AI sound effect system self-adaptive to seat position and angle and adjusting method

Abstract

The invention discloses a seat position and angle self-adaptive AI sound effect system, which comprises a data acquisition layer used for acquiring geometric parameters and environmental noise information of a seat in real time and calculating real-time space coordinates of ears of a passenger relative to a loudspeaker based on the geometric parameters; the AI algorithm layer comprises a CNN-LSTM hybrid neural network model and is used for receiving the real-time space coordinates and the environmental noise information as input and outputting sound effect adjustment parameters, and the sound effect adjustment parameters comprise loudness gain G, sound channel delay tau and equalizer EQ parameters theta; and the execution control layer is used for carrying out real-time processing and rendering on the multichannel audio signals according to the sound effect adjustment parameters and outputting the processed signals to a loudspeaker for playing. According to the invention, the problem that the sound effect cannot dynamically adapt to the position change of the seat in real time in the prior art is solved, automatic real-time adaptation of the sound effect is realized, sound field positioning deviation and frequency response distortion are effectively avoided, and the comfort and personalization of vehicle-mounted audio experience are improved.

Description

Technical Field

[0001] This invention relates to the fields of automotive electronics and intelligent audio technology, and more specifically, to an AI sound system and adjustment method that adapts to the seat position and angle. Background Technology

[0002] With the continuous evolution of automotive smart cockpit technology, users' demands for in-vehicle audio experience have gradually shifted from static, fixed preset modes to dynamic, real-time adaptive modes. However, existing technologies face multiple obstacles in achieving this transformation. Traditional in-vehicle audio systems generally use preset modes for management, such as a driver-only mode or a rear-seat optimized mode. Such solutions have significant limitations when the seat position changes due to sliding, height adjustment, or angle adjustment.

[0003] When changes in parameters such as seat back angle and headrest height alter the relative spatial relationship between the passenger and the speakers, a significant shift in sound field localization occurs. For example, human voice localization shifts from the central region to a lateral position, accompanied by distortion in frequency response characteristics, particularly nonlinear attenuation of low-frequency energy. Furthermore, existing systems fail to effectively integrate seat pressure distribution data with passenger posture angle information, making it impossible to accurately identify differences in passenger body shape and dynamic posture changes. For instance, when passengers are in a forward-leaning working posture or a reclining resting posture, their ear sensitivity to sound changes, but the system lacks targeted sound compensation strategies, resulting in a uniform sound effect scheme that fails to meet diverse auditory needs.

[0004] Current technical architectures largely rely on manual sound effect switching after seat adjustment, forming a passive response mechanism. This lacks a real-time closed loop for data collection and execution feedback, and also fails to introduce AI-driven dynamic decision-making capabilities. The adjustment process suffers from significant delays and cannot be personalized by incorporating passengers' historical auditory preferences, resulting in a disjointed overall experience.

[0005] Although some technical solutions attempt to improve sound performance through sound field calibration, these methods have failed to establish a precise mapping relationship between seat dynamic parameters and sound compensation parameters, nor have they integrated artificial intelligence real-time decision-making mechanisms. Therefore, they cannot systematically solve the problem of synergy between dynamic adjustment, real-time adaptation and personalized optimization. Summary of the Invention

[0006] The purpose of this invention is to provide an AI sound effect system and adjustment method that adapts to the seat position and angle, which realizes automatic real-time adaptation of sound effects, effectively avoids sound field positioning deviation and frequency response distortion, and improves the comfort and personalization of the in-vehicle audio experience.

[0007] The above-mentioned technical objective of the present invention is achieved through the following technical solution:

[0008] An AI sound system that adapts to seat position and angle includes:

[0009] The data acquisition layer is used to acquire the geometric parameters of the seat and environmental noise information in real time, and calculate the real-time spatial coordinates of the passenger's ear relative to the speaker based on the geometric parameters.

[0010] The AI ​​algorithm layer includes a CNN-LSTM hybrid neural network model, which receives the real-time spatial coordinates and environmental noise information as input and outputs sound effect adjustment parameters, including loudness gain G, channel delay τ and equalizer EQ parameter θ.

[0011] The execution control layer is used to process and render the multi-channel audio signal in real time according to the sound effect adjustment parameters, and output the processed signal to the speaker for playback.

[0012] Furthermore, the data acquisition layer includes:

[0013] Angle sensor used to measure the leg angle of the seat. and the upturned corner of the seat cushion ;

[0014] Displacement sensor, used to measure the seat's fore-and-aft displacement x and height displacement z;

[0015] A noise sensor is used to collect environmental noise information.

[0016] Furthermore, the real-time spatial coordinates are calculated using the following method:

[0017] Establish a three-dimensional coordinate system with the seat design reference point R as the origin, and calculate the spatial coordinates of the passenger's ear relative to the speaker using the following formula. ;

[0018]

[0019] in, Indicates the fore-and-aft displacement of the seat. This represents the horizontal distance from point R to the passenger's ear. For the body and leg angles, The seat cushion has an upturned corner. Let R be the height of point R along the Y-axis. Let R be the coordinates of point R along the axis.

[0020] Furthermore, the AI ​​algorithm layer performs the following operations:

[0021] The spatial coordinates The CNN-LSTM hybrid neural network model is pre-trained with environmental noise information as input, and outputs loudness gain G, channel delay τ and equalizer EQ parameters θ. The audio signal is dynamically adjusted in real time using the parameter group {G,τ,θ}.

[0022] The loudness gain G is used to fuse distance attenuation and AI compensation, correcting for loudness attenuation caused by distance. The formula used is as follows:

[0023]

[0024]

[0025] In the formula, For reference distance, For real-time ear distance, The compensation value is the output of the CNN-LSTM hybrid neural network model, and N is the environmental noise information;

[0026] The channel delay τ is used to correct for sound field localization deviations and ensure that sound arrives at the ear synchronously. The formula used is as follows:

[0027]

[0028] In the formula, For real-time ear distance, For reference distance, For the speed of sound propagation, Optimized vocal tract delay compensation values ​​for CNN-LSTM hybrid neural network models;

[0029] The equalizer (EQ) parameters are spatial rotation angle parameters used to support three-dimensional sound field localization calculations.

