Personalized sound effect configuration method, system and device

Abstract

The application discloses a personalized sound effect configuration method, system and device, relates to the technical field of computers, and the method comprises the following steps: constructing a hearing sensation parameter model; constructing a sound effect objective feature model; performing feature extraction on the preference feedback information to generate target hearing preference features; determining target hearing sensation parameters according to the target hearing sensation mode; based on the hearing sensation parameter model, taking the target hearing sensation parameters as constraints, adopting a gradient descent algorithm to perform inversion solving in a parameter space of sound effect objective features to determine target sound effect objective features matched with the target hearing preference features; and based on the sound effect objective feature model, taking the target sound effect objective features as constraints, adopting a Monte Carlo algorithm to perform inversion solving in a parameter space of sound effect processing parameters to generate target sound effect processing parameters matched with the target sound effect objective features. The application has the effect of realizing precise configuration of personalized sound effects.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to methods, systems and devices for personalized sound effect configuration. Background Technology

[0002] In fields with extremely high requirements for sound reproduction accuracy and individual adaptability, conventional "standard presets" or tuning schemes geared towards the average person are completely inadequate. For example, when adjusting a monitoring system for a recording engineer, it is necessary not only to compensate for the frequency response differences caused by their individual auricle and ear canal, but also to accurately reproduce the subtle sensations they rely on in their work, such as "distinct instrument outlines" and "discernible spatial layers." In virtual auditory training applications, a three-dimensional sound field needs to be generated based on the spatial auditory characteristics of a specific operator, while simultaneously meeting the mandatory requirement that "a certain type of warning sound must not be masked by ambient sound under any circumstances." When customizing hearing aids for individuals with hearing loss, it is not enough to simply compensate based on the audiogram; their personal feelings, such as "they want conversations to sound clear but not harsh," must also be combined with the current noisy environment and translated into the device's processing parameters.

[0003] For the aforementioned problems, existing technologies mainly focus on the following approaches. One is to directly provide several preset modes for users to choose from. This method completely ignores individual auditory differences and subjective preferences; the same mode may mean completely different auditory biases for different users. Another approach is to perform fixed equalization compensation based on measured hearing curves. Although this can correct some objective hearing loss, it does not consider the user's subjective preference for sound style. For example, two people with the same hearing curve, one who prefers mellow high frequencies and the other who prefers bright high frequencies, will get the same result from this method. The third approach is to have users compare the played samples in pairs and then use optimization algorithms to iteratively search. However, this method can only explore a few low-dimensional parameters such as equalizer gain and cannot reach high-dimensional control dimensions such as filter structure, dynamic processor, or reverberation parameters. Furthermore, this method is essentially an unconstrained global preference search. When the acoustic requirements of the scene itself conflict with the user's preferences, the search results will inevitably favor one side at the expense of the other, failing to obtain a compromise solution that takes both into account.

[0004] The common drawback of the aforementioned technologies is that they have an extremely limited range of adjustable parameters and lack the ability to simultaneously handle multiple constraints from different levels. When the configuration target must simultaneously accommodate personal auditory preferences and the acoustic requirements of a specific scenario, a demanding intersection is formed in a continuous and highly dimensional parameter space. Current technology cannot perform direct inversion calculations from requirement description to parameter set within this space. As a result, such highly customized sound effect configurations can only rely on time-consuming item-by-item adjustments by professionals, leading to poor transferability of the solutions and a high dependence on the experience of the adjuster. Summary of the Invention

[0005] In order to achieve precise configuration of personalized sound effects while satisfying both personal auditory preferences and acoustic requirements of specific scenarios, this application provides a method, system, and device for configuring personalized sound effects.

[0006] Firstly, this application provides a method for configuring personalized sound effects, employing the following technical solution: Personalized sound effect configuration methods include: A hearing parameter model is constructed, wherein the hearing parameter model is trained based on a first sample data, the first sample data including sound effect objective feature samples, user hearing preference feature samples, and hearing parameter samples; Construct an objective feature model of sound effects, wherein the objective feature model of sound effects is trained based on second sample data, the second sample data including sound effect processing parameter samples and sound effect objective feature samples; Obtain the target user's preference feedback information for differentiated sound effect samples, and extract features from the preference feedback information to generate target auditory preference features; Obtain the target listening mode selected by the target user, and determine the target listening parameters based on the target listening mode; Based on the aforementioned auditory parameter model, and with the target auditory parameters as constraints, the gradient descent algorithm is used to perform inversion and solution in the parameter space of the objective features of sound effects, thereby determining the objective features of sound effects that match the target auditory preference features. Based on the objective feature model of the sound effect, and with the objective feature of the target sound effect as a constraint, the Monte Carlo algorithm is used to perform inversion and solution in the parameter space of the sound effect processing parameters to generate target sound effect processing parameters that match the objective feature of the target sound effect, and the sound effect is configured according to the target sound effect processing parameters.

[0007] By employing the above technical solution, a hearing parameter model is constructed. This model is trained using first sample data, which includes objective sound effect feature samples, user auditory preference feature samples, and hearing parameter samples. Then, an objective sound effect feature model is constructed, trained using second sample data, which includes sound effect processing parameter samples and objective sound effect feature samples. Next, the target user's preference feedback information for differentiated sound effect samples is obtained, and feature extraction is performed on this preference feedback information to generate target auditory preference features. The target user selects a target listening mode, and the target listening parameters are determined based on the target listening mode. Then, based on the listening parameter model and with the target listening parameters as constraints, the gradient descent algorithm is used to invert and solve the objective features of the sound effect in the parameter space to determine the objective features of the target sound effect that match the target listening preference features. Then, based on the objective feature model and with the objective features of the target sound effect as constraints, the Monte Carlo algorithm is used to invert and solve the objective features of the sound effect in the parameter space of the sound effect processing parameters to generate the target sound effect processing parameters that match the objective features of the target sound effect. Finally, the sound effect is configured according to the target sound effect processing parameters. This invention constructs a listening parameter model and an objective sound effect feature model, unifying and quantifying the user's subjective auditory preferences and the objective acoustic requirements of the scene into constraints that can be jointly computed. Then, using the target listening parameters as constraints, the gradient descent algorithm is used to invert and solve the target sound effect objective features in the sound effect objective feature space. Next, the Monte Carlo algorithm is used to invert from the target sound effect objective features to a high-dimensional sound effect processing parameter space. The two-stage inversion decomposes the complex nonlinear problem into subproblems that can be solved sequentially. Users only need to provide preference feedback and select a listening mode to automatically generate a personalized sound effect configuration that takes into account personal preferences and scene requirements and is precisely matched. This solves the problem that current technologies cannot handle multiple constraints simultaneously and are difficult to converge stably in high-dimensional parameter spaces, reducing the professional threshold and time cost of highly customized sound effect configurations.

[0008] Optionally, the step of constructing the auditory parameter model, wherein the auditory parameter model is trained based on first sample data, the first sample data including objective sound effect feature samples, user auditory preference feature samples, and auditory parameter samples, includes: Obtain the first sample data, wherein each sample in the first sample data includes objective sound effect features, user auditory preference features and corresponding auditory parameters; The first sample data is divided into a first training set and a first validation set; Construct a first model to be trained, wherein the first model to be trained takes the objective features of the sound effect and the user's auditory preference features as inputs and the auditory parameters as outputs; Based on the first training set, the first model to be trained is iteratively trained, and based on the first validation set, the model obtained in each iteration is evaluated. When the corresponding evaluation result meets the first preset training termination condition, training is stopped, and the corresponding model is used as the auditory parameter model.

