Sagittal plane auditory localization method and device based on Bayesian reasoning and medium

Abstract

The invention relates to the technical field of auditory modeling, in particular to a sagittal plane auditory positioning method and device based on Bayesian reasoning and a medium. According to the method, the perception evidence and the spatial prior information are integrated based on the Bayesian criterion so as to simulate the auditory positioning behavior of the human on the sagittal plane. A sensing likelihood function is constructed by calculating spectral cross-correlation between input spectrum features and a feature template, and an adaptive correlation-similarity mapping mechanism is introduced to relieve the discomfort problem in a spectrum dynamic sound source localization task, so that the stability and accuracy of polar angle estimation are improved. A personalized head-related transfer function and positioning response data of a user are utilized to fit model parameters, so that a polar angle estimation result of the model in a complex listening scene is closer to real positioning performance of the user.

Description

Technical Field

[0001] The present invention relates to the technical field of auditory modeling, and in particular to a sagittal plane auditory localization method, device and medium based on Bayesian reasoning. Background Art

[0002] The human auditory system uses monaural spectral cues to estimate the polar angular direction of sound sources in the sagittal plane. These cues are primarily formed by directional filtering of sound by the torso, head, and especially the pinna. Although some progress has been made in uncovering these mechanisms, the neural processing of spectral cues in auditory localization remains incompletely understood, especially in complex acoustic scenes.

[0003] Computational auditory models qualitatively describe and explain the human auditory perception mechanism by simulating the auditory processing stage of the auditory system. Traditional auditory localization models perform well when the sound source spectrum is relatively flat, generating human-like localization predictions. However, their localization performance degrades significantly when the sound source has dynamic spectral characteristics. These models typically use template matching criteria based on the L1 norm or Gaussian likelihood. When both the sound source spectrum and head-related transfer functions (HRTFs) are unknown, polar angle estimation becomes a mathematically ill-posed problem. Summary of the Invention

[0004] In view of this, the present invention provides a sagittal plane auditory localization method, device and medium based on Bayesian reasoning to solve the ill-posed problem of polar angle estimation in the prior art.

[0005] In a first aspect, the present invention provides a sagittal plane auditory localization method based on Bayesian reasoning, the method comprising: obtaining a user's personalized head-related transfer function and positioning response data; extracting spectral features of the left and right ears respectively from binaural input signals; according to the Bayesian criterion, performing cross-correlation calculation on the spectral features and feature templates of each polar angle direction on the target sagittal plane to determine the perceptual likelihood function in each polar angle direction, and determining the posterior probability distribution of each polar angle by combining the perceptual likelihood function with the spatial prior distribution, wherein the feature template is determined by the personalized head-related transfer function; selecting the polar angle prediction value of the model from the posterior probability distribution based on the maximum a posteriori estimation criterion, wherein the model is a model constructed based on spectral feature extraction and Bayesian reasoning; fitting model parameters to minimize the deviation between the model prediction and the actual positioning response under a preset threshold; and performing polar angle direction localization on the binaural input signals based on the personalized fitting model, wherein the personalized fitting model is a model after parameter fitting.

[0006] In this paper, perceptual evidence and spatial prior information are integrated based on the Bayesian criterion to simulate human auditory localization behavior in the sagittal plane. A perceptual likelihood function is constructed by calculating the spectral cross-correlation between input spectral features and feature templates to alleviate the ill-posedness problem in spectral dynamic sound source localization tasks, thereby improving the stability and accuracy of polar angle estimation. The model parameters are fitted using the user's personalized head-related transfer function and localization response data, making the model's polar angle estimation results in complex listening scenarios more similar to the user's actual localization performance.

[0007] In an optional embodiment, obtaining a user's personalized head-related transfer function and positioning response data includes: obtaining the user's personalized head-related transfer function based on personalized measurements or physiological parameter matching; performing a psychoacoustic localization experiment to collect the user's positioning response data in all acquisition directions on a spherical grid, and resampling the user's personalized head-related transfer function to match the spatial resolution of the positioning response data.

[0008] In this paper, head-related transfer functions are obtained through personalized measurement or physiological parameter matching, and positioning response data is collected in conjunction with psychoacoustic localization experiments. The HRTF is then resampled to match the data's spatial resolution. This allows for precise adaptation to the user's physiological characteristics, enhancing the degree of HRTF personalization and matching the positioning response data with the HRTF's spatial characteristics. This provides more realistic, user-friendly basic data for subsequent tasks such as polar angle estimation, enhancing the accuracy and robustness of model predictions.

[0009] In an optional embodiment, spectral features of the left and right ears are respectively extracted from the binaural input signals, including: extracting initial eigenvectors of the binaural input signals based on a gammatone filter bank that approximates basilar membrane filtering and half-wave rectification of hair cell conduction; determining spectral amplitude profile features based on the root mean square amplitude of the initial eigenvectors; and determining noisy features by combining the spectral amplitude profile features and additive Gaussian noise, with the noisy features being used as spectral features.

[0010] In this paper, a gammatone filter bank and half-wave rectification are used to extract initial features that align with physiological mechanisms; the root mean square amplitude determines the spectral amplitude profile, focusing on key energy features; and additive Gaussian noise simulates perceptual noise. This simulated auditory system processing results in extracted spectral features that more closely resemble human auditory perception, providing realistic physiological input for polar angle estimation and other applications, improving the model's robustness and perceptual simulation accuracy in complex acoustic environments.

[0011] In an optional embodiment, according to the Bayesian criterion, the spectral features and the feature templates of each polar angle direction on the target sagittal plane are cross-correlated and calculated to determine the perceptual likelihood function in each polar angle direction, and the posterior probability distribution of each polar angle is determined by combining the perceptual likelihood function with the spatial prior distribution, including: calculating the cross-correlation coefficient based on the spectral features and the feature templates of each polar angle direction on the target sagittal plane, retaining the positive correlation coefficient and setting the non-positive correlation coefficient to zero, to obtain a unilateral correlation vector related to each polar angle direction, the unilateral correlation vector including a left-side correlation vector and a right-side correlation vector; using a psychometric function to map the unilateral correlation vector into a similarity vector related to each polar angle direction, the similarity vector including a left-side similarity vector and a right-side similarity vector, the perceptual selectivity parameter and the perceptual sensitivity parameter of the psychometric function using a fixed value or using an adaptive correlation-similarity mapping mechanism for adaptive adjustment; using a binaural weighting function to weightedly sum and normalize the left-side similarity vector and the right-side similarity vector to obtain the perceptual likelihood function related to each polar angle direction; and determining the posterior probability distribution of each polar angle by combining the perceptual likelihood function and the spatial prior distribution.