[0030] Furthermore, the training process of the CNN-LSTM hybrid neural network model employs a multi-objective loss function:

[0031]

[0032] Where Loss is a multi-objective loss function. These are the weighting coefficients for loudness gain. The loudness gain predicted by the model. For true loudness gain, This is the weighting coefficient for the channel delay. For the vocal tract delay predicted by the model, For true channel delay, These are the weighting coefficients of the equalizer parameters. The equalizer EQ parameters predicted by the model. For actual equalizer (EQ) parameters, It is the Euclidean norm.

[0033] An AI-powered sound effect adjustment method that adapts to seat position and angle includes the following steps:

[0034] Real-time acquisition of seat geometry parameters and environmental noise information;

[0035] The real-time spatial coordinates of the passenger's ear relative to the speaker are calculated based on the collected geometric parameters;

[0036] The real-time spatial coordinates and environmental noise information are input into a pre-trained CNN-LSTM hybrid neural network model;

[0037] The loudness gain G, channel delay τ, and EQ parameter θ output by the CNN-LSTM hybrid neural network model are obtained, and the multi-channel audio signal is rendered and played in real time using the parameter group {G,τ,θ}.

[0038] In summary, the present invention has the following beneficial effects:

[0039] By acquiring seat geometry parameters and environmental noise information in real time, calculating the spatial coordinates of the passenger's ears, processing output sound effect adjustment parameters with AI algorithms, and performing real-time rendering, the problem of sound effects not being able to dynamically adapt to changes in seat position in real time in existing technologies has been solved. It has achieved automatic real-time adaptation of sound effects, effectively avoiding sound field positioning offset and frequency response distortion, and improving the comfort and personalization of the in-vehicle audio experience. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the AI ​​sound effect system for adaptive seat position and angle described in this invention.

[0041] Figure 2 This is a flowchart of the AI ​​sound effect method for adaptive seat position and angle described in this invention. Detailed Implementation

[0042] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below with reference to the figures and specific embodiments.

[0043] like Figure 1 As shown, the present invention proposes an AI sound effect system that adapts to seat position and angle. The system includes a data acquisition layer, an AI algorithm layer, and an execution control layer.

[0044] The data acquisition layer is configured to acquire the seat's geometric parameters and environmental noise information in real time, and calculate the real-time spatial coordinates of the passenger's ear relative to the speaker based on these geometric parameters.

[0045] The AI ​​algorithm layer contains a CNN-LSTM hybrid neural network model, which is configured to receive real-time spatial coordinates and environmental noise information as input and output sound effect adjustment parameters, including loudness gain G, channel delay τ, and equalizer EQ parameters θ.

[0046] The execution control layer is configured to process and render multi-channel audio signals in real time according to sound effect adjustment parameters, and output the processed signals to the speakers for playback.

[0047] For ease of understanding, some key technical concepts in this invention are explained below:

[0048] The data acquisition layer is configured to acquire the basic data required for system operation, such as the geometry of the seat and the acoustic conditions of the environment.

[0049] Geometric parameters are quantitative data that indicate the current physical position and posture of a seat, such as the seat's fore-aft position, height, and the angles of the backrest and seat cushion.

[0050] Environmental noise information refers to real-time noise data inside or outside a vehicle, which is used to assess the impact of the current acoustic environment on the sound experience.

[0051] Real-time spatial coordinates describe the precise position of a passenger's ear relative to the in-vehicle speakers in three-dimensional space, and these coordinates are updated in real time as the seat is dynamically adjusted.

[0052] The AI ​​algorithm layer is the core of the system's intelligent decision-making. It analyzes the input data through complex computational models and generates sound effect adjustment instructions.

[0053] The CNN-LSTM hybrid neural network model is a deep learning model that combines the features of convolutional neural networks (CNN) and long short-term memory networks (LSTM). It excels at processing temporal data and extracting spatial features, and is used to predict and optimize sound effect parameters.

[0054] The sound effect adjustment parameters are specific values ​​output by the AI ​​algorithm layer to guide the sound effect system in making adjustments. They include loudness gain G, channel delay τ, and equalizer EQ parameters θ.

[0055] Loudness gain G is used to adjust the overall volume of an audio signal to compensate for distance attenuation or the effects of ambient noise.

[0056] Channel delay τ is used to adjust the arrival time of audio signals from different channels to correct sound field positioning deviations and ensure sound synchronization.

[0057] The EQ parameter θ is used to adjust the intensity of the audio signal in different frequency ranges to optimize the timbre balance and frequency response characteristics.

[0058] The execution control layer is configured to receive instructions from the AI ​​algorithm layer and translate them into actual processing operations on the audio signal, thereby driving the speaker to play.

[0059] Multichannel audio signals refer to signals that contain multiple independent audio channels, which can achieve richer spatial sound effects.

[0060] A loudspeaker is a device that converts electrical signals into sound waves, and it is responsible for playing the processed audio signals.

[0061] The main technical features of the AI ​​sound effect system for adaptive seat position and angle of the present invention can be described in detail as follows:

[0062] The data acquisition layer is configured to acquire the seat's geometric parameters and ambient noise information in real time, and calculate the real-time spatial coordinates of the passenger's ear relative to the speakers based on these geometric parameters. The seat's geometric parameters can be acquired in various ways. For example, a series of sensors can be placed on the seat, triggering signals when the seat moves to a specific position to indicate its current state. Ambient noise information can be collected by deploying one or more general-purpose microphones within the vehicle compartment. Based on the acquired geometric parameters, the real-time spatial coordinates of the passenger's ear relative to the speakers can be estimated using a pre-defined geometric model. For example, a simplified in-vehicle spatial model can be established, treating the passenger's ear as a reference point fixed to the seat. Using the known geometric parameters of the seat and the fixed positions of the speakers, basic geometric calculation methods, such as trigonometric functions or vector operations, can be employed to determine the distance and angular relationships between this reference point and each speaker.