[0009] By adopting the above technical solution, in order to construct the auditory parameter model, first sample data is obtained. Each sample in the first sample data includes objective sound effect features, user auditory preference features, and corresponding auditory parameters. Then, the first sample data is divided into a first training set and a first validation set. Then, a first model to be trained is constructed. The first model to be trained takes the objective sound effect features and user auditory preference features as input and the auditory parameters as output. Then, the first model to be trained is iteratively trained based on the first training set, and the model obtained in each iteration is evaluated based on the first validation set. When the corresponding evaluation result meets the first preset training termination condition, the training is stopped, and the corresponding model is used as the auditory parameter model.

[0010] Optionally, the step of constructing the objective feature model of sound effects, wherein the objective feature model of sound effects is obtained by training based on second sample data, the second sample data including sound effect processing parameter samples and objective feature samples of sound effects, includes: Obtain the second sample data, wherein each sample in the second sample data includes sound effect processing parameters and corresponding objective sound effect features; The second sample data is divided into a second training set and a second validation set; Construct a second model to be trained, wherein the second model to be trained takes the sound effect processing parameters as input and the objective features of the sound effect as output; Based on the second training set, the second model to be trained is iteratively trained, and based on the second validation set, the model obtained in each iteration is evaluated. When the corresponding evaluation result meets the second preset training termination condition, the training is stopped, and the corresponding model is used as the sound effect objective feature model.

[0011] By adopting the above technical solution, in order to construct the sound effect objective feature model, second sample data is obtained. Each sample in the second sample data includes sound effect processing parameters and corresponding sound effect objective features. Then, the second sample data is divided into a second training set and a second validation set. Then, a second model to be trained is constructed. The second model to be trained takes the sound effect processing parameters as input and the sound effect objective features as output. Then, the second model to be trained is iteratively trained based on the second training set, and the model obtained in each iteration is evaluated based on the second validation set. When the corresponding evaluation result meets the second preset training termination condition, the training is stopped, and the corresponding model is used as the sound effect objective feature model.

[0012] Optionally, the step of obtaining the target user's preference feedback information for differentiated sound effect samples, and extracting features from the preference feedback information to generate target auditory preference features includes: Obtain the target user's preference feedback information for multiple sets of differentiated sound effect samples, wherein the preference feedback information includes at least one of rating information, ranking information, and selection information; The preference feedback information is quantified to generate a preference quantification value; Extract auditory preference attributes from the preference quantification value, wherein the auditory preference attributes include at least one of loudness preference attribute, pitch preference attribute, spatial perception preference attribute, and timbre preference attribute; The auditory preference attributes are vectorized and encoded to generate the target auditory preference features of the target user.

[0013] By adopting the above technical solution, in order to generate target auditory preference features, the target user's preference feedback information on multiple sets of differentiated sound effect samples is obtained. The preference feedback information includes at least one of rating information, ranking information, and selection information. Then, the preference feedback information is quantified to generate preference quantification values. Then, auditory preference attributes are extracted from the preference quantification values. The auditory preference attributes include at least one of loudness preference attributes, pitch preference attributes, spatial perception preference attributes, and timbre preference attributes. Then, the auditory preference attributes are vectorized and encoded to generate the target user's target auditory preference features.

[0014] Optionally, the step of extracting the auditory preference attribute from the preference quantification value includes: Based on the acoustic properties of the differentiated sound effect samples, the preference quantization values ​​are grouped to obtain a subset of preference quantization values ​​for each acoustic property, wherein the acoustic properties include at least one of loudness, frequency, spatial and timbre attributes. Statistical analysis is performed on each subset of the aforementioned preference quantification values ​​to generate a preference intensity distribution along the acoustic attribute dimension; Based on the preference intensity distribution of each acoustic attribute dimension, the feature values ​​corresponding to each acoustic attribute dimension are determined to generate auditory preference attributes, wherein the feature values ​​include at least one of mean, median, peak value, or variance.

[0015] By adopting the above technical solution, in order to extract auditory preference attributes, the preference quantification values ​​are grouped based on the acoustic attributes of differentiated sound effect samples to obtain a subset of preference quantification values ​​for each acoustic attribute. The acoustic attributes include at least one of loudness, frequency, spatial and timbre attributes. Then, statistical analysis is performed on each subset of preference quantification values ​​to generate a preference intensity distribution for each acoustic attribute dimension. Based on the preference intensity distribution for each acoustic attribute dimension, the feature values ​​corresponding to each acoustic attribute dimension are determined to generate auditory preference attributes. The feature values ​​include at least one of mean, median, peak value or variance.

[0016] Optionally, the step of determining the target hearing parameters based on the target hearing mode includes: Obtain the target listening mode selected by the target user; Based on the target hearing pattern, a matching query is performed in a preset hearing pattern library to determine the corresponding target hearing parameters. The preset hearing pattern library includes preset hearing pattern data and preset hearing parameter data, and is used to characterize the mapping relationship between the preset hearing pattern data and the preset hearing parameter data.

[0017] By adopting the above technical solution, in order to generate target listening parameters, the target listening mode selected by the target user is obtained, and then a matching query is performed in the preset listening mode library according to the target listening mode to determine the corresponding target listening parameters. The preset listening mode library includes preset listening mode data and preset listening parameter data. The preset listening mode library is used to represent the mapping relationship between the preset listening mode data and the preset listening parameter data.

[0018] Optionally, the step of determining the target sound effect objective features that match the target auditory preference features by using the gradient descent algorithm to invert and solve the parameter space of the sound effect objective features based on the auditory parameter model, with the target auditory parameters as constraints, includes: Initialize the objective features of the sound effects in the parameter space of the objective features of the sound effects, and set the maximum number of iterations of gradient descent and the convergence tolerance error; The target auditory preference features are used as fixed inputs, and the target auditory preference features and the current objective sound effect features are input into the auditory parameter model to output the corresponding predicted auditory parameters. Construct a loss function, wherein the loss function consists of a difference measure term and a regularization term, the difference measure term is used to characterize the deviation between the predicted auditory parameters and the target auditory parameters, and the regularization term is used to constrain the amplitude of the current sound effect objective features; Calculate the gradient of the loss function with respect to the objective features of the sound effect; Along the opposite direction of the gradient, update the objective features of the sound effect according to a preset learning rate, and project the updated objective features of the sound effect into the parameter space of the objective features of the sound effect; If the current iteration count reaches the maximum iteration count, or the value of the loss function is continuously lower than the convergence tolerance error, then the current objective sound effect feature is taken as the target objective sound effect feature; otherwise, the iteration continues.