[0012] In the present invention, the Bayesian criterion is combined with cross-correlation calculation, psychometric function, and binaural weighting function to accurately simulate the auditory perception process: retaining positive correlation to focus on effective matching, psychometric function mapping subjective perception, binaural weighting integrating binaural information, and combining spatial prior information to make Bayesian reasoning more consistent with the human auditory localization mechanism, which can effectively solve the ill-posed problem of dynamic sound source polar angle estimation and improve the accuracy and reliability of polar angle estimation.

[0013] In an optional embodiment, when the perceptual selectivity parameter and the perceptual sensitivity parameter are adaptively adjusted using an adaptive correlation-similarity mapping mechanism, a psychometric function is used to map the unilateral correlation vector into a similarity vector associated with each polar angular direction, including: arranging the unilateral correlation vectors of each polar angular direction in descending order, and determining the maximum value and second peak value of the unilateral correlation vector; performing linear regression on a preset psychometric function based on the maximum value of the unilateral correlation vector and the maximum value of the preset similarity vector, as well as the second peak value of the unilateral correlation vector and the minimum value of the preset similarity vector, to determine the adaptive parameters; and mapping the unilateral correlation vector into a similarity vector associated with each polar angular direction based on the psychometric function determined by the adaptive parameters.

[0014] In the present invention, an adaptive correlation-to-similarity mapping (ACSM) mechanism can be introduced into the psychometric function to simulate the active direction exploration behavior of humans when the perceptual confidence is low.

[0015] In an optional embodiment, the polar angle prediction value of the model is selected from the posterior probability distribution based on the maximum a posteriori estimation criterion, including: according to the maximum a posteriori estimation criterion, selecting the direction with the largest probability in the posterior distribution as the polar angle prediction value of the target sound source, and introducing response noise, wherein the response noise obeys the von Mises-Fisher distribution.

[0016] In the present invention, real auditory noise interference is simulated by adding von Mises-Fisher distributed response noise in the direction of the maximum probability polar angle.

[0017] In an optional embodiment, model parameters are fitted to minimize the deviation between the model prediction and the true positioning response under a preset threshold, including: taking the polar angle prediction containing response noise as the predicted value and the polar angle direction in the positioning response data as the true value; adjusting the preset parameters so that the deviation between the predicted value and the true value meets the preset requirements, the preset parameters including the spectral feature extraction stage, the Bayesian inference stage and the free parameters in the response noise.

[0018] In the present invention, parameters are adjusted based on the predicted values and true values to optimize the model's adaptability to complex acoustic environments, improve the accuracy and robustness of polar angle estimation, and make the model more consistent with the perceptual behavior pattern of human auditory positioning.

[0019] In an optional embodiment, the preset parameters are adjusted so that the deviation between the predicted value and the true value meets the preset requirements, including: respectively calculating the polarity positioning performance indicators corresponding to the predicted value and the true value, the polarity positioning performance indicators include local polarity error and quadrant error rate, and the predicted value is a model prediction value corresponding to the current model parameters; determining the prediction deviation between the predicted value and the true value based on the polarity positioning performance indicator, the prediction deviation includes the prediction deviation corresponding to the local polarity error and the quadrant error rate respectively; determining the joint prediction deviation based on the prediction deviation corresponding to the local polarity error and the quadrant error rate respectively; adjusting the initial parameters based on the relationship between the joint prediction deviation and the joint prediction deviation threshold, and repeating the polarity positioning performance indicator, prediction deviation, determination of the joint prediction deviation and parameter adjustment process until the relationship between the joint prediction deviation and the joint prediction deviation threshold meets the preset requirements, and the joint prediction deviation threshold is determined based on the threshold of the polarity positioning performance indicator.

[0020] In this paper, by calculating polarity localization metrics (local polarity error and quadrant error rate) between the predicted and true values, we determine the prediction deviation and joint deviation, and iteratively adjust parameters based on their relationship with thresholds. This allows for precise quantification of the difference between model prediction and true performance. By constraining parameter optimization with a joint deviation threshold, we adapt model parameters to the user's auditory characteristics, improve polar angle estimation accuracy, and enhance the model's adaptability and robustness to complex acoustic scenarios.

[0021] In a second aspect, the present invention provides a sagittal plane auditory localization device based on Bayesian reasoning, the device comprising: a data acquisition module for acquiring a user's personalized head-related transfer function and positioning response data; a feature extraction module for extracting spectral features of the left and right ears respectively from binaural input signals; a Bayesian reasoning module for performing cross-correlation calculations on the spectral features and feature templates of each polar angle direction on the target sagittal plane according to the Bayesian criterion, determining the perceptual likelihood function in each polar angle direction, and determining the posterior probability distribution of each polar angle by combining the perceptual likelihood function with the spatial prior distribution, wherein the feature template is determined by the personalized head-related transfer function; a perceptual decision module for selecting a polar angle prediction value of a model from the posterior probability distribution based on a maximum a posteriori estimation criterion, wherein the model is a model constructed based on spectral feature extraction and Bayesian reasoning; a parameter fitting module for fitting model parameters to minimize the deviation between the model prediction and the actual positioning response under a preset threshold; and a direction prediction module for performing polar angle direction localization on the binaural input signals based on the personalized fitting model, wherein the personalized fitting model is a model after parameter fitting.

[0022] In a third aspect, the present invention provides a computer device comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the sagittal plane auditory localization method based on Bayesian reasoning of the above-mentioned first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0023] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the sagittal plane auditory localization method based on Bayesian reasoning of the above-mentioned first aspect or any corresponding embodiment thereof.

[0024] In a fifth aspect, the present invention provides a computer program product comprising computer instructions for causing a computer to execute the sagittal plane auditory localization method based on Bayesian reasoning according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0026] Figure 1 is a flow chart of a sagittal plane auditory localization method based on Bayesian reasoning according to an embodiment of the present invention; Figure 2is a flow chart of another sagittal plane auditory localization method based on Bayesian reasoning according to an embodiment of the present invention; Figure 3 This is a block diagram of a sagittal plane auditory localization model based on Bayesian reasoning according to an embodiment of the present invention; Figure 4 Schematic diagram of the effect of high-frequency speech attenuation on polarity localization performance according to an embodiment of the present invention; Figure 5 Schematic diagram of the effect of non-personalized HRTF on polarity positioning performance according to an embodiment of the present invention; Figure 6 Schematic diagram of the effect of spectrum ripple on polar positioning performance according to an embodiment of the present invention; Figure 7 Schematic diagram of polar localization response of a typical digital listener under different SNRs according to an embodiment of the present invention; Figure 8 Schematic diagram of polar localization performance of five digital listeners under different SNRs according to an embodiment of the present invention; Figure 9 Schematic diagram of polarity localization performance of speech stimulation in the absence of high-frequency information according to an embodiment of the present invention; Figure 10 is a structural block diagram of a sagittal plane auditory localization device based on Bayesian reasoning according to an embodiment of the present invention; Figure 11 Schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] As discussed in the background, the human auditory system's physiological structures, including the auricle (outer ear), external auditory canal, and head, filter sounds incident from different directions. Specifically, when a sound source is located at a certain polar angle within the sagittal plane, the path of the sound entering the auricle is altered by the folds and curvature of the auricle, causing sound waves of different frequencies to interfere, reflect, or absorb, resulting in specific spectral characteristics. The brain memorizes or learns the correspondence between these spectral characteristics and polar angles, allowing it to determine the vertical direction of the sound source based solely on single-ear input.