[0063] The AI ​​algorithm layer comprises a CNN-LSTM hybrid neural network model, configured to receive real-time spatial coordinates and environmental noise information as input and output sound effect adjustment parameters, including loudness gain G, channel delay τ, and equalizer (EQ) parameters θ. The system maps the received real-time spatial coordinates and environmental noise information to predefined sound effect adjustment parameters based on a pre-defined rule base. For example, when an increase in the distance between a passenger's ear and a speaker is detected, the system can increase the loudness gain G of that speaker according to preset rules. The channel delay τ can be calculated simply based on the speed of sound propagation and distance difference. The equalizer (EQ) parameter θ can be adjusted according to a preset frequency response curve to compensate for attenuation or enhancement of specific frequencies. Alternatively, the AI ​​algorithm layer can employ a shallow neural network model, learning from limited sample data to establish a non-linear relationship between real-time spatial coordinates, environmental noise information, and sound effect adjustment parameters, thereby achieving initial adaptive adjustment of the sound effects.

[0064] The execution control layer is configured to process and render the multi-channel audio signal in real time according to the audio effect adjustment parameters, and output the processed signal to the speakers for playback. The execution control layer can be implemented by a digital signal processor (DSP) or a microcontroller. This processor receives the audio effect adjustment parameters output from the AI ​​algorithm layer and performs real-time digital processing on the multi-channel audio signal. For example, loudness gain G can be adjusted by adjusting the digital volume control module; channel delay τ can be precisely controlled by digital delay lines; and equalizer (EQ) parameters θ can be adjusted by digital filters to change the frequency response of the audio signal. The processed multi-channel audio signal is then sent to the corresponding speakers for playback.

[0065] This invention acquires seat geometric parameters and environmental noise information in real time through a data acquisition layer and calculates the real-time spatial coordinates of the passenger's ears. The AI ​​algorithm layer utilizes a CNN-LSTM hybrid neural network model to output sound effect adjustment parameters based on this information. The execution control layer then processes and renders the multi-channel audio signals in real time. Therefore, this invention effectively solves the problems of sound field positioning shift, frequency response distortion, and lack of real-time personalized adaptation in traditional in-vehicle audio systems during dynamic seat adjustments, achieving millisecond-level adaptive compensation for sound effects and enhancing the user's auditory experience in smart cockpit scenarios.

[0066] In some of the embodiments of the present invention described above, a data acquisition layer is proposed to acquire seat geometric parameters and environmental noise information in real time. However, in its implementation, since the specific sensor configuration is not clearly defined, it may be impossible to accurately measure key dynamic parameters of the seat such as angle, displacement and environmental noise, thereby affecting the accuracy and completeness of real-time spatial coordinate calculation, and causing problems such as sound field shift, frequency response distortion and insufficient environmental adaptability during sound effect adjustment.

[0067] To address this, the present invention further proposes that the data acquisition layer includes an angle sensor, a displacement sensor, and a noise sensor. The angle sensor is used to measure the leg angle of the seat. and the upturned corner of the seat cushion The displacement sensor is used to measure the seat's forward and backward displacement (x) and vertical displacement (z); the noise sensor is used to collect environmental noise information.

[0068] The angle sensor is a device capable of sensing and quantifying the degree of rotation or tilt of an object. Its function is to accurately capture the posture changes of various components of the seat, and these angular parameters are directly related to the passenger's sitting posture and ear orientation. In practical applications, angle sensors can be implemented in various ways. For example, a combination sensor of a microelectromechanical system (MEMS) gyroscope and accelerometer can be used, integrated into the seat frame or key rotating joints. An inertial measurement unit (IMU) can then sense the tilt angle and rotational changes of various seat components in real time, thereby obtaining the body-leg angle. and the upturned corner of the seat cushion .

[0069] The displacement sensor is a device used to measure changes in the linear position or distance of an object. Its function is to accurately track the position of the seat in three-dimensional space, which is crucial for determining the real-time distance between the passenger's ear and the speaker. In practical applications, displacement sensors can be implemented in various ways. For example, a linear potentiometer or resistance ruler can be used, mounted on the seat rails and lifting mechanism, to obtain the seat's fore-aft displacement (x) and height displacement (z) by measuring the relative position change of the slider or lifting rod. Another implementation method is to use a Hall effect sensor array, arranging magnets and a Hall sensor array on the seat rails or lifting mechanism. Displacement is accurately measured by detecting changes in magnetic field strength, offering advantages such as non-contact operation and long lifespan.

[0070] The noise sensor is a device that converts sound waves into electrical signals. Its function is to collect environmental noise information inside the vehicle cabin in real time, providing a data basis for subsequent sound effect compensation. In practical applications, noise sensors can be implemented in various ways. For example, electret microphones can be used, which are low-cost, small in size, and stable in performance, and can be widely deployed inside the vehicle cabin, such as in the ceiling, dashboard, or near seats, to collect ambient noise. Another approach is to use a MEMS microphone array. By forming an array of multiple MEMS microphones and combining it with beamforming and other technologies, noise sources from different directions inside the vehicle can be more accurately identified and separated, providing more accurate environmental noise information.

[0071] Through the above technical solutions, the data acquisition layer can comprehensively and accurately acquire the dynamic parameters of the seat and environmental noise information. Angle sensors precisely capture the leg angle and seat cushion tilt angle, directly reflecting changes in passenger posture and providing crucial input for real-time spatial coordinate calculations, effectively avoiding sound field positioning shifts caused by posture deviations. Displacement sensors accurately measure the seat's fore-and-aft and height displacements, and combined with angle parameters, fully describe the seat's position in three-dimensional space, laying the foundation for calculating the coordinates of the passenger's ear relative to the speaker and ensuring real-time response to position changes. Noise sensors collect environmental noise information in real time, enabling the system to effectively compensate for noise and improve the clarity and adaptability of sound effects in noisy environments. The synergistic effect of these sensors ensures the accuracy and comprehensiveness of data acquisition, solves potential data loss or inaccuracy issues, and provides reliable data support for precise sound effect adjustment by the AI ​​algorithm layer and execution control layer, thereby significantly improving the adaptive capability of the sound system and the user experience.

[0072] In some of the embodiments of the present invention described above, a data acquisition layer is proposed to obtain the geometric parameters of the seat and environmental noise information. However, in its implementation, there are shortcomings in how to accurately calculate the real-time spatial coordinates of the passenger's ear relative to the speaker based on these parameters to ensure the accuracy and real-time performance of the sound effect adjustment. Specifically, there is a lack of a unified calculation benchmark and quantification method, which may introduce errors in the spatial coordinate calculation and affect the accuracy of subsequent adaptive sound effect adjustment.