[0019] By adopting the above technical solution, in order to determine the objective features of the target sound effect, the objective features of the sound effect are initialized in the parameter space of the objective features of the sound effect, and the maximum number of iterations of gradient descent and the convergence tolerance error are set. Then, the target auditory preference features are used as fixed inputs, and the target auditory preference features and the current objective features of the sound effect are input into the auditory parameter model, and the corresponding predicted auditory parameters are output. Then, a loss function is constructed, which consists of a difference measure term and a regularization term. The difference measure term is used to characterize the deviation between the predicted auditory parameters and the target auditory parameters, and the regularization term is used to constrain the amplitude of the current objective features of the sound effect. Then, the gradient of the loss function with respect to the objective features of the sound effect is calculated. Then, the objective features of the sound effect are updated in the opposite direction of the gradient according to the preset learning rate, and the updated objective features of the sound effect are projected into the parameter space of the objective features of the sound effect. Then, if the current iteration number reaches the maximum iteration number, or the function value of the loss function is continuously lower than the convergence tolerance error, the current objective features of the sound effect are taken as the target objective features of the sound effect; otherwise, the iteration continues.

[0020] Optionally, the step of generating target sound effect processing parameters that match the target sound effect objective features by using the Monte Carlo algorithm to perform inversion and solution in the parameter space of the sound effect processing parameters based on the objective feature model of the sound effect and constrained by the objective features of the target sound effect includes: Define the sample size, sampling method, and proposed distribution for Monte Carlo sampling; Based on the number of samples and the sampling method, random sampling is performed within the parameter space of the sound effect processing parameters to generate an initial sound effect processing parameter sample set, which serves as the current parameter candidate set. Input each set of sound effect processing parameters in the current parameter candidate set into the sound effect objective feature model, and output the corresponding predicted sound effect objective features; Calculate the deviation between each of the predicted sound effect objective features and the target sound effect objective features, and construct an evaluation function based on the deviation, wherein the evaluation function is used to measure the likelihood of the predicted sound effect objective features reproducing the target sound effect objective features; New candidate parameters are extracted from the proposed distribution, the corresponding acceptance probability is calculated using the evaluation function, and the current parameter candidate set is updated based on the acceptance probability to generate a new parameter candidate set. Determine whether the iteration termination condition is met. If it is met, terminate the iteration. Otherwise, use the new parameter candidate set as the current parameter candidate set and continue iterating. The iteration termination condition includes the convergence of the evaluation function or reaching a preset number of samplings. After the iteration terminates, the sound effect processing parameter with the optimal function value of the evaluation function is selected from the current parameter candidate set as the target sound effect processing parameter.

[0021] By adopting the above technical solution, in order to generate target sound effect processing parameters, the number of samples, sampling method, and proposal distribution of Monte Carlo sampling are set. Then, based on the number of samples and sampling method, random sampling is performed within the parameter space of the sound effect processing parameters to generate an initial sound effect processing parameter sample set, which serves as the current parameter candidate set. Then, each set of sound effect processing parameters in the current parameter candidate set is input into the sound effect objective feature model, outputting the corresponding predicted sound effect objective features. Next, the deviation between each predicted sound effect objective feature and the target sound effect objective feature is calculated, and an evaluation function is constructed based on the deviation. The evaluation function is used to measure the predicted sound effect objective features. The likelihood of the objective features of the target sound effect is restored. Then, new candidate parameters are extracted from the proposal distribution. The corresponding acceptance probability is calculated using the evaluation function. The current parameter candidate set is updated according to the acceptance probability to generate a new parameter candidate set. Then, it is determined whether the iteration termination condition is met. If it is met, the iteration is terminated. Otherwise, the new parameter candidate set is used as the current parameter candidate set, and the iteration continues. The iteration termination condition includes the convergence of the evaluation function or reaching a preset number of samples. After the iteration terminates, the sound effect processing parameter with the optimal function value of the evaluation function is selected from the current parameter candidate set as the target sound effect processing parameter.

[0022] Secondly, this application also provides a personalized sound effect configuration system, which adopts the following technical solution: Personalized sound configuration system, including: The auditory parameter model construction module is used to construct an auditory parameter model, wherein the auditory parameter model is obtained by training based on first sample data, the first sample data including sound effect objective feature samples, user auditory preference feature samples, and auditory parameter samples; The sound effect objective feature model construction module is used to construct a sound effect objective feature model, wherein the sound effect objective feature model is obtained by training based on second sample data, the second sample data including sound effect processing parameter samples and sound effect objective feature samples; The preference feature extraction module is used to obtain the target user's preference feedback information on differentiated sound effect samples, and to extract features from the preference feedback information to generate target auditory preference features; The target hearing parameter determination module is used to obtain the target hearing mode selected by the target user and determine the target hearing parameters based on the target hearing mode; The first inversion module is used to perform inversion and solution in the parameter space of objective sound effect features based on the auditory parameter model, with the target auditory parameters as constraints, using the gradient descent algorithm to determine the target objective sound effect features that match the target auditory preference features. The second inversion module is used to perform inversion and solution in the parameter space of the sound effect processing parameters based on the objective feature model of the sound effect, with the objective feature of the target sound effect as a constraint, using the Monte Carlo algorithm to generate target sound effect processing parameters that match the objective feature of the target sound effect, and to configure the sound effect according to the target sound effect processing parameters.

[0023] Thirdly, this application also provides a computer device, which adopts the following technical solution: A computer device includes a memory and a processor, the memory storing a computer program executable on the processor, the processor executing the computer program to implement the method described in the first aspect.

[0024] In summary, this application includes at least the following beneficial technical effects: constructing a hearing parameter model, wherein the hearing parameter model is trained based on first sample data, the first sample data including objective sound effect feature samples, user auditory preference feature samples, and hearing parameter samples; then constructing an objective sound effect feature model, wherein the objective sound effect feature model is trained based on second sample data, the second sample data including sound effect processing parameter samples and objective sound effect feature samples; then obtaining the target user's preference feedback information for differentiated sound effect samples, and extracting features from the preference feedback information to generate target auditory preference features. Then, the target user's selected target listening mode is obtained, and the target listening parameters are determined based on the target listening mode. Then, based on the listening parameter model, with the target listening parameters as constraints, the gradient descent algorithm is used to perform inversion and solution in the parameter space of the objective sound effect features to determine the objective sound effect features that match the target listening preference features. Then, based on the objective sound effect feature model, with the objective sound effect features as constraints, the Monte Carlo algorithm is used to perform inversion and solution in the parameter space of the sound effect processing parameters to generate the target sound effect processing parameters that match the objective sound effect features. Finally, the sound effect is configured based on the objective sound effect processing parameters. This invention constructs a listening parameter model and an objective sound effect feature model, unifying and quantifying the user's subjective auditory preferences and the objective acoustic requirements of the scene into constraints that can be jointly computed. Then, using the target listening parameters as constraints, the gradient descent algorithm is used to invert and solve the target sound effect objective features in the sound effect objective feature space. Next, the Monte Carlo algorithm is used to invert from the target sound effect objective features to a high-dimensional sound effect processing parameter space. The two-stage inversion decomposes the complex nonlinear problem into subproblems that can be solved sequentially. Users only need to provide preference feedback and select a listening mode to automatically generate a personalized sound effect configuration that takes into account personal preferences and scene requirements and is precisely matched. This solves the problem that current technologies cannot handle multiple constraints simultaneously and are difficult to converge stably in high-dimensional parameter spaces, reducing the professional threshold and time cost of highly customized sound effect configurations. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of the overall process of an embodiment of this application.

[0026] Figure 2 This is a schematic diagram of the system structure of this application.