[0028] The head-related transfer function (HRTF) describes the spectral changes of sound from a free field to the eardrum. For example, HRTFs at different polar angles are pre-measured to create a template library of polar angle-spectral features. The perceived spectrum is then matched against templates in this library to determine the polar angle of the sound source. This matching is most effective when the sound source spectrum is flat. However, if the sound source spectrum is dynamic, the time-varying spectrum is coupled with the HRTF features, making it impossible for template matching to separate the "sound source characteristics" from the "directional characteristics," resulting in reduced accuracy in polar angle estimation.

[0029] Based on this, this embodiment uses the perceptual likelihood function based on spectral cross-correlation in Bayesian reasoning to effectively solve the ill-posed problem of the auditory localization model when predicting the polar angle direction of the spectral dynamic signal.

[0030] To make the purpose, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts shall fall within the scope of protection of the present invention.

[0031] According to an embodiment of the present invention, an embodiment of a sagittal plane auditory localization method based on Bayesian reasoning is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0032] In this embodiment, a sagittal plane auditory localization method based on Bayesian reasoning is provided, which can be used in electronic devices such as computers, mobile phones, tablet computers, etc. Figure 1 is a flow chart of a sagittal plane auditory localization method based on Bayesian reasoning according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps: Step S101 obtains the user's personalized head-related transfer function (HRTF) and localization response data. The HRTF describes the spectral modulation characteristics produced by scattering, reflection, and diffraction of physiological structures such as the head, auricle, and torso when sound propagates from free space to the entrance of the ear canal. It is specifically determined by the spatial direction (lateral and polar angles) of the sound source and the sound frequency. That is, sound sources in different directions have different HRTFs. Due to individual differences in human acoustic structure, HRTFs vary from person to person. Therefore, this embodiment determines personalized HRTFs for different users. Personalized HRTFs can be determined through measurement or modeling, and are not further described here. Localization response data represents the result of the user's directional judgment of sound sources in different directions, reflecting the user's actual perceived sound source direction.

[0033] Step S102 extracts spectral features for the left and right ears from the binaural input signals. The spectral feature extraction process simulates the processing of the human auditory system, meaning the extracted spectral features represent internal auditory perception. The binaural input signals can be acquired using two microphones at the entrances of the left and right ear canals in a real-world or laboratory acoustic environment (e.g., an anechoic chamber or reverberation chamber).

[0034] In step S103, according to the Bayesian criterion, the spectral features are cross-correlated with the feature templates at each polar angle on the target sagittal plane to determine the perceptual likelihood function at each polar angle. The perceptual likelihood function is then combined with the spatial prior distribution to determine the posterior probability distribution at each polar angle. The feature templates are determined by the personalized head-related transfer function.

[0035] In anatomy, the sagittal plane is a vertical plane that divides an organism (such as the head) into two symmetrical left and right halves (similar to a plane that cuts the head from the tip of the nose to the back of the skull). In auditory localization, sound source localization within the sagittal plane primarily involves determining vertical directions (such as above, front, below, and back), as distinct from horizontal (left and right) localization. The polar angle refers to the angular deviation from a reference direction (such as straight ahead) to the target direction in a polar coordinate system.

[0036] In this embodiment, the polar angle is defined in the binaural polar coordinate system, which together with the lateral angle determines the spatial direction of the sound source relative to the user. , the lateral angle range is from (right) to 90° (left), while the polar angle range is from (front and lower) to 270° (back and lower). The target sagittal plane is responded by the transverse angle Decide.

[0037] Specifically, the feature template uniformly resamples the user's full-space DTF (Directional Transfer Function) data to obtain DTF data for a fixed number of directions. By removing the direction-independent Common Transfer Function (CTF) from the HRTF, the direction-dependent DTF data is obtained. Finally, through the aforementioned spectral feature extraction stage, the spectral feature template library is obtained.

[0038] After determining the feature template, the extracted input spectral features are cross-correlated with the feature template to measure the feature matching degree; the cross-correlation result is then converted into a perceptual likelihood function, reflecting the probability that "the spectral feature comes from a certain polar angle direction"; then, combined with spatial priors (such as the statistical law of the probability of occurrence of polar angles), the likelihood and prior are fused through Bayes' theorem to obtain the posterior probability distribution of each polar angle, realizing the probabilistic inference of the polar angle direction, so that the model fits the user's auditory physiology and spatial statistical characteristics.

[0039] Step S104 , selecting a polar angle prediction value of a model from the posterior probability distribution based on a maximum a posteriori estimation criterion, wherein the model is a model constructed based on spectrum feature extraction and Bayesian reasoning.

[0040] Specifically, after determining the posterior probability distribution, the polar angle direction corresponding to the maximum probability, that is, the predicted value, is determined from the posterior probability distribution based on the maximum a posteriori estimation criterion.

[0041] Step S105 : Fitting model parameters to minimize the deviation between the model prediction and the actual positioning response under a preset threshold.

[0042] Specifically, during the model parameter fitting process, the obtained predicted value and positioning response data are compared. If the deviation between the two does not meet the preset conditions, the parameter fitting process is continued until the deviation between the two meets the preset conditions.

[0043] Step S106: Polar angular orientation of the binaural input signals is performed based on the personalized fitting model, which is a parameter-fitted model. Specifically, based on the fitting parameters from the previous step, the polar angular orientation of any binaural input signal can be estimated, ultimately determining the direction of the sound source in the sagittal plane. That is, steps S102 to S104 are performed on any binaural input signal to ultimately determine the polar angular orientation of the sound source.

[0044] The sagittal plane auditory localization method based on Bayesian reasoning provided by an embodiment of the present invention integrates perceptual evidence and spatial prior information based on the Bayesian criterion to simulate human auditory localization behavior in the sagittal plane. By calculating the spectral cross-correlation between the input spectral features and the feature template, a perceptual likelihood function is constructed to alleviate the ill-posed problem in the spectral dynamic sound source localization task, thereby improving the stability and accuracy of the polar angle estimation. The model parameters are fitted using the user's personalized head-related transfer function and positioning response data, so that the polar angle estimation results of the model in complex listening scenarios are closer to the user's actual positioning performance.