[0073] To address this, the present invention further proposes the following method for calculating real-time spatial coordinates:

[0074] Establish a three-dimensional coordinate system with the seat design reference point R as the origin, and calculate the spatial coordinates of the passenger's ear relative to the speaker using the following formula. ;

[0075]

[0076] in, Indicates the fore-and-aft displacement of the seat. This represents the horizontal distance from point R to the passenger's ear. For the body and leg angles, The corners of the seat cushion are upturned. Let R be the height of point R along the Y-axis. Let R be the coordinates of point R along the axis.

[0077] This real-time spatial coordinate calculation method aims to transform the dynamic geometric parameters of the seat into precise positional information of the passenger's ear in three-dimensional space, providing accurate input for subsequent sound effect algorithms. Its function is to establish a quantitative relationship between seat status and listening position, ensuring the accuracy of the physical basis for sound effect adjustment.

[0078] A three-dimensional coordinate system is established with the seat design reference point R as the origin. R typically refers to the H point (Hip Point) in automotive seat design, representing the reference point where the occupant's buttocks meet the seat cushion and backrest. Establishing a three-dimensional coordinate system with R as the origin provides a relatively stable reference benchmark with respect to the occupant's posture. This allows subsequent calculations of the ear's spatial coordinates to exclude the influence of overall vehicle movement or differences in seat installation position, focusing instead on changes in ear position caused by variations in the occupant's own posture.

[0079] The spatial coordinates of the passenger's ear relative to the speaker are calculated using the formula described above. This formula is the core mathematical model used to comprehensively map various geometric parameters of the seat (such as displacement and angle) to the three-dimensional spatial coordinates of the passenger's ear. Its purpose is to provide a precise and quantifiable method to track changes in the passenger's ear position relative to the speaker in real time, thus providing an accurate physical basis for adaptive adjustment of sound effects. This formula can be directly programmed into the onboard processor or digital signal processor (DSP), receiving sensor data from the data acquisition layer as input and outputting the calculated spatial coordinates in real time.

[0080] Here, x represents the fore-and-aft displacement of the seat. This parameter directly affects the distance between the passenger's ear and the front speaker, and is one of the key factors influencing sound field localization and loudness attenuation. This fore-and-aft displacement can be obtained by directly measuring the movement distance of the seat rail using a displacement sensor (such as a linear potentiometer or Hall effect sensor). Alternatively, it can be calculated by accumulating the pulse signals fed back from the seat motor encoder.

[0081] This parameter represents the horizontal distance from point R to the passenger's ear. It reflects the projected distance from the seat reference point R to the passenger's ear on the horizontal plane. It is a crucial component in determining the ear's lateral and longitudinal position, reflecting differences in passenger body shape and posture. This horizontal distance can be preset based on ergonomic data or dynamically adjusted by estimating passenger body shape and posture using seat pressure sensors, visual recognition systems, etc.

[0082] The body-to-leg angle, typically referring to the angle between a passenger's thigh and torso, reflects their sitting posture, such as whether they are leaning forward, upright, or reclining. Changes in this angle affect the degree of torso tilt, thus altering the vertical position of the ears. The body-to-leg angle can be measured using angle sensors (such as gyroscopes or accelerometers) mounted on the seat cushion or backrest, or indirectly estimated by analyzing seat pressure distribution data.

[0083] The seat cushion tilt angle refers to the angle at which the front end of the seat cushion is tilted relative to the horizontal plane. Adjusting this angle changes the relative height of the passenger's hips and legs, thus affecting the posture of the entire torso and consequently the vertical position of the ears. The seat cushion tilt angle can be obtained by directly measuring the tilt angle of the seat cushion using an angle sensor. Alternatively, it can be calculated using the encoder feedback signal from the seat adjustment motor.

[0084] The height of point R along the Y-axis represents the height of the seat reference point R on the Y-axis of the established three-dimensional coordinate system (typically the vehicle width direction). It is primarily used to correct the vertical reference position of point R, ensuring the accuracy of ear coordinate calculations. This height can be obtained by measuring the overall vertical movement distance of the seat using a seat height displacement sensor (such as a linear potentiometer). Alternatively, it can be calculated using the encoder feedback signal from the seat adjustment motor.

[0085] The coordinates of point R in the axial direction represent the coordinates of the seat reference point R on the X-axis of the established three-dimensional coordinate system (typically the vehicle's fore-aft direction). It is used to correct the reference position of point R in the fore-aft direction, and together with the seat's fore-aft displacement, determines the final fore-aft position of the ear. These coordinates can be obtained by measuring the overall fore-aft movement of the seat using a fore-aft displacement sensor (such as a linear potentiometer). Alternatively, they can be calculated using the encoder feedback signal from the seat adjustment motor.

[0086] Through the above technical solution, this invention can accurately calculate the real-time spatial coordinates of the passenger's ear relative to the speaker. A three-dimensional coordinate system is established with the seat design reference point R as the origin, providing a stable and standardized reference benchmark, effectively avoiding subjectivity in coordinate system selection and calculation errors. By integrating key geometric parameters such as the seat's fore-and-aft displacement, the horizontal distance from point R to the passenger's ear, the leg angle, the seat cushion's upward angle, the height of point R in the Y-axis direction, and the coordinates of point R in the axial direction, and using specific mathematical formulas for calculation, real-time, dynamic tracking of the passenger's ear position is achieved. This precise spatial coordinate calculation capability provides highly accurate input data for the subsequent AI algorithm layer to output loudness gain G, channel delay τ, and equalizer EQ parameters θ, thereby ensuring the accuracy and real-time performance of sound effect adjustment. Therefore, when the seat position or angle changes dynamically, the system can sense and quantify the change in ear position in milliseconds, effectively eliminating sound field positioning offset and frequency response distortion, significantly improving the adaptive compensation effect of in-vehicle audio and the user's listening experience.

[0087] In some of the embodiments of the present invention described above, it is proposed to calculate the real-time spatial coordinates of the passenger's ear relative to the speaker to adjust the sound effect. However, in its implementation, there is a lack of an efficient method to integrate distance attenuation, environmental noise and AI compensation, which leads to inaccurate adjustment of sound effect parameters and the inability to correct loudness attenuation and sound field positioning deviation caused by distance in real time, thereby affecting the adaptability and accuracy of the sound effect.