[0027] Figure 3 This is a structural block diagram of the computer device described in this application. Detailed Implementation

[0028] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0029] This application discloses a method for configuring personalized sound effects.

[0030] Reference Figure 1 Personalized sound effect configuration methods include: Step S11: Construct a listening parameter model, wherein the listening parameter model is trained based on the first sample data, which includes sound effect objective feature samples, user listening preference feature samples, and listening parameter samples.

[0031] It should be noted that in step S11, the auditory parameter model is pre-trained using the first sample data. Each sample in the first sample data includes objective sound effect features, user auditory preference features, and corresponding auditory parameters. The auditory parameter model learns how users with various auditory preferences will subjectively generate auditory parameters under various objective sound effect features. It can be understood as a predictive model from objective acoustic features and user preference features to subjective auditory parameters. The construction of this model allows subjective auditory perception to be quantified into calculable constraints, providing a model foundation for subsequent auditory goal-oriented inversion.

[0032] Step S12: Construct an objective feature model of sound effects. The objective feature model of sound effects is obtained by training based on the second sample data, which includes sound effect processing parameter samples and objective feature samples of sound effects.

[0033] It should be noted that in step S12, the sound effect objective feature model is pre-trained using the second sample data. Each set of samples in the second sample data contains sound effect processing parameters and corresponding sound effect objective features. The sound effect objective feature model learns the mapping relationship between sound effect processing parameters and sound effect objective features. Sound effect processing parameters refer to the underlying parameters that can be directly applied to the audio signal processing link, such as the gain of each frequency point of the equalizer, the cutoff frequency and quality factor of the filter, the threshold and ratio of the compressor, and the pre-delay and diffusion of the reverb. Sound effect objective features are quantitative indicators that can be extracted from the processed audio signal and describe its overall acoustic characteristics.

[0034] Step S13: Obtain the target user's preference feedback information for differentiated sound effect samples, and extract features from the preference feedback information to generate target auditory preference features.

[0035] It should be noted that, in the specific implementation of step S13, multiple sound effect samples with obvious differences in auditory perception can be played to the target user, and their feedback information can be collected. The feedback can take the form of rating, pairwise comparison selection, preference ranking, etc. Then, feature extraction is performed on the collected preference feedback information, and it is transformed into a structured target auditory preference feature vector. This feature vector is a numerical description of the user's personal auditory preference. It will be used as one of the input conditions for subsequent inversion solution to ensure that the final generated sound effect configuration matches the user's subjective preferences, rather than a general template.

[0036] Step S14: Obtain the target listening mode selected by the target user, and determine the target listening parameters based on the target listening mode.

[0037] It should be noted that step S14 introduces scenario-based acoustic requirements. The target listening mode can be explicitly selected by the user, such as "concert hall," "audio podcast," "nighttime movie viewing," and "competitive game." Each listening mode corresponds to a set of preset target listening parameters. These parameters define the subjective listening indicators expected to be achieved in that scenario, such as speech intelligibility needing to reach a certain threshold, sound field width needing to be within a certain range, and low-frequency masking effect needing to be below a certain limit. The target listening parameters do not depend on user preferences but on the objective requirements of the scenario itself for sound reproduction. This step determines the hard constraints that must be met in the inversion solution.

[0038] Step S15: Based on the auditory parameter model, with the target auditory parameters as constraints, the gradient descent algorithm is used to perform inversion and solution in the parameter space of the objective features of the sound effect, and the objective features of the target sound effect that match the target auditory preference features are determined.

[0039] It should be noted that after obtaining the target auditory preference features and target auditory parameters, it is necessary to find an optimal point in the parameter space of the objective features of the sound effect. The solution process is based on the auditory parameter model, with the target auditory parameters as the constraint boundary that must be satisfied. The consistency between the target auditory preference features and the model prediction results is used as the optimization objective. The gradient descent algorithm is used to iteratively update the objective features of the sound effect along the gradient direction until convergence. The final objective features of the target sound effect are a solution at the acoustic feature level that simultaneously meets the two conditions of "what the user likes" and "what the scene requires".

[0040] Step S16: Based on the objective feature model of sound effects, and with the objective features of the target sound effects as constraints, the Monte Carlo algorithm is used to perform inversion and solution in the parameter space of the sound effect processing parameters to generate target sound effect processing parameters that match the objective features of the target sound effects, and sound effect configuration is performed according to the target sound effect processing parameters.

[0041] It should be noted that, due to the high dimensionality and multi-peak, non-convex characteristics of the audio effect processing parameter space, gradient-based methods are prone to getting trapped in local optima. Therefore, the Monte Carlo algorithm is used for solving the problem. Specifically, a large number of random samples are taken within the parameter space of the audio effect processing parameters. Each set of sampled parameters is input into the audio effect objective feature model, and the deviation between the output audio effect objective features and the target audio effect objective features determined in step S15 is calculated. Based on the magnitude of the deviation, the sampled parameters are filtered or weighted to gradually approximate the parameter combination that satisfies the constraints. The final target audio effect processing parameters generated are the final results that can be directly configured into the audio processing link, thereby completing the personalized audio effect configuration for the target user.

[0042] In the above implementation, a hearing parameter model is constructed, wherein the hearing parameter model is trained based on a first sample data, the first sample data including sound effect objective feature samples, user auditory preference feature samples, and hearing parameter samples. Then, a sound effect objective feature model is constructed, wherein the sound effect objective feature model is trained based on a second sample data, the second sample data including sound effect processing parameter samples and sound effect objective feature samples. Then, the target user's preference feedback information for differentiated sound effect samples is obtained, and feature extraction is performed on the preference feedback information to generate target auditory preference features. Finally, the target... The system identifies the user's selected target listening mode and determines the target listening parameters based on that mode. Then, based on the listening parameter model and constrained by the target listening parameters, it uses a gradient descent algorithm to invert and solve the objective features of the sound effects in the parameter space. This determines the objective features of the sound effects that match the target auditory preference features. Finally, based on the objective features of the sound effects model and constrained by the objective features of the sound effects, it uses a Monte Carlo algorithm to invert and solve the objective features of the sound effects in the parameter space. This generates the objective sound effect processing parameters that match the objective features of the sound effects, and the sound effects are configured based on these parameters. This invention constructs a listening parameter model and an objective sound effect feature model, unifying and quantifying the user's subjective auditory preferences and the objective acoustic requirements of the scene into constraints that can be jointly computed. Then, using the target listening parameters as constraints, the gradient descent algorithm is used to invert and solve the target sound effect objective features in the sound effect objective feature space. Next, the Monte Carlo algorithm is used to invert from the target sound effect objective features to a high-dimensional sound effect processing parameter space. The two-stage inversion decomposes the complex nonlinear problem into subproblems that can be solved sequentially. Users only need to provide preference feedback and select a listening mode to automatically generate a personalized sound effect configuration that takes into account personal preferences and scene requirements and is precisely matched. This solves the problem that current technologies cannot handle multiple constraints simultaneously and are difficult to converge stably in high-dimensional parameter spaces, reducing the professional threshold and time cost of highly customized sound effect configurations.

[0043] As a further implementation of the method, a listening parameter model is constructed, wherein the listening parameter model is trained based on a first sample data, the first sample data including sound effect objective feature samples, user auditory preference feature samples, and listening parameter samples, the steps of which include: Step S21: Obtain the first sample data, wherein each sample in the first sample data includes objective sound effect features, user auditory preference features, and corresponding auditory parameters.