[0045] In this embodiment, a sagittal plane auditory localization method based on Bayesian reasoning is provided. The process includes the following steps: Step S201: Obtain the user's personalized head-related transfer function and positioning response data.

[0046] Specifically, the above step S201 includes: Step S2011: Obtain the user's personalized head-related transfer function (HRTF) based on personalized measurements or physiological parameter matching. Specifically, when using personalized measurements to obtain this function, professional acoustic equipment can be used to transmit sound signals (such as broadband noise or swept-frequency signals) of known frequencies into the ear canal from different spatial directions (covering common lateral and polar angles) in an anechoic chamber or other environment. Simultaneously, a miniature microphone placed at the ear canal entrance (or a simulated ear canal location) collects the sound signals, which have been filtered by physiological structures such as the head and auricle. For example, a multi-speaker array can be used to surround the subject, playing signals sequentially at preset angles while simultaneously collecting the received signals from the ear canal. By comparing the spectral differences between the free-field signal (without head obstruction) and the collected ear canal signal, HRTFs corresponding to different directions are calculated, reflecting the filtering characteristics of the specific user's head and auricle for sounds from different directions.

[0047] When using physiological parameter matching to determine this function, a database can be established based on a large amount of measured HRTFs data. Physiological parameters that are strongly correlated with HRTFs (such as head size, three-dimensional shape of the auricle, and length and diameter of the ear canal) can be extracted. After measuring these physiological parameters for a new user, the HRTFs of individuals with the most similar physiological characteristics are matched in the database to serve as an approximation of the user's personalized HRTFs. For example, parameters such as the user's auricle shape and head circumference can be obtained through 3D scanning and input into a machine learning model. The model then selects sample HRTFs from the database that have a high degree of match with the physiological parameters, and after interpolation and optimization, assigns them to the new user.

[0048] Step S2012, execute the psychoacoustic localization experiment to collect the user's localization response data in all acquisition directions on the spherical grid, and resample the user's personalized head-related transfer function to match the spatial resolution of the localization response data. Specifically, the psychoacoustic localization experiment refers to an experiment in which users make subjective orientation judgments on sound sources in different spatial positions in a controllable acoustic environment (such as an anechoic room or a virtual acoustic environment), with the purpose of obtaining behavioral data on the user's actual perception of the direction of the sound source. The spherical grid refers to dividing the sphere centered on the user in three-dimensional space into discrete "grid points" according to certain rules. Each grid point corresponds to a spatial direction (using a grid containing polar angles). , lateral angle interaural polar coordinate description).

[0049] After determining the localization response data for all acquisition directions on the spherical grid through psychoacoustic localization experiments, it is necessary to further determine whether the spatial resolution of the personalized head-related transfer function determined in the previous step is consistent with the spatial resolution of the localization response data. If the two are inconsistent, for example, if the HRTFs are sampled at a polar angle of 5° and an azimuth angle of 10°, and the localization response data are sampled at a polar angle of 10° and an azimuth angle of 15°, the HRTFs need to be resampled.

[0050] Specifically, if the HRTF sampling density is higher than the positioning response data, for the sparser grid points in the positioning response data, the data of the surrounding HRTF sampling points are used to interpolate (such as spherical interpolation, linear interpolation, etc.) to calculate the HRTF value corresponding to the sparse point. For example, if the spectrum data of several adjacent HRTF sampling points are known, the spherical interpolation algorithm is used to fit the HRTFs of the target sparse point. If the HRTF sampling density is lower than the positioning response data, the point closest (in the spatial direction) to the positioning response data grid point is selected from the HRTF sampling points as the matching point; or the HRTF sampling data is filtered, decimated, or other downsampling operations are performed to make its spatial resolution consistent with the positioning response data.

[0051] Step S202: extracting the spectrum features of the left and right ears from the binaural input signals respectively.

[0052] Specifically, the above step S202 includes: Step S2021 , extracting initial feature vectors of binaural input signals based on a gammatone filter bank that approximates basilar membrane filtering and half-wave rectification of hair cell conduction.

[0053] Among them, the basilar membrane in the human cochlea has frequency selectivity, and different positions respond differently to sounds of different frequencies, similar to a group of filters, to achieve frequency decomposition of sound. The gammatone filterbank is a mathematical model that simulates the frequency decomposition characteristics of the basilar membrane. Specifically, each gammatone filter has a specific center frequency, performs bandpass filtering on the input signal, and outputs the signal components of the corresponding frequency range. In this embodiment, the filter bank is used to divide the 0.7-18kHz frequency range into 28 frequency bands, and each frequency band corresponds to a filter. In this way, the complex sound signal is decomposed into 28 channels according to the frequency, simulating the response of the basilar membrane to sounds of different frequencies. The hair cells in the cochlea can convert the mechanical signal of the basilar membrane vibration into an electrical signal, and its response is nonlinear. Half-wave rectification is a simple process to simulate this nonlinear characteristic.

[0054] Based on this, the 28-dimensional (frequency band) initial eigenvector of each side (ear) signal is obtained according to the gammatone filter bank that approximates basilar membrane filtering and half-wave rectification of hair cell conduction. The initial eigenvector of each frequency band is expressed by the following formula: (1) Where, Represents the gammatone filter, subscript is the frequency band index, is the time index, Indicates stimulus (source) signal Unilateral head-related impulse response The monaural input signal obtained after linear convolution, superscript To approximate the spectral resolution of the human cochlea, a gammatone filter bank was set to an equivalent rectangular bandwidth of 1, dividing the signal into 28 bands in the frequency range of 0.7 to 18 kHz. A max function was used to retain the non-negative components and implement half-wave rectification to simulate the nonlinearity of hair cell conduction.