[0088] In response, this invention further proposes that the AI ​​algorithm layer perform the following operations:

[0089] The spatial coordinates The CNN-LSTM hybrid neural network model is pre-trained with environmental noise information as input, and outputs loudness gain G, channel delay τ and equalizer EQ parameters θ. The audio signal is dynamically adjusted in real time using the parameter group {G,τ,θ}.

[0090] The loudness gain G is used to fuse distance attenuation and AI compensation, correcting for loudness attenuation caused by distance. The formula used is as follows:

[0091]

[0092]

[0093] In the formula, For reference distance, For real-time ear distance, The compensation value is the output of the CNN-LSTM hybrid neural network model, and N is the environmental noise information;

[0094] Specifically, the spatial coordinates The seat's geometric parameters (such as body-leg angle) are obtained through the data acquisition layer. The corners of the seat cushion are raised. The real-time position information of the passenger's ear relative to the speaker, calculated from the front-to-back displacement (x) and height-to-height displacement (z), provides precise three-dimensional data of the passenger's listening position, forming the basis for adaptive sound effect adjustment. The environmental noise information, collected by noise sensors, reflects the real-time noise level and spectral characteristics inside the carriage. Its purpose is to provide environmental interference data for the AI ​​model, enabling it to effectively suppress the impact of noise on hearing or compensate based on noise characteristics during sound effect adjustment. The pre-trained CNN-LSTM hybrid neural network model is a deep learning architecture that combines the advantages of convolutional neural networks (CNN) and long short-term memory networks (LSTM). CNN excels at extracting spatial features from input data, such as identifying specific geometric patterns from spatial coordinates; LSTM excels at processing sequential data and capturing temporal dependencies, such as the dynamic trends of seat position or environmental noise. Pre-training means that the model has been trained and optimized with a large amount of simulated data or real-world driving scenario data before deployment, enabling it to learn the complex mapping relationship between seat dynamic parameters, environmental noise, and optimal sound effect adjustment parameters.

[0095] Loudness gain G is a parameter used to adjust the overall volume of the audio signal. Its core function is to integrate distance attenuation and AI compensation to correct for loudness attenuation caused by changes in the distance between the passenger's ear and the speaker. When the passenger's ear is farther from the speaker, the loudness naturally attenuates, and the G value increases accordingly to compensate; conversely, it decreases. Furthermore, the AI ​​compensation component can intelligently adjust based on factors such as environmental noise to ensure consistent perceived loudness in different scenarios. The reference distance... This is a preset, ideal listening distance, typically a standard listening position determined during system design. The real-time ear distance is the actual distance between the passenger's ear and the speaker, expressed in spatial coordinates. The calculation yielded the result.

[0096] The compensation value output by the CNN-LSTM hybrid neural network model This is an additional loudness compensation amount intelligently generated by a pre-trained CNN-LSTM hybrid neural network model based on the input (spatial coordinates and environmental noise information). Its function is to compensate for complex factors that a purely physical distance attenuation model cannot consider, such as the influence of environmental noise on loudness perception and the differences in human ear sensitivity to loudness at different frequencies, thereby achieving more refined loudness adjustment. N represents the environmental noise information, which is used as the output compensation value of the CNN-LSTM hybrid neural network model in the calculation of loudness gain G. An important input is the loudness compensation, which enables the model to dynamically adjust the loudness compensation based on the intensity and spectral characteristics of real-time ambient noise. For example, it can appropriately increase the loudness to maintain clarity when the noise is high, or suppress it at specific noise frequencies.

[0097] The channel delay τ is used to correct for sound field localization deviations and ensure that sound arrives at the ear synchronously. The formula used is as follows:

[0098]

[0099] In the formula, For real-time ear distance, For reference distance, For the speed of sound propagation, Optimized vocal tract delay compensation values ​​for CNN-LSTM hybrid neural network models;

[0100] Channel delay τ is a parameter used to adjust the time difference between the arrival times of audio signals from different channels at the passenger's ear. Its function is to correct sound field localization deviations, ensuring that sound arrives at the ear synchronously, thereby maintaining the accuracy and stability of the sound image. By precisely calculating and applying the delay, the position of the sound source in space can be simulated, allowing the listener to perceive that the sound comes from the expected direction. The speed of sound propagation... It is a physical constant, typically taken as the speed of sound in air (approximately 343 m / s), and its function is to convert distance differences into time differences, forming the basis for calculating the vocal tract delay τ. The vocal tract delay compensation value optimized by the CNN-LSTM hybrid neural network model... It is an additional channel delay compensation amount intelligently generated by the pre-trained CNN-LSTM hybrid neural network model based on the input. Its function is to compensate for complex factors that the pure physical distance difference model cannot consider, such as sound field distortion caused by sound wave reflection and refraction in the carriage, as well as the psychoacoustic effect of human ear on sound source localization, thereby achieving more accurate sound field localization correction.

[0101] The equalizer (EQ) parameters are spatial rotation angle parameters. The head-related transfer function (HRTF) can be used to convert these spatial rotation angles into EQ parameters to support 3D sound field localization calculations. The EQ parameter θ is used to adjust the gain or attenuation of the audio signal at different frequencies; here, it specifically refers to the spatial rotation angle parameter. Its function is to support 3D sound field localization calculations. By adjusting the response at specific frequencies, it can simulate the angular changes of the sound source in 3D space, compensating for frequency response distortion caused by changes in the angle of the passenger's ears (such as head tilt), thereby enhancing the immersion and realism of the 3D sound field.

[0102] Real-time dynamic adjustment of audio signals using parameter sets {G,τ,θ} refers to the system's ability to instantly and continuously modify and optimize multi-channel audio signals based on the loudness gain G, channel delay τ, and equalizer (EQ) parameters θ output by the AI ​​algorithm layer. Its purpose is to ensure that sound effect adjustments remain highly synchronized with changes in seat position, angle, and ambient noise, avoiding auditory delays or discomfort. Implementation methods can include: integrating corresponding algorithm modules into a digital signal processor (DSP) to receive parameters output by the AI ​​model and adjust the gain, delay, and frequency response of the audio stream with extremely low latency; or utilizing a high-performance embedded system to achieve fast and precise control of the audio signal through software-defined radio (SDR) or dedicated audio processing chips.