[0044] Step S22: Divide the first sample data into a first training set and a first validation set.

[0045] Step S23: Construct the first model to be trained, wherein the first model to be trained takes the objective features of sound effects and the user's auditory preference features as inputs and the auditory parameters as outputs.

[0046] Step S24: Based on the first training set, iteratively train the first model to be trained, and based on the first validation set, evaluate the model obtained in each iteration. When the corresponding evaluation result meets the first preset training termination condition, stop training and use the corresponding model as the auditory parameter model.

[0047] In the above implementation, in order to construct the auditory parameter model, first sample data is obtained, wherein each sample in the first sample data includes objective sound effect features, user auditory preference features, and corresponding auditory parameters. Then, the first sample data is divided into a first training set and a first validation set. Then, a first model to be trained is constructed, wherein the first model to be trained takes the objective sound effect features and user auditory preference features as input and the auditory parameters as output. Then, the first model to be trained is iteratively trained based on the first training set, and the model obtained in each iteration is evaluated based on the first validation set. When the corresponding evaluation result meets the first preset training termination condition, the training is stopped, and the corresponding model is used as the auditory parameter model.

[0048] As a further implementation of the method, a sound effect objective feature model is constructed, wherein the sound effect objective feature model is trained based on second sample data, the second sample data including sound effect processing parameter samples, and the step of sound effect objective feature samples includes: Step S31: Obtain the second sample data, wherein each sample in the second sample data includes sound effect processing parameters and corresponding objective sound effect features.

[0049] Step S32: Divide the second sample data into a second training set and a second validation set.

[0050] Step S33: Construct a second model to be trained, wherein the second model to be trained takes sound effect processing parameters as input and sound effect objective features as output.

[0051] Step S34: Based on the second training set, iteratively train the second model to be trained, and based on the second validation set, evaluate the model obtained in each iteration. When the corresponding evaluation result meets the second preset training termination condition, stop training and use the corresponding model as the sound effect objective feature model.

[0052] In the above implementation, in order to construct the sound effect objective feature model, second sample data is obtained, wherein each sample in the second sample data includes sound effect processing parameters and corresponding sound effect objective features. Then, the second sample data is divided into a second training set and a second validation set. Then, a second model to be trained is constructed, wherein the second model to be trained takes the sound effect processing parameters as input and the sound effect objective features as output. Then, based on the second training set, the second model to be trained is iteratively trained, and based on the second validation set, the model obtained in each iteration is evaluated. When the corresponding evaluation result meets the second preset training termination condition, the training is stopped, and the corresponding model is used as the sound effect objective feature model.

[0053] As a further implementation of the method, the steps of obtaining the target user's preference feedback information on differentiated sound effect samples, and extracting features from the preference feedback information to generate target auditory preference features include: Step S41: Obtain the target user's preference feedback information for multiple sets of differentiated sound effect samples, wherein the preference feedback information includes at least one of rating information, ranking information and selection information.

[0054] Step S42: Quantify the preference feedback information to generate a preference quantification value.

[0055] It's important to note that user-provided preference feedback takes many forms. For example, ratings might be discrete scores like "5 points" or "4 points," rankings might be sequential relationships like "sample A is better than sample B is better than sample C," and choices might be binary results like "choose A between A and B." This raw feedback cannot be directly used for feature extraction and subsequent model calculations; it requires quantification to convert it into comparable and computable preference quantification values. For ratings, the scores can be used directly or normalized. For rankings, ordinal values ​​can be assigned based on the ranking position or latent scores can be estimated using comparison models like the Bradley-Terry model. For choices, the frequency of winning can be converted into preference probabilities or preference strength values. After quantification, all feedback is mapped to the same numerical scale, providing standardized input for extracting auditory preference attributes.

[0056] Step S43: Extract auditory preference attributes from the preference quantification value, wherein the auditory preference attributes include at least one of loudness preference attributes, pitch preference attributes, spatial preference attributes, and timbre preference attributes.

[0057] It's important to note that the preference quantification value reflects a user's overall liking for a complete sound effect sample. However, this overall evaluation value implicitly includes the user's tendencies across multiple auditory dimensions. For example, a user might give a sample a high score because of its strong bass response or its wide soundstage. The process of extracting auditory preference attributes involves decomposing the overall preference quantification value into multiple independent auditory perception dimensions, including but not limited to loudness preference, pitch preference, spatial preference, and timbre preference. In practice, based on the known objective parameters of differentiated sound effect samples across each auditory dimension, combined with the user's preference quantification value for each sample, the direction and intensity of the user's tendency in each auditory dimension can be calculated. This process decomposes the user's vague overall preference into a structured, multi-dimensional auditory preference profile, providing quantifiable conditions for subsequently generating objective features of target sound effects that precisely match the user's auditory preferences.

[0058] Step S44: Vectorize the auditory preference attributes to generate the target auditory preference features of the target user.

[0059] In the above implementation, in order to generate target auditory preference features, the target user's preference feedback information on multiple sets of differentiated sound effect samples is obtained. The preference feedback information includes at least one of rating information, ranking information, and selection information. Then, the preference feedback information is quantified to generate preference quantification values. Then, auditory preference attributes are extracted from the preference quantification values. The auditory preference attributes include at least one of loudness preference attributes, pitch preference attributes, spatial perception preference attributes, and timbre preference attributes. Then, the auditory preference attributes are vectorized and encoded to generate the target user's target auditory preference features.

[0060] As a further implementation of the method, the step of extracting auditory preference attributes from the preference quantification value includes: Step S51: Based on the acoustic attributes of the differentiated sound effect samples, the preference quantization values ​​are grouped to obtain a subset of preference quantization values ​​for each acoustic attribute. The acoustic attributes include at least one of loudness attribute, frequency attribute, spatial attribute, and timbre attribute.

[0061] Step S52: Perform statistical analysis on each subset of preference quantification values ​​to generate the preference intensity distribution of the acoustic attribute dimension.

[0062] Step S53: Based on the preference intensity distribution of each acoustic attribute dimension, determine the feature value corresponding to each acoustic attribute dimension to generate auditory preference attributes, wherein the feature value includes at least one of mean, median, peak value or variance.

[0063] In the above implementation, in order to extract auditory preference attributes, the preference quantification values ​​are grouped based on the acoustic attributes of the differentiated sound effect samples to obtain a subset of preference quantification values ​​for each acoustic attribute. The acoustic attributes include at least one of loudness, frequency, spatial and timbre attributes. Then, statistical analysis is performed on each subset of preference quantification values ​​to generate a preference intensity distribution for each acoustic attribute dimension. Based on the preference intensity distribution for each acoustic attribute dimension, the feature values ​​corresponding to each acoustic attribute dimension are determined to generate auditory preference attributes. The feature values ​​include at least one of mean, median, peak value or variance.

[0064] As a further implementation of the method, the step of determining the target hearing parameters based on the target hearing pattern includes: Step S61: Obtain the target listening mode selected by the target user.

[0065] Step S62: Based on the target hearing mode, perform a matching query in the preset hearing mode library to determine the corresponding target hearing parameters. The preset hearing mode library includes preset hearing mode data and preset hearing parameter data. The preset hearing mode library is used to represent the mapping relationship between the preset hearing mode data and the preset hearing parameter data.