[0055] Step S2022: Determine a spectral magnitude profile feature based on the root mean square amplitude of the initial feature vector. Specifically, the root mean square amplitude over the entire time range is collected from the initial feature vector to obtain a spectral magnitude profile (SMP) feature. The SMP feature is expressed as follows: (2) In addition, since the initial feature vector extracted above includes vectors of 28 frequency bands, the SMP features on the 28 frequency bands are combined into a monaural spectrum feature vector, which is expressed by the following formula: (3) In step S2023, the noisy feature is determined by combining the spectrum amplitude profile feature and the additive Gaussian noise. The noisy feature is used as the spectrum feature. Specifically, after determining the spectrum amplitude profile, the additive Gaussian noise is added to simulate the limited perceptual accuracy of the feature extraction stage. Therefore, the noisy feature after adding the additive Gaussian noise is expressed as follows: (4) Step S203: Based on the Bayesian criterion, the spectral features are cross-correlated with the feature templates at each polar angle on the target sagittal plane to determine the perceptual likelihood function at each polar angle. The perceptual likelihood function is then combined with the spatial prior distribution to determine the posterior probability distribution at each polar angle. The feature templates are determined using a personalized head-related transfer function. Specifically, the above step S203 includes: Step S2031: Calculate the cross-correlation coefficient based on the spectral features and the feature templates of each polar angle direction on the target sagittal plane, retain the positive correlation coefficients, and set the non-positive correlation coefficients to zero, to obtain the unilateral correlation vectors related to each polar angle direction. The unilateral correlation vectors include the left-side correlation vector and the right-side correlation vector. Specifically, the cross-correlation coefficients are calculated one by one between the spectral features and the feature templates of each polar angle direction on the target sagittal plane. The unilateral correlation vectors are determined using the following formula: (5) In the formula, the feature template It is calculated based on the user's personalized HRTFs, which includes the polar angle of each target sagittal plane in formula (4) The target sagittal plane is determined by the lateral angle response. Since the extracted spectral features include the spectral features of the left ear and the spectral features of the right ear, the spectral features on both sides are cross-correlated with the feature template, and the determined correlation vectors include the left correlation vector and the right correlation vector.

[0056] In step S2032, a psychometric function is used to map the unilateral correlation vector into a similarity vector associated with each polar angle direction. The similarity vector includes a left-side similarity vector and a right-side similarity vector. The perceptual selectivity parameter and the perceptual sensitivity parameter of the psychometric function are fixed values or adaptively adjusted using an adaptive correlation-similarity mapping mechanism. Specifically, human auditory perception is not a simple "feature matching" and has nonlinearity of subjective perception, sensitivity differences, etc. The psychometric function simulates this nonlinear mapping of "feature to perceptual similarity" and converts the obtained correlation vector into a similarity vector that is more in line with human subjective perception. Specifically, the similarity vector is calculated using the following formula: (6) Where, and They represent the perceptual selectivity parameter and the perceptual sensitivity parameter respectively.

[0057] In an optional implementation, fixed values can be used for the parameters in the psychometric function. Furthermore, in order to better fit the perception law of human hearing in uncertain scenarios, an adaptive correlation-to-similarity mapping (ACSM) mechanism is introduced to simulate the active direction exploration behavior of humans when the perception confidence is low. Based on this, the parameters and Determined as follows: Step a1: Arrange the one-sided correlation vectors of each polar angle direction in descending order, and determine the maximum value and the second peak value of the one-sided correlation vectors.

[0058] Step a2, based on the maximum value of the unilateral correlation vector and the maximum value of the preset similarity vector, as well as the second peak value of the unilateral correlation vector and the minimum value of the preset similarity vector, a linear regression solution is performed on the preset psychometric function to determine the adaptive parameters. Among them, the preset maximum similarity value and the preset minimum similarity value can be predetermined. These two values can limit the value range of the similarity, avoiding the failure of the model to depict the "low confidence" scene due to extreme values, thereby achieving the simulation of human perception ambiguity. In this embodiment, the preset minimum similarity value is Similarity_min = 0.01, preset maximum similarity value Similarity_max = 0.99.

[0059] Thus, based on the maximum value of the one-sided correlation vector C_max and the preset maximum similarity vector and the second peak of the unilateral correlation vector C_2ndmax and the two points determined by the preset similarity vector minimum value ( C_max , Similarity_max )and( C_2ndmax , Similarity_min ) Perform linear regression to solve the adaptive parameters and .

[0060] Step a3: Map the unilateral correlation vector to the similarity vector related to each polar angle direction based on the psychometric function determined by the adaptive parameters. Specifically, after the adaptive parameters are determined, they are substituted into the psychometric function, and the left-side correlation vector and the right-side correlation vector determined in step S2031 are substituted into the function to obtain the left-side similarity vector. and the similarity vector on the right .

[0061] In step S2033, a binaural weighting function is used to perform weighted summation and normalization on the left-side similarity vector and the right-side similarity vector to obtain a perceptual likelihood function associated with each polar angle direction. Specifically, the determined perceptual likelihood function is expressed by the following formula: (7) Where, is the weighting function, binaural weight coefficient , in this model it is assumed that the lateral angular response Known.

[0062] Step S2034: Determine the posterior probability distribution of each polar angle by combining the perceptual likelihood function and the spatial prior distribution. The spatial prior distribution is defined as a finite variance Gaussian distribution centered on the horizontal plane in the vertical dimension: (8) Where, the variance of the posterior hemisphere is Greater than the variance of the front hemisphere .

[0063] Based on the spatial prior distribution, the posterior probability distribution is expressed as follows: (9) Step S204: selecting the polar angle prediction value of the model from the posterior probability distribution based on the maximum a posteriori estimation criterion.

[0064] Specifically, the above step S204 includes: Step S2041: Based on the maximum a posteriori estimation criterion, the direction with the highest probability in the posterior distribution is selected as the polar angle prediction value of the target sound source, and response noise is introduced. The response noise follows a von Mises-Fisher distribution. Specifically, based on the maximum a posteriori estimation criterion, the direction with the highest probability in the posterior distribution is selected as the polar angle prediction value of the target sound source. Response noise is then introduced to simulate the uncertainty inherent in motion response or decision-making processes, thereby obtaining the final prediction result. The prediction value after adding noise is expressed as follows: (10) Where, Subject to the mean of zero and the concentration parameter is The concentration can be converted to the standard deviation of the von Mises-Fisher distribution. .

[0065] Step S205 : Fitting model parameters to minimize the deviation between the model prediction and the actual positioning response under a preset threshold.

[0066] Specifically, the above step S205 includes: Step S2051: taking the polar angle prediction including the response noise as the predicted value and the polar angle direction in the positioning response data as the true value.

[0067] Step S2052: Adjust the preset parameters so that the deviation between the predicted value and the true value meets the preset requirements. The preset parameters include the free parameters in the spectrum feature extraction stage, the Bayesian inference stage, and the response noise. Specifically, in this embodiment, the adjusted parameters include the variance of the front hemisphere. , the variance of the posterior hemisphere , the parameters of additive Gaussian noise , concentration standard deviation , perceived selectivity parameter , perception sensitivity parameter .

[0068] In an optional embodiment, adjusting the preset parameters so that the deviation between the predicted value and the true value meets the preset requirements includes the following steps: Step b1, respectively calculating the polarity positioning performance indicators corresponding to the predicted value and the true value, the polarity positioning performance indicators include local polarity error and quadrant error rate, and the predicted value is the model prediction value corresponding to the current model parameters.