[0103] Through the above technical solution, this invention, based on existing real-time spatial coordinate calculation, further introduces an AI algorithm layer to perform refined and intelligent real-time dynamic adjustment of sound effect parameters. Specifically, the real-time spatial coordinates of the passenger's ears and environmental noise information are input into a pre-trained CNN-LSTM hybrid neural network model. This model can comprehensively consider distance attenuation, environmental interference, and complex acoustic psychological effects, outputting highly optimized loudness gain G, channel delay τ, and equalizer EQ parameter θ. The loudness gain G, by integrating physical distance attenuation and AI intelligent compensation, can accurately correct the loudness attenuation caused by changes in seat position and adaptively adjust according to environmental noise, ensuring a consistent and comfortable listening experience under any circumstances. The channel delay τ, by combining physical distance difference and AI-optimized compensation values, corrects sound field positioning deviation in real time, ensuring that multi-channel audio signals arrive at the passenger's ears synchronously, thereby maintaining the stability and accuracy of the sound image and avoiding sound field shift. The equalizer EQ parameter θ, as a spatial rotation angle parameter, can support three-dimensional sound field positioning calculation, further compensating for frequency response distortion caused by changes in the passenger's head posture, and enhancing the immersiveness of the three-dimensional sound field. Ultimately, by using the parameter group {G,τ,θ} to dynamically adjust the multi-channel audio signal in real time, this invention achieves millisecond-level sound effect response, significantly improving the accuracy, real-time performance, and adaptability of sound effect adjustment. It effectively solves the problems of inaccurate sound effect parameter adjustment, inability to correct distance attenuation and sound field positioning deviation in real time in traditional solutions, thereby providing passengers with a more personalized and immersive listening experience.

[0104] In some of the embodiments of the present invention, a CNN-LSTM hybrid neural network model is proposed to output sound effect adjustment parameters. However, during the training process, since multiple output targets such as loudness gain, channel delay and equalizer EQ parameters need to be optimized at the same time, there is a lack of an effective loss function to balance the optimization weights of these parameters. This may lead to optimization conflicts or deviations between parameters during model training, affecting prediction accuracy and sound effect adaptation.

[0105] To address this, the present invention further proposes a multi-objective loss function for the training process of the CNN-LSTM hybrid neural network model:

[0106]

[0107] Where Loss is a multi-objective loss function. These are the weighting coefficients for loudness gain. The loudness gain predicted by the model. For true loudness gain, This is the weighting coefficient for the channel delay. For the vocal tract delay predicted by the model, For true channel delay, These are the weighting coefficients of the equalizer parameters. The equalizer EQ parameters predicted by the model. For actual equalizer (EQ) parameters, It is the Euclidean norm.

[0108] A multi-objective loss function is a mathematical function used in machine learning model training to simultaneously optimize multiple interrelated or potentially conflicting output objectives (such as loudness gain, channel delay, and equalizer (EQ) parameters). It combines the loss terms corresponding to each objective (e.g., through weighted summation) into a single scalar value to guide model parameter updates. Its aim is to find a balance between different objectives, preventing the model from severely compromising the performance of other objectives while optimizing one. For example, besides weighted summation, Pareto optimization can be used to find a set of non-dominated solutions among multiple objectives, or dynamic weight adjustment strategies can be employed to automatically adjust the weights of each objective based on training progress.

[0109] The weighting coefficient of loudness gain is used to quantify the relative importance of loudness gain in the overall optimization objective. By adjusting this coefficient, the sensitivity of the model to loudness gain prediction errors during training can be controlled. For example, when the accuracy of loudness gain is crucial to user experience, it can be assigned a higher weighting coefficient, prompting the model to prioritize optimizing the prediction accuracy of loudness gain. This weighting coefficient can be determined through hyperparameter tuning methods such as empirical setting, cross-validation, grid search, or Bayesian optimization. In supervised learning, the loudness gain predicted by the model refers to the loudness gain value calculated by the CNN-LSTM hybrid neural network model based on input data (such as real-time spatial coordinates and ambient noise information). The true loudness gain refers to the actual loudness gain value in the training dataset that corresponds to the input data and has been precisely measured or annotated by experts. The goal of model training is to make the predicted value as close as possible to the true value; the difference between the two constitutes the error term of the loudness gain.

[0110] Similar to the weighting coefficient for loudness gain, the weighting coefficient for channel delay is used to specify the relative priority of channel delay in multi-objective optimization. For example, if the accuracy of sound field localization is a core requirement of the system design, a higher weight can be assigned to channel delay to ensure that the model has higher accuracy in predicting channel delay. This weighting coefficient can also be set using various hyperparameter tuning techniques. The channel delay predicted by the model is the channel delay value calculated by the CNN-LSTM hybrid neural network model based on the input data. The true channel delay is the actual channel delay value in the training dataset corresponding to the input data. The model learns how to accurately adjust sound field localization by minimizing the difference between the predicted and true values.

[0111] The weighting coefficients of the equalizer parameters are used to balance the contribution of the equalizer EQ parameters to the overall loss function. For example, if a high degree of fine-tuning of timbre and frequency response is required, the weights of the equalizer parameters can be appropriately increased, prompting the model to predict the EQ parameters more accurately. The method for determining these weighting coefficients is similar to that for loudness gain and channel delay. The equalizer EQ parameters predicted by the model are the values ​​calculated by the CNN-LSTM hybrid neural network model based on the input data. The actual equalizer EQ parameters are the corresponding values ​​in the training dataset. By comparing the predicted and actual values, the model learns how to dynamically adjust the frequency response of the audio based on seat position and ambient noise.

[0112] The Euclidean norm, also known as the L2 norm, is a commonly used distance metric for calculating the straight-line distance between two points in a vector space. In loss functions, it is typically used to quantify the magnitude of the error between the model's predicted values ​​and the true values. For example, for parameters in vector form (such as EQ parameters), the Euclidean norm can calculate the distance between the predicted vector and the true vector, thus providing a comprehensive measure of error. Its advantage lies in penalizing larger errors more severely, helping the model converge to a more accurate solution.