[0066] In the above implementation, in order to generate target hearing parameters, the target hearing mode selected by the target user is obtained, and then a matching query is performed in the preset hearing mode library according to the target hearing mode to determine the corresponding target hearing parameters. The preset hearing mode library includes preset hearing mode data and preset hearing parameter data. The preset hearing mode library is used to represent the mapping relationship between the preset hearing mode data and the preset hearing parameter data.

[0067] As a further implementation of the method, based on the auditory parameter model, and constrained by the target auditory parameters, the gradient descent algorithm is used to inversely solve the objective features of the sound effect in the parameter space of the objective features of the sound effect, and the step of determining the objective features of the target sound effect that match the target auditory preference features includes: Step S71: Initialize the objective features of the sound effects in the parameter space of the objective features of the sound effects, and set the maximum number of iterations of gradient descent and the convergence tolerance error.

[0068] It should be noted that the parameter space of the objective sound effect features is determined as follows: this parameter space is spanned by the reasonable value ranges of each dimension included in the objective sound effect features. Specifically, the value range of each dimension can be determined based on the statistical distribution of the objective sound effect feature samples in each dimension in the first sample data. For example, the minimum and maximum values ​​in the samples can be used as boundaries, or hard upper and lower limits can be set according to the physical constraints of the actual acoustic system (such as reverberation time not being negative, frequency response fluctuation amplitude being limited to the achievable range of the filter, etc.). Thus, each dimension corresponds to a continuous numerical range, and the value ranges of multiple dimensions combine to form a bounded high-dimensional continuous space, that is, the parameter space of the objective sound effect features. The gradient descent algorithm will search for the optimal solution in this space, and the projection operation in step S75 also uses the boundary of this space as the clipping basis to ensure that the objective sound effect features generated during the iteration process are always within the range of the actually achievable acoustic state.

[0069] Step S72: The target auditory preference features are used as fixed inputs, and the target auditory preference features and the current objective sound effect features are input into the auditory parameter model, and the corresponding predicted auditory parameters are output.

[0070] It's important to note that during the inversion process, the target auditory preference feature, serving as a fixed input, is a feature vector extracted and encoded from the target user's preference feedback information. It represents the user's personal preferences across multiple auditory dimensions, including loudness, pitch, spatial awareness, and timbre. For this configuration task, the user is fixed; therefore, this feature vector remains unchanged throughout the entire inversion iteration, does not participate in gradient updates, and does not change with the number of iterations. In contrast, the objective sound effect features are the variables to be optimized, and their values ​​are updated along the gradient direction in each iteration. Fixing the target auditory preference feature means that the auditory parameter model uses "the preference of this specific user" as one of the constant conditions in each forward inference, approximating the target auditory parameters only by adjusting the objective sound effect features, thus ensuring that the final solution always matches the target user's personal auditory preferences.

[0071] Step S73: Construct a loss function, which consists of a difference measure and a regularization term. The difference measure is used to characterize the deviation between the predicted auditory parameters and the target auditory parameters, and the regularization term is used to constrain the amplitude of the current objective features of the sound effect.

[0072] Step S74: Calculate the gradient of the loss function with respect to the objective features of the sound effect.

[0073] It should be noted that step S74 uses automatic or numerical differentiation methods to calculate the partial derivatives of the loss function with respect to each dimension of the sound effect objective features at the current point, obtaining the gradient vector. The gradient vector indicates the direction in which the loss function increases most rapidly in the parameter space, that is, in which direction and by what proportion should each dimension of the sound effect objective features change to cause the loss function to increase most rapidly. Since subsequent steps need to update parameters in the opposite direction of the gradient to reduce the loss, the accuracy of the gradient calculation directly affects the convergence efficiency and the quality of the final solution.

[0074] Step S75: Update the objective features of the sound effect along the opposite direction of the gradient according to the preset learning rate, and project the updated objective features of the sound effect into the parameter space of the objective features of the sound effect.

[0075] It should be noted that moving in the opposite direction of the gradient means adjusting the objective features of the sound effects in the direction that reduces the loss function, and the learning rate controls the step size of each move. The updated objective features of the sound effects may have some dimensions that exceed their reasonable value range. The projection operation clips or scales them back to the effective range of the parameter space dimension by dimension, ensuring that the objective features of the sound effects obtained after each iteration always correspond to the physically achievable acoustic state, and avoiding the generation of meaningless or unreachable feature combinations.

[0076] Step S76: If the current iteration count reaches the maximum iteration count, or the function value of the loss function is continuously lower than the convergence tolerance error, then the current sound effect objective feature is taken as the target sound effect objective feature; otherwise, continue iterating.

[0077] In the above implementation, to determine the objective features of the target sound effect, the objective features of the sound effect are initialized in the parameter space of the objective features of the sound effect, and the maximum number of iterations of gradient descent and the convergence tolerance error are set. Then, the target auditory preference features are used as fixed inputs, and the target auditory preference features and the current objective features of the sound effect are input into the auditory parameter model to output the corresponding predicted auditory parameters. Then, a loss function is constructed, which consists of a difference metric term and a regularization term. The difference metric term is used to characterize the deviation between the predicted auditory parameters and the target auditory parameters, and the regularization term is used to constrain the amplitude of the current objective features of the sound effect. Then, the gradient of the loss function with respect to the objective features of the sound effect is calculated. Then, the objective features of the sound effect are updated in the opposite direction of the gradient according to the preset learning rate, and the updated objective features of the sound effect are projected into the parameter space of the objective features of the sound effect. Then, if the current number of iterations reaches the maximum number of iterations, or the function value of the loss function is continuously lower than the convergence tolerance error, the current objective features of the sound effect are taken as the target objective features of the sound effect; otherwise, the iteration continues.

[0078] As a further implementation of the method, based on the objective feature model of sound effects, and constrained by the objective features of the target sound effects, the Monte Carlo algorithm is used to perform inversion and solution in the parameter space of the sound effect processing parameters to generate target sound effect processing parameters that match the objective features of the target sound effects. This step includes: Step S81: Set the number of samples, sampling method, and proposed distribution for Monte Carlo sampling.

[0079] It should be noted that step S81 sets the basic operating parameters for the Monte Carlo inversion solution. The sample size determines the scale of random sampling in the parameter space in each iteration. Its setting must balance search sufficiency and computational cost. Too few samples may cause the search to miss high-quality solution regions, while too many samples will increase the computational burden of a single round. It can usually be determined comprehensively based on the dimensionality of the parameter space and the target solution accuracy. The sampling method specifies how to distribute points in the parameter space. Uniform sampling can be used to fully cover the space, or sampling of key areas based on prior knowledge can be used to improve search efficiency. The proposal distribution determines how to generate new candidate solutions from the current candidate solutions during the iteration process. Usually, a multivariate Gaussian distribution or Cauchy distribution centered on the current candidate solution can be used. The variance parameter of the distribution controls the search step size: a larger variance indicates a stronger global exploration capability, while a smaller variance indicates a better local fine-grained search capability. It can be set specifically according to the actual characteristics of the parameter space.