[0069] Specifically, the initial value and value range of the preset parameters can be set as follows: The initial value is 11.5° and the range is [5°, 25°]; The initial value is 11.5° and the range is [5°, 40°]; The initial value is 3.5° and the range is [0.5°, 10°]; The initial value is 17° and the range is [5°, 25°]; The initial value is 6 , in the range [1,40] ; The initial value is 0.7 and the range is [0.1, 1].

[0070] Among them, the polar positioning performance index of the positioning response data at 0°, ±20°, ±40°, and ±60°, with an interval of 20°, can be calculated based on the predicted value determined by the initial parameters and the true value in the positioning response data. This index includes local polar errors (PE) and quadrant error rate (QE). When calculating the polar positioning performance index, based on the initial parameters, each target direction is repeated a preset number of times, such as K times. Specifically, the two indicators are expressed by the following formulas: (11) (12) In the formula, when calculating the polar positioning performance index of the true value, Indicates the polar angle (true value) actually perceived by humans in the kth trial. When calculating the polar positioning performance index of the true value, represents the predicted value of the polar angle; Indicates the target polar angle (theoretical value) in the kth trial. Represents the local polar angle responses whose absolute error with the target polar angle is less than 90 degrees, The function limits the angle difference to [ , 180°]. PE is the root mean square error of the trials with polar angle errors less than 90 degrees, and QE is the percentage of trials with polar angle errors exceeding 90 degrees.

[0071] Step b2: Determine the prediction deviation between the predicted value and the true value based on the polarity positioning performance index. The prediction deviation includes the prediction deviation corresponding to the local polarity error and the quadrant error rate. Specifically, the prediction deviation is expressed by the following formula: (13) Where, The interval index representing the lateral angle, represents the number of lateral angle intervals, Represents the occurrence rate of the sound source in each lateral angle interval. In addition, It represents the indicator PE or QE, and the superscripts A and P represent actual (real) performance and predicted performance, respectively.

[0072] Step b3: determining a joint prediction deviation based on the prediction deviations corresponding to the local polarity error and the quadrant error rate; specifically, the joint prediction deviation is expressed by the following formula: (14) In the formula, the chance rate ( and ) represents the listener's positioning performance when the target location is completely guessed.

[0073] Step b4, adjust the initial parameters based on the relationship between the joint prediction deviation and the joint prediction deviation threshold, and repeat the polarity positioning performance index, prediction deviation, determination of the joint prediction deviation and parameter adjustment process until the relationship between the joint prediction deviation and the joint prediction deviation threshold meets the preset requirements, and the joint prediction deviation threshold is determined based on the threshold of the polarity positioning performance index.

[0074] Among them, for the polarity positioning performance index, the following formula is used to determine the real-person positioning performance and predictive positioning performance Is the relative deviation between the two smaller than the target threshold? : (15) Where, the target threshold Including thresholds corresponding to two performance indicators and ,in, , , combining these two thresholds, and formula (14) and formula (15) can be converted to obtain the joint prediction deviation threshold Then the joint prediction deviation determined based on formula (14) is compared with the joint prediction deviation threshold. Converge to threshold If the following is true, the preset requirements are met and the parameters at this time can be solidified; otherwise, the parameters need to be adjusted and the above steps S202 to S204 are repeated until the joint prediction deviation d converges to the threshold In the following, the obtained parameters are used as the final parameters.

[0075] In addition, it should be noted that the ACSM mechanism is not introduced in the parameter adjustment process because the model calibration relies on the positioning data obtained based on broadband noise stimulation.

[0076] Step 206: Perform polar angle positioning on the binaural input signals based on the personalized fitting model, wherein the personalized fitting model is a model after parameter fitting. Figure 1 Step S106 of the illustrated embodiment will not be described in detail here.

[0077] In this paper, a perceptual likelihood function based on spectral cross-correlation effectively addresses the ill-posedness problem of auditory localization models when predicting the direction of spectral dynamic signals. Furthermore, the solution incorporates an adaptive correlation-to-similarity mapping (ACSM) mechanism to simulate the active directional exploration behavior of humans when perceptual confidence is low.

[0078] As a specific application example of the embodiment of the present invention, Figure 2 As shown in FIG, the sagittal plane auditory localization method based on Bayesian reasoning is implemented using the following process: Step S1, obtaining the user's personalized head-related transfer functions (HRTFs) and positioning response data; Step S2, extracting the spectrum features of the left and right ears from the binaural input signals respectively as representations of internal auditory perception; Step S3: Based on the Bayesian criterion, the extracted spectral features are cross-correlated with the user's feature template in the target sagittal plane direction to obtain the perceptual likelihood function in the candidate polar angle direction. This function is then combined with the spatial prior distribution to form the posterior probability distribution of the target polar angle. Step S4: According to the maximum a posteriori estimation criterion, the direction with the highest probability in the posterior distribution is selected as the polar angle prediction value of the target sound source. Response noise is then introduced to simulate the uncertainty inherent in the motion response or decision-making process, thereby obtaining the final prediction result of the model. Step S5: searching for a set of optimal model parameters within a preset parameter range so that the average polarity positioning error between the model prediction value and the true value is lower than a target threshold, and using this set of parameters as the user's personalized model parameters; In step S6, based on the calibrated personalized model, the polar angle direction of any binaural input signal can be estimated, and the direction of the sound source in the sagittal plane can be finally determined.

[0079] Taking a user as an example, the sagittal plane auditory localization method based on Bayesian reasoning includes: The localization response data of a normal hearing subject were collected. The target sound sources were evenly distributed on the surface of a virtual sphere centered on the listener. The target positions were described using the lateral angle and polar angle in the binaural polar coordinate system, where the lateral angle range was from (right) to 90° (left), the polar angle range is from The local polar errors (PE) and quadrant error rate (QE) of the real-person positioning response data were calculated at seven lateral angle intervals centered at 0°, ±20°, ±40°, and ±60° and spaced 20° apart as the true localization performance values. Each subject was measured at no fewer than 750 directions, and the measurement scale affected the quality of the parameter fitting. In this example, the sound source signal was 500 milliseconds of broadband Gaussian noise.