[0113] Through the above technical solution, this invention can effectively solve the optimization conflict or deviation problem that may exist between multiple output targets such as loudness gain, channel delay, and equalizer (EQ) parameters during the training process of CNN-LSTM hybrid neural network models. Specifically, the loss function weights and sums the prediction errors of each target (loudness gain, channel delay, and EQ parameters) and uses the Euclidean norm to accurately quantify the error, enabling the model to comprehensively consider the optimization needs of all key sound effect parameters during training. The introduction of weight coefficients for loudness gain, channel delay, and equalizer parameters allows for flexible adjustment of the optimization priority of different parameters according to the actual application scenario and the importance of the auditory experience, thereby avoiding the potential for neglecting one aspect while optimizing another that may result from optimizing a single target. For example, when sound field positioning accuracy is prioritized, the weight of channel delay can be increased; when volume balance is more critical, loudness gain can be emphasized. This refined weight control ensures that the model can achieve higher overall accuracy and robustness when predicting loudness gain G, channel delay τ, and equalizer (EQ) parameters θ. Ultimately, this enables the AI ​​sound system that adapts to seat position and angle to provide more precise and balanced real-time sound adjustment, significantly improving the passenger's auditory experience under different seat postures and environmental noise, eliminating sound field shift and frequency response distortion, and achieving millisecond-level adaptive compensation.

[0114] In some of the solutions described above in this invention, an AI sound effect system for adaptive seat position and angle is proposed to achieve dynamic sound effect adaptation. However, the adjustment response time is long and the real-time performance is insufficient during the execution process. The lack of an efficient data closed-loop processing mechanism makes it impossible to meet the requirements of millisecond-level adaptive compensation, thereby causing sound field positioning offset and frequency response distortion problems.

[0115] See Figure 2 The present invention further proposes an adjustment method for an AI sound effect system that adapts to the seat position and angle, the method comprising the following steps:

[0116] Real-time acquisition of seat geometry and ambient noise information. This step aims to obtain real-time data on the current physical state of the seat and the in-vehicle acoustic environment. The seat geometry reflects the changes in the passenger's position and posture relative to the speakers inside the vehicle, forming the basis for calculating sound field changes. Ambient noise information is used to assess the level of interference in the current listening environment in order to perform noise compensation or optimization.

[0117] The real-time spatial coordinates of the passenger's ear relative to the speaker are calculated based on the collected geometric parameters. The purpose of this step is to transform the abstract seat geometric parameters into concrete spatial coordinates that can be used for acoustic calculations, accurately describing the position of the passenger's ear relative to a fixed speaker in three-dimensional space. This is crucial for achieving precise sound field localization and compensation. Specifically, a three-dimensional coordinate system with the seat design reference point R as the origin can be pre-established. Based on the seat's geometric parameters and the pre-defined relative positional relationship between the passenger's ear and point R, the spatial coordinates of the passenger's ear relative to the speaker are calculated in real-time using a geometric transformation formula. For example, the formula described in claim 3 can be used for calculation. Alternatively, a mapping relationship between the passenger's head model and the seat's geometric parameters can be established, and machine learning or deep learning models can be used to directly predict the real-time spatial coordinates of the passenger's ear using the collected geometric parameters as input.

[0118] The real-time spatial coordinates and environmental noise information are input into a pre-trained CNN-LSTM hybrid neural network model. This step is the core intelligent decision-making process of this method, utilizing the pre-trained AI model to perform in-depth analysis of the real-time acquired spatial information and environmental noise to predict the optimal sound effect adjustment parameters. The CNN-LSTM hybrid neural network model can effectively process temporal data and spatial features, capturing complex nonlinear relationships. Specifically, the real-time calculated spatial coordinates of the passenger's ear and the real-time collected environmental noise information can be used as input vectors and fed into a CNN-LSTM hybrid neural network model trained on a large amount of acoustic data and user preference data. The CNN part can extract spatial features, and the LSTM part can process temporal changes, jointly predicting the loudness gain G, channel delay τ, and equalizer EQ parameters θ. Alternatively, the spatial coordinates and environmental noise information can be feature-engineered, for example, extracting high-level features such as distance, angle, and noise type, and then these features can be input into the pre-trained CNN-LSTM hybrid neural network model.

[0119] Finally, the loudness gain G, channel delay τ, and EQ parameter θ output by the CNN-LSTM hybrid neural network model are obtained, and the multi-channel audio signal is rendered and played in real time using the parameter set {G,τ,θ}. This step is the execution stage of sound effect adjustment. Based on the optimized parameters predicted by the AI ​​model, the multi-channel audio signal being played is subjected to real-time acoustic processing to compensate for sound field shift, frequency response distortion, and noise interference, ensuring that passengers obtain the best listening experience. Specifically, the audio processing unit receives the loudness gain G, channel delay τ, and equalizer EQ parameter θ output by the CNN-LSTM hybrid neural network model. The loudness gain G is used to adjust the volume of each channel to compensate for distance attenuation and AI compensation; the channel delay τ is used to apply a precise delay to each channel signal to correct the time difference of sound waves reaching the ear and ensure accurate sound image positioning; the equalizer EQ parameter θ is used to adjust the frequency response of the audio signal to compensate for frequency response distortion or achieve three-dimensional sound field positioning. These processed multi-channel audio signals are then played through speakers. Alternatively, the parameters output by the model can be directly mapped to control instructions from a digital signal processor (DSP). The DSP then adjusts various parameters of the audio signal in real time based on these instructions, achieving millisecond-level response speeds.