[0080] Step S82: Based on the number of samples and the sampling method, randomly sample within the parameter space of the sound effect processing parameters to generate an initial sound effect processing parameter sample set as the current parameter candidate set.

[0081] Step S83: Input the sound effect processing parameters of each group in the current parameter candidate set into the sound effect objective feature model, and output the corresponding predicted sound effect objective features.

[0082] Step S84: Calculate the deviation between each predicted sound effect objective feature and the target sound effect objective feature, and construct an evaluation function based on the deviation. The evaluation function is used to measure the likelihood of the predicted sound effect objective features reproducing the target sound effect objective features.

[0083] Step S85: Extract new candidate parameters from the proposal distribution, calculate the corresponding acceptance probability using the evaluation function, and update the current parameter candidate set based on the acceptance probability to generate a new parameter candidate set.

[0084] It should be noted that step S85 performs candidate solution generation and acceptance / rejection determination in Monte Carlo sampling. Specifically, taking a candidate parameter in the current parameter candidate set as a benchmark, random perturbation sampling is performed according to the proposal distribution set in step S81 to generate a new candidate parameter. Then, the new candidate parameter is input into the sound effect objective feature model to obtain its predicted sound effect objective features. The evaluation function values ​​of the new candidate parameter and the current candidate parameter are calculated separately using the evaluation function. The acceptance probability is determined based on the ratio of the two evaluation function values. If the evaluation function value of the new candidate parameter is better, it indicates that its corresponding sound effect processing parameter can more accurately restore the target sound effect objective features. In this case, the new candidate parameter is accepted with a higher probability, or even a probability of 1, replacing the old candidate parameter with a poor evaluation in the candidate set. If the evaluation function value of the new candidate parameter is worse, it is accepted with a certain probability, which is determined by the specific value of the ratio of the evaluation function values. This probabilistic acceptance mechanism enables the search process to escape the local optimum region, avoiding the algorithm from converging to a suboptimal solution too early, while the overall trend still converges towards the direction of better evaluation function values. After this round of updates, a new set of parameter candidates is formed as the basis for the next round of iterations.

[0085] Step S86: Determine whether the iteration termination condition is met. If it is met, terminate the iteration. Otherwise, use the new parameter candidate set as the current parameter candidate set and continue iterating. The iteration termination condition includes the convergence of the evaluation function or reaching the preset number of samplings.

[0086] Step S87: After the iteration terminates, select the sound effect processing parameter with the optimal function value of the evaluation function from the current parameter candidate set as the target sound effect processing parameter.

[0087] In the above implementation, to generate target sound effect processing parameters, the number of samples, sampling method, and proposal distribution for Monte Carlo sampling are set. Then, based on the number of samples and sampling method, random sampling is performed within the parameter space of the sound effect processing parameters to generate an initial sound effect processing parameter sample set, which serves as the current parameter candidate set. Next, each set of sound effect processing parameters in the current parameter candidate set is input into the sound effect objective feature model, outputting the corresponding predicted sound effect objective features. Then, the deviation between each predicted sound effect objective feature and the target sound effect objective feature is calculated, and an evaluation function is constructed based on the deviation. The evaluation function is used to measure the predicted sound effect objective features. The likelihood of the objective features of the target sound effect is characterized. Then, new candidate parameters are extracted from the proposal distribution. The corresponding acceptance probability is calculated using the evaluation function. The current parameter candidate set is updated according to the acceptance probability to generate a new parameter candidate set. Then, it is determined whether the iteration termination condition is met. If it is met, the iteration is terminated. Otherwise, the new parameter candidate set is used as the current parameter candidate set, and the iteration continues. The iteration termination condition includes the convergence of the evaluation function or reaching a preset number of samples. After the iteration terminates, the sound effect processing parameter with the optimal function value of the evaluation function is selected from the current parameter candidate set as the target sound effect processing parameter.

[0088] This application also discloses a personalized sound effect configuration system.

[0089] refer to Figure 2 Personalized sound effect configuration system, including: The auditory parameter model construction module is used to construct the auditory parameter model, which is trained based on the first sample data. The first sample data includes objective sound effect feature samples, user auditory preference feature samples, and auditory parameter samples. The sound effect objective feature model construction module is used to construct the sound effect objective feature model. The sound effect objective feature model is obtained by training based on the second sample data, which includes sound effect processing parameter samples and sound effect objective feature samples. The preference feature extraction module is used to obtain the target user's preference feedback information on differentiated sound effect samples, and to extract features from the preference feedback information to generate target auditory preference features; The target hearing parameter determination module is used to obtain the target hearing mode selected by the target user and determine the target hearing parameters based on the target hearing mode; The first inversion module is used to solve the objective sound effect features that match the target auditory preference features by using the gradient descent algorithm to invert the model based on the auditory parameter model and with the target auditory parameters as constraints. The second inversion module is used to perform inversion and solution in the parameter space of the sound effect processing parameters based on the objective feature model of the sound effect, with the objective feature of the target sound effect as a constraint, using the Monte Carlo algorithm to generate target sound effect processing parameters that match the objective feature of the target sound effect, and to configure the sound effect according to the target sound effect processing parameters.

[0090] The personalized sound effect configuration system of the present invention can implement any of the methods in the personalized sound effect configuration method, and the specific working process of the personalized sound effect configuration system of the present invention can refer to the corresponding process in the personalized sound effect configuration method described above.

[0091] This application also discloses a computer device.

[0092] refer to Figure 3 A computer device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement any of the above-described methods for personalized sound effect configuration.

[0093] The above are all preferred embodiments of this application and are not intended to limit the scope of protection of this application. Any feature disclosed in this specification (including the abstract and drawings) may be replaced by other equivalent or similar features unless specifically stated otherwise. That is, unless specifically stated otherwise, each feature is only one example of a series of equivalent or similar features.

Claims

1. A method for configuring personalized sound effects, characterized in that, include: A hearing parameter model is constructed, wherein the hearing parameter model is trained based on a first sample data, the first sample data including sound effect objective feature samples, user hearing preference feature samples, and hearing parameter samples; Construct an objective feature model of sound effects, wherein the objective feature model of sound effects is trained based on second sample data, the second sample data including sound effect processing parameter samples and sound effect objective feature samples; Obtain the target user's preference feedback information for differentiated sound effect samples, and extract features from the preference feedback information to generate target auditory preference features; Obtain the target listening mode selected by the target user, and determine the target listening parameters based on the target listening mode; Based on the aforementioned auditory parameter model, and with the target auditory parameters as constraints, the gradient descent algorithm is used to perform inversion and solution in the parameter space of the objective features of sound effects, thereby determining the objective features of sound effects that match the target auditory preference features. Based on the objective feature model of the sound effect, and with the objective feature of the target sound effect as a constraint, the Monte Carlo algorithm is used to perform inversion and solution in the parameter space of the sound effect processing parameters to generate target sound effect processing parameters that match the objective feature of the target sound effect, and the sound effect is configured according to the target sound effect processing parameters.