[0080] Set the initial model parameters ( , , , , , ) are as follows: , , , , , . Figure 3 A block diagram of a sagittal plane auditory localization model based on Bayesian reasoning provided in an embodiment of the present invention, such as Figure 3 As shown in the figure, the head-related impulse responses in different directions are convolved with broadband white noise as binaural input signals, and the monaural spectral feature vector is calculated according to equations (1) to (3): , then consider additive Gaussian noise (Eq. (4)) to simulate the monaural spectral features containing the auditory system’s perceptual ambiguity . According to this feature calculation process, the noise-free monaural spectral feature vector is extracted from the head-related transfer functions (HRTFs) of the subject in different directions as a matching template. The spectral features extracted from the binaural input signal are cross-correlated with the user's feature template on the target sagittal plane one by one (Equation (5)), and finally the perceptual likelihood function for the candidate polar angle direction is obtained through correlation-similarity mapping (Equation (6)) and binaural weighting (Equation (7)). According to the Bayesian criterion of Equation (8), the posterior distribution of the target polar angle is calculated by combining the spatial prior distribution and the likelihood function of Equation (9). In the decision stage (Equation (10)), the maximum value of the posterior distribution is searched as the polar angle estimate by relying on the maximum a posteriori estimation method, and finally the motion response representing the inherent uncertainty of human auditory perception is added to obtain the polar angle prediction value. .

[0081] In this embodiment, the subject's existing HRTF dataset is resampled to 1500 directions to be compatible with the non-uniform HRTF acquisition grid, and the simulation is repeated 300 times for 1500 directions to simulate the inherent randomness in the model estimation. In this embodiment, the lateral angle of the target sound source is assumed to be known in the model prediction process to focus on the model's polar angle prediction performance. In this embodiment, the correlation-similarity mapping of Equation (6) can adopt an adaptive version. After enabling this mechanism, the model can dynamically adjust the parameters according to the cross-correlation results of the real-time spectrum. and , thereby increasing attention to the peak correlation area. This mechanism helps improve the model's performance in complex auditory environments, making it closer to the human auditory behavior of actively exploring the environment, and enhancing the consistency and rationality of the model with the actual perception process.

[0082] Based on the real-person positioning data and the model prediction values, PE (Equation (11)) and QE (Equation (12)) in different lateral angle intervals are calculated as the actual value and predicted value of the positioning performance, respectively. The prediction deviation between the model prediction value and the human response data under the two performance indicators is calculated according to Equation (13): , and then the joint prediction deviation is obtained by weighted integration based on the respective probability of occurrence of performance indicators (Equation (14)). According to the relative deviation threshold corresponding to the two performance indicators (Equation (15)) calculates the joint prediction residual threshold Search for a set of optimal model parameters within a given parameter range so that the joint prediction deviation is lower than the target threshold , and use this set of parameters as the personalized model parameters for the user. In this embodiment, in order to minimize the prediction deviation , the fmincon function with interior point method in Matlab was used, and the multi-starting point optimization strategy was applied to run the local optimization algorithm from multiple initial points to find the global optimal solution.

[0083] Five models with personalized parameter calibration were used as "digital listeners" to conduct five positioning experiments to evaluate the impact of different factors on the model positioning performance at the overall level. Experiments 1-3 reproduced the positioning experimental design in previous studies, respectively evaluating the impact of high-frequency attenuation of speech, non-personalized HRTF, and spectral ripples on positioning performance, and obtaining corresponding real-person data for comparison with the prediction results of the positioning model proposed in this invention. Experiments 4-5 aim to explore the impact of the adaptive correlation-similarity mapping mechanism on the model positioning performance in complex acoustic environments, specifically including different signal-to-noise ratio (SNR) conditions and the lack of high-frequency cues. Listener-specific fitting parameters were used in the first three experiments, while the parameters related to similarity mapping in Experiments 4-5 were used. , It will be dynamically adjusted according to the Adaptive Correlation-to-Similarity Mapping (ACSM) mechanism.

[0084] Figure 4 The effect of different degrees of high-frequency speech attenuation on the model's localization performance is demonstrated. The experimental stimulus signal consists of 260 monosyllabic speech samples selected from a broadband corpus, with an average duration of 710ms and a frequency range of 0.3-16 kHz. The speech passes through an 8 kHz cutoff frequency and a stopband attenuation of 0dB. dB, and The stimuli were low-pass filtered at 100 dB, and broadband Gaussian white noise was used as a baseline stimulus for comparison. The above stimulus signals were evenly placed at 76 random locations on a virtual sphere around the listener's head for auditory localization experiments, with 50 repetitions for each target direction. The mean absolute polar angle error (MAPE) and QE (Equation (12)) were used as performance metrics to evaluate the model:

[0085] The inverse tangent function By considering and The final experimental results show that the model prediction performance shows good consistency with real-life data, that is, the polarity positioning error gradually increases with the increase of high-frequency attenuation.

[0086] Figure 5 The impact of non-personalized HRTFs on model localization performance was demonstrated. The experiment simulated the localization performance (PE and QE) of five digital listeners using their own (Own) and others' (Other) HRTFs within a ±30° lateral angle range while locating 250ms of Gaussian white noise. The results showed that using non-personalized virtual auditory playback significantly increased polar localization errors.

[0087] Figure 6 The effect of spectral corrugation on polar localization performance is demonstrated. The experiment simulated the localization performance of five digital listeners when localizing sound source signals with different corrugation densities and corrugation depths. Under baseline conditions, the model's localization performance was slightly degraded compared to real-person data. The polar error rate (PER) was used to evaluate the model's sagittal plane localization performance, which was achieved by statistically analyzing the model response that deviated from the linear prediction of baseline regression by more than 45 degrees. Under the corrugation density condition, as the density increased, the increase in the PER for human listeners reached a maximum of around 1 ripple per octave. Subsequently, the PER increase gradually decreased, approaching 0 at 8 ripples per octave.

[0088] Figure 7 The figure shows the speech localization responses of a typical subject under seven different SNR conditions with and without ACSM enabled. Each column in the figure represents the localization prediction results under different SNR conditions. In the polar angle localization scatter plot, the dots located on or clustered near the positive diagonal line represent correct or nearly correct polar angle localization responses to the direction of the target sound source, while the dots distributed near the negative diagonal line represent confused responses. At lower SNRs, the polar angle responses when ACSM is not enabled are mainly concentrated near 180°, while the responses after enabling are concentrated at and 180°, at which point the regression fit deviation is shifted to 90°. The above results indicate that ACSM helps prevent the polar angle response from being pulled toward the hemisphere with larger prior variance, making it closer to the direction exploration behavior of humans under low perceptual confidence. The average speech localization results of the five digital listeners are Figure 8 The three sub-graphs show the influence of SNR on the positioning performance of the model. The positioning performance of the model is determined by the positioning gain after linear regression analysis ( ), positioning deviation ( ) and positioning accuracy (coefficient of determination )Evaluate: (17) Among them, the response Representative Target The predicted polar angle, regression parameter and Represents positioning gain and deviation respectively, which can be obtained by - The data were estimated using the least squares method. This can be approximated by calculating the square of the Pearson correlation coefficient between the target and the response. It can be seen that the model positioning performance gradually improves with the increase of SNR after ACSM is enabled, which is consistent with the human auditory perception pattern.