[0120] Through the aforementioned adjustment method, this invention enables real-time, dynamic, and precise adjustment of the AI ​​sound system for adaptive seat position and angle. Specifically, it collects the seat's geometric parameters and environmental noise information in real time, providing an immediate and comprehensive data foundation for subsequent intelligent decision-making, effectively avoiding the errors and delays caused by traditional manual or delayed data collection. Based on these collected geometric parameters, the system accurately calculates the real-time spatial coordinates of the passenger's ear relative to the speaker, allowing the system to accurately grasp changes in the passenger's listening position and providing precise spatial basis for sound field localization and compensation. Subsequently, these real-time spatial coordinates and environmental noise information are input into a pre-trained CNN-LSTM hybrid neural network model. This model can deeply learn and integrate complex spatial acoustic characteristics and environmental noise effects, intelligently predicting the optimal loudness gain G, channel delay τ, and EQ parameter θ. Finally, the system uses these parameter sets generated in real time by the AI ​​model to render and play multi-channel audio signals in real time, thereby achieving millisecond-level adaptive compensation for sound field localization offset, frequency response distortion, and environmental noise interference. Compared to existing technologies, this method significantly improves the real-time performance and accuracy of sound effect adjustment by constructing a closed-loop system of "data acquisition-AI decision-making-execution feedback". In particular, the introduction of the AI ​​algorithm layer enables the system to go beyond simple geometric compensation, integrating distance attenuation, AI compensation, sound field positioning deviation correction, and three-dimensional sound field positioning calculation. This allows it to provide a personalized and high-quality auditory experience instantly and seamlessly as the seat position and angle dynamically change, effectively solving the problems of long adjustment response time and insufficient real-time performance, and ensuring the dynamic adaptability of sound effects.

[0121] Through the above process, this system achieves millisecond-level adaptive compensation for changes in seat position and angle. Compared with the problem of "disconnect between sound field and seat adjustment" in existing technologies, this system eliminates sound field shift and frequency response distortion caused by seat adjustment through real-time data acquisition, precise spatial coordinate calculation, and AI intelligent decision-making. For example, when user A adjusts from a normal sitting posture to a reclining resting posture, traditional systems cannot compensate accordingly, which may lead to a decrease in high-frequency sensitivity. However, this system captures posture changes through an angle sensor, and the EQ parameter θ output by the AI ​​algorithm layer can provide targeted compensation for high frequencies, thus solving the problem of "lack of passenger posture adaptation". In addition, this system constructs a closed-loop system of "data acquisition-AI decision-making-execution feedback", achieving real-time and accurate sound effect adjustment, avoiding the problems of "insufficient AI coordination" in existing technologies, which leads to lag in adjustment response and lack of personalized experience. The system can automatically and seamlessly provide an optimized auditory experience without the user having to manually switch sound effect modes.

[0122] In this document, the terms "upper," "lower," "front," "back," "left," "right," "top," "bottom," "inner," "outer," "vertical," and "horizontal," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only used for the clarity of expressing the technical solution and for the convenience of description, and therefore should not be construed as limiting the present invention.

[0123] In this document, the terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, which includes not only the elements listed but also other elements not expressly listed.

[0124] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. An AI sound system that adapts to seat position and angle, characterized in that, include: The data acquisition layer is used to acquire the geometric parameters of the seat and environmental noise information in real time, and calculate the real-time spatial coordinates of the passenger's ear relative to the speaker based on the geometric parameters. The AI ​​algorithm layer includes a CNN-LSTM hybrid neural network model, which receives the real-time spatial coordinates and environmental noise information as input and outputs sound effect adjustment parameters, including loudness gain G, channel delay τ and equalizer EQ parameter θ. The execution control layer is used to process and render the multi-channel audio signal in real time according to the sound effect adjustment parameters, and output the processed signal to the speaker for playback.

2. The AI ​​sound effect system for adaptive seat position and angle according to claim 1, characterized in that, The data acquisition layer includes: Angle sensor used to measure the leg angle of the seat. and the upturned corner of the seat cushion ; Displacement sensor, used to measure the seat's fore-and-aft displacement x and height displacement z; A noise sensor is used to collect environmental noise information.

3. The AI ​​sound effect system for adaptive seat position and angle according to claim 2, characterized in that, The real-time spatial coordinates are calculated using the following method: Establish a three-dimensional coordinate system with the seat design reference point R as the origin, and calculate the spatial coordinates of the passenger's ear relative to the speaker using the following formula. ; in, Indicates the fore-and-aft displacement of the seat. This represents the horizontal distance from point R to the passenger's ear. For the body and leg angles, The seat cushion has an upturned corner. Let R be the height of point R along the Y-axis. Let R be the coordinates of point R along the axis.

4. The AI ​​sound effect system for adaptive seat position and angle according to claim 3, characterized in that, The AI ​​algorithm layer performs the following operations: The spatial coordinates The CNN-LSTM hybrid neural network model is pre-trained with environmental noise information as input, and outputs loudness gain G, channel delay τ and equalizer EQ parameters θ. The audio signal is dynamically adjusted in real time using the parameter group {G,τ,θ}. The loudness gain G is used to fuse distance attenuation and AI compensation, correcting for loudness attenuation caused by distance. The formula used is as follows: In the formula, For reference distance, For real-time ear distance, The compensation value is the output of the CNN-LSTM hybrid neural network model, and N is the environmental noise information; The channel delay τ is used to correct for sound field localization deviations and ensure that sound arrives at the ear synchronously. The formula used is as follows: In the formula, For real-time ear distance, For reference distance, For the speed of sound propagation, Optimized vocal tract delay compensation values ​​for CNN-LSTM hybrid neural network models; The equalizer (EQ) parameters are spatial rotation angle parameters used to support three-dimensional sound field localization calculations.

5. The AI ​​sound effect system for adaptive seat position and angle according to claim 3, characterized in that, The training process of the CNN-LSTM hybrid neural network model employs a multi-objective loss function: Where Loss is a multi-objective loss function. These are the weighting coefficients for loudness gain. The loudness gain predicted by the model. For true loudness gain, This is the weighting coefficient for the channel delay. For the vocal tract delay predicted by the model, For true channel delay, These are the weighting coefficients of the equalizer parameters. The equalizer EQ parameters predicted by the model. For actual equalizer (EQ) parameters, It is the Euclidean norm.

6. A method for adjusting an AI sound system based on the adaptive seat position and angle according to any one of claims 1-5, characterized in that, Including the following steps: S1. Real-time acquisition of seat geometric parameters and environmental noise information; S2. Calculate the real-time spatial coordinates of the passenger's ear relative to the speaker based on the collected geometric parameters; S3. Input the real-time spatial coordinates and environmental noise information into the pre-trained CNN-LSTM hybrid neural network model; S4. Obtain the loudness gain G, channel delay τ, and EQ parameter θ output by the CNN-LSTM hybrid neural network model, and use the parameter group {G,τ,θ} to render and play the multi-channel audio signal in real time.

Images