2. The personalized sound effect configuration method according to claim 1, characterized in that, The step of constructing the auditory parameter model, wherein the auditory parameter model is trained based on first sample data, the first sample data including objective sound effect feature samples, user auditory preference feature samples, and auditory parameter samples, includes: Obtain the first sample data, wherein each sample in the first sample data includes objective sound effect features, user auditory preference features and corresponding auditory parameters; The first sample data is divided into a first training set and a first validation set; Construct a first model to be trained, wherein the first model to be trained takes the objective features of the sound effect and the user's auditory preference features as inputs and the auditory parameters as outputs; Based on the first training set, the first model to be trained is iteratively trained, and based on the first validation set, the model obtained in each iteration is evaluated. When the corresponding evaluation result meets the first preset training termination condition, training is stopped, and the corresponding model is used as the auditory parameter model.

3. The personalized sound effect configuration method according to claim 1, characterized in that, The step of constructing an objective sound effect feature model, wherein the objective sound effect feature model is trained based on second sample data, the second sample data including sound effect processing parameter samples and objective sound effect feature samples, includes: Obtain the second sample data, wherein each sample in the second sample data includes sound effect processing parameters and corresponding objective sound effect features; The second sample data is divided into a second training set and a second validation set; Construct a second model to be trained, wherein the second model to be trained takes the sound effect processing parameters as input and the objective features of the sound effect as output; Based on the second training set, the second model to be trained is iteratively trained, and based on the second validation set, the model obtained in each iteration is evaluated. When the corresponding evaluation result meets the second preset training termination condition, the training is stopped, and the corresponding model is used as the sound effect objective feature model.

4. The personalized sound effect configuration method according to claim 1, characterized in that, The step of obtaining the target user's preference feedback information for differentiated sound effect samples, and extracting features from the preference feedback information to generate target auditory preference features includes: Obtain the target user's preference feedback information for multiple sets of differentiated sound effect samples, wherein the preference feedback information includes at least one of rating information, ranking information, and selection information; The preference feedback information is quantified to generate a preference quantification value; Extract auditory preference attributes from the preference quantification value, wherein the auditory preference attributes include at least one of loudness preference attribute, pitch preference attribute, spatial perception preference attribute, and timbre preference attribute; The auditory preference attributes are vectorized and encoded to generate the target auditory preference features of the target user.

5. The personalized sound effect configuration method according to claim 4, characterized in that, The step of extracting auditory preference attributes from the preference quantification value includes: Based on the acoustic properties of the differentiated sound effect samples, the preference quantization values ​​are grouped to obtain a subset of preference quantization values ​​for each acoustic property, wherein the acoustic properties include at least one of loudness, frequency, spatial and timbre attributes. Statistical analysis is performed on each subset of the aforementioned preference quantification values ​​to generate a preference intensity distribution along the acoustic attribute dimension; Based on the preference intensity distribution of each acoustic attribute dimension, the feature values ​​corresponding to each acoustic attribute dimension are determined to generate auditory preference attributes, wherein the feature values ​​include at least one of mean, median, peak value, or variance.

6. The personalized sound effect configuration method according to claim 1, characterized in that, The step of determining the target hearing parameters based on the target hearing mode includes: Obtain the target listening mode selected by the target user; Based on the target listening mode, a matching query is performed in a preset listening mode library to determine the corresponding target listening parameters. The preset listening mode library includes preset listening mode data and preset listening parameter data. The preset listening mode library is used to characterize the mapping relationship between the preset listening mode data and the preset listening parameter data.

7. The personalized sound effect configuration method according to claim 1, characterized in that, The step of determining the target sound effect objective features that match the target auditory preference features by using the gradient descent algorithm to invert and solve the problem in the parameter space of the objective sound effect features based on the auditory parameter model, with the target auditory parameters as constraints, includes: Initialize the objective features of the sound effects in the parameter space of the objective features of the sound effects, and set the maximum number of iterations of gradient descent and the convergence tolerance error; The target auditory preference features are used as fixed inputs, and the target auditory preference features and the current objective sound effect features are input into the auditory parameter model to output the corresponding predicted auditory parameters. Construct a loss function, wherein the loss function consists of a difference measure term and a regularization term, the difference measure term is used to characterize the deviation between the predicted auditory parameters and the target auditory parameters, and the regularization term is used to constrain the amplitude of the current sound effect objective features; Calculate the gradient of the loss function with respect to the objective features of the sound effect; Along the opposite direction of the gradient, update the objective features of the sound effect according to a preset learning rate, and project the updated objective features of the sound effect into the parameter space of the objective features of the sound effect; If the current iteration count reaches the maximum iteration count, or the value of the loss function is continuously lower than the convergence tolerance error, then the current objective sound effect feature is taken as the target objective sound effect feature; otherwise, the iteration continues.

8. The personalized sound effect configuration method according to claim 1, characterized in that, The step of generating target sound effect processing parameters that match the target sound effect objective features by using the Monte Carlo algorithm to invert and solve the problem in the parameter space of the sound effect processing parameters based on the objective feature model of the sound effect, with the objective features of the target sound effect as constraints, includes: Define the sample size, sampling method, and proposed distribution for Monte Carlo sampling; Based on the number of samples and the sampling method, random sampling is performed within the parameter space of the sound effect processing parameters to generate an initial sound effect processing parameter sample set, which serves as the current parameter candidate set. Input each set of sound effect processing parameters in the current parameter candidate set into the sound effect objective feature model, and output the corresponding predicted sound effect objective features; Calculate the deviation between each of the predicted sound effect objective features and the target sound effect objective features, and construct an evaluation function based on the deviation, wherein the evaluation function is used to measure the likelihood of the predicted sound effect objective features reproducing the target sound effect objective features; New candidate parameters are extracted from the proposed distribution, the corresponding acceptance probability is calculated using the evaluation function, and the current parameter candidate set is updated based on the acceptance probability to generate a new parameter candidate set. Determine whether the iteration termination condition is met. If it is met, terminate the iteration. Otherwise, use the new parameter candidate set as the current parameter candidate set and continue iterating. The iteration termination condition includes the convergence of the evaluation function or reaching a preset number of samplings. After the iteration terminates, the sound effect processing parameter with the optimal function value of the evaluation function is selected from the current parameter candidate set as the target sound effect processing parameter.

9. A personalized sound effect configuration system, characterized in that, include: The auditory parameter model construction module is used to construct an auditory parameter model, wherein the auditory parameter model is obtained by training based on first sample data, the first sample data including sound effect objective feature samples, user auditory preference feature samples, and auditory parameter samples; The sound effect objective feature model construction module is used to construct a sound effect objective feature model, wherein the sound effect objective feature model is obtained by training based on second sample data, the second sample data including sound effect processing parameter samples and sound effect objective feature samples; The preference feature extraction module is used to obtain the target user's preference feedback information on differentiated sound effect samples, and to extract features from the preference feedback information to generate target auditory preference features; The target hearing parameter determination module is used to obtain the target hearing mode selected by the target user and determine the target hearing parameters based on the target hearing mode; The first inversion module is used to perform inversion and solution in the parameter space of objective sound effect features based on the auditory parameter model, with the target auditory parameters as constraints, using the gradient descent algorithm to determine the target objective sound effect features that match the target auditory preference features; The second inversion module is used to perform inversion and solution in the parameter space of the sound effect processing parameters based on the objective feature model of the sound effect, with the objective feature of the target sound effect as a constraint, using the Monte Carlo algorithm to generate target sound effect processing parameters that match the objective feature of the target sound effect, and to configure the sound effect according to the target sound effect processing parameters.

10. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method of any one of claims 1 to 8.

Images