frequency bands, provide a basis for parameter adjustment, and perform band-targeted analysis; Detect the noise reduction failure scenario and trigger emergency optimization.

9. An adaptive audio adjustment method based on headphone noise reduction according to claim 1, characterized in that, The driven reinforcement learning model uses the deep deterministic policy gradient model, which is suitable for continuous action spaces; Input the current state and output an action; Its calculation formula is: R = w1 ⋅ SNR improvement value + w2 ⋅ user satisfaction - w3 ⋅ power consumption; User satisfaction: Quantified according to the user's operation direction; Power consumption penalty: Linearly related to the processor load and noise reduction intensity; The training process is: Offline pre-training: Initialize the model using historical datasets; Online fine-tuning: When new data arrives, perform mini-batch gradient descent through the edge computing chip; Policy deployment: Push the optimized parameters to the ANC filter module and the audio equalizer.

10. An adaptive audio adjustment system based on headphone noise reduction, characterized in that, Including: A noise feature module, which is used to collect multi-source noise through the cooperation of the front-feed microphone and the feedback microphone hardware, analyze the multi-source noise to obtain noise features, and calculate a dynamic threshold based on the noise features to generate a noise reduction target curve; A multi-dimensional perception module, which is used to obtain user information through multi-dimensional perception, fuse user heart rate, skin conductance, and accelerometer data, and identify the user's state through a pre-trained classification model; A dynamic noise reduction module, which is used to dynamically configure the noise reduction strategy parameters in the corresponding scenario based on the noise features and user state labels, and perform noise reduction during the audio playback process; A closed-loop optimization module, which is used to evaluate the noise reduction efficiency by calculating the real-time SNR improvement value before and after noise reduction, and optimize the policy parameters by combining the user's manual adjustment records to drive the reinforcement learning model.

Images