[0089] Figure 9 The low-pass speech localization results of five digital listeners with and without ACSM are presented. The simulation results show that the model prediction performance after ACSM is turned on is close to the lower SNR condition ( ), that is, the ACSM mechanism can simulate the direction exploration behavior of humans in low perception confidence scenarios.

[0090] In this embodiment, a sagittal plane auditory localization device based on Bayesian reasoning is also provided. This device is used to implement the above-mentioned embodiments and preferred embodiments, and the details already described will not be repeated. As used below, the term "module" can refer to a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation using hardware, or a combination of software and hardware, is also possible and contemplated.

[0091] This embodiment provides a sagittal plane auditory localization device based on Bayesian reasoning, such as Figure 10 Shown, including: A data acquisition module 11 is used to obtain the user's personalized head-related transfer function and positioning response data; A feature extraction module 12 is used to extract the spectrum features of the left and right ears from the binaural input signals respectively; A Bayesian inference module 13 is configured to perform cross-correlation calculations on the spectral features and feature templates at each polar angle on the target sagittal plane according to the Bayesian criterion, determine the perceptual likelihood function at each polar angle, and determine the posterior probability distribution at each polar angle by combining the perceptual likelihood function with the spatial prior distribution. The feature templates are determined by a personalized head-related transfer function. A perception decision module 14 is configured to select a polar angle prediction value of a model from a posterior probability distribution based on a maximum a posteriori estimation criterion, wherein the model is a model constructed based on spectral feature extraction and Bayesian reasoning; a parameter fitting module 15 for fitting model parameters to minimize the deviation between the model prediction and the true positioning response under a preset threshold; The direction prediction module 16 is used to perform polar angle positioning on the binaural input signals based on a personalized fitting model, where the personalized fitting model is a model after parameter fitting.

[0092] The further functional description of each of the above modules is the same as that of the above corresponding embodiments and will not be repeated here.

[0093] In an optional embodiment, the data acquisition module is specifically used to: obtain the user's personalized head-related transfer function based on personalized measurements or physiological parameter matching; perform a psychoacoustic localization experiment to collect the user's localization response data in all acquisition directions on a spherical grid, and resample the user's personalized head-related transfer function to match the spatial resolution of the localization response data.

[0094] In an optional embodiment, the feature extraction module is specifically used to: extract the initial feature vectors of the binaural input signal based on a gammatone filter group that approximates basilar membrane filtering and half-wave rectification of hair cell conduction; determine the spectrum amplitude profile feature based on the root mean square amplitude of the initial feature vector; and determine the noisy feature by combining the spectrum amplitude profile feature and additive Gaussian noise, and the noisy feature is used as the spectrum feature.

[0095] In an optional embodiment, the Bayesian reasoning module includes: A correlation vector determination module is used to calculate the mutual correlation coefficient based on the spectral features and the feature templates of each polar angle direction on the target sagittal plane, retain the positive correlation coefficient, and set the non-positive correlation coefficient to zero, to obtain the unilateral correlation vector related to each polar angle direction. The unilateral correlation vector includes a left correlation vector and a right correlation vector; a similarity vector determination module, configured to map the unilateral correlation vector into similarity vectors associated with each polar angle direction using a psychometric function, wherein the similarity vectors include a left-side similarity vector and a right-side similarity vector, wherein the perceptual selectivity parameter and the perceptual sensitivity parameter of the psychometric function are fixed values or adaptively adjusted using an adaptive correlation-similarity mapping mechanism; A perceptual likelihood function determination module is used to perform a weighted summation and normalization process on the left-side similarity vector and the right-side similarity vector using a binaural weighting function to obtain a perceptual likelihood function associated with each polar angle direction; The posterior distribution determination module is used to determine the posterior probability distribution of each polar angle by combining the perceptual likelihood function and the spatial prior distribution.

[0096] In an optional embodiment, when the perceptual selectivity parameter and the perceptual sensitivity parameter are adaptively adjusted using an adaptive correlation-similarity mapping mechanism, the similarity vector determination module is specifically used to: arrange the unilateral correlation vectors of each polar angle direction in descending order, and determine the maximum value and second peak value of the unilateral correlation vector; perform linear regression on a preset psychological measurement function based on the maximum value of the unilateral correlation vector and the preset maximum value of the similarity vector, as well as the second peak value of the unilateral correlation vector and the preset minimum value of the similarity vector, to determine the adaptive parameters; and map the unilateral correlation vector into similarity vectors related to each polar angle direction based on the psychological measurement function determined by the adaptive parameters.

[0097] In an optional embodiment, the perception decision module is specifically used to: select the direction with the largest probability in the posterior distribution as the polar angle prediction value of the target sound source according to the maximum a posteriori estimation criterion, and introduce response noise, wherein the response noise obeys the von Mises-Fisher distribution.

[0098] In an optional embodiment, the parameter fitting module includes: A predicted value and true value determination module is used to use the polar angle prediction containing response noise as the predicted value and the polar angle direction in the positioning response data as the true value; The adjustment submodule is used to adjust the preset parameters so that the deviation between the predicted value and the true value meets the preset requirements. The preset parameters include the spectrum feature extraction stage, the Bayesian inference stage, and the free parameters in the response noise.

[0099] In an optional embodiment, the adjustment submodule is specifically used to: calculate the polarity positioning performance indicators corresponding to the predicted value and the true value respectively, the polarity positioning performance indicators include local polarity error and quadrant error rate, and the predicted value is a model prediction value corresponding to the current model parameters; determine the prediction deviation between the predicted value and the true value based on the polarity positioning performance indicator, the prediction deviation includes the prediction deviation corresponding to the local polarity error and the quadrant error rate respectively; determine the joint prediction deviation based on the prediction deviation corresponding to the local polarity error and the quadrant error rate respectively; adjust the initial parameters based on the relationship between the joint prediction deviation and the joint prediction deviation threshold, and repeat the polarity positioning performance indicator, prediction deviation, joint prediction deviation determination and parameter adjustment process until the relationship between the joint prediction deviation and the joint prediction deviation threshold meets the preset requirements, and the joint prediction deviation threshold is determined based on the threshold of the polarity positioning performance indicator.

[0100] The embodiment of the present invention also provides a computer device having the above Figure 10 The sagittal plane auditory localization device based on Bayesian inference is shown.

[0101] See also Figure 11 , Figure 11is a structural diagram of a computer device provided by an optional embodiment of the present invention, such as Figure 11 As shown, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components utilize different buses to communicate with each other and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in the memory or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Equally, multiple computer devices can be connected, and each device provides part of the necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 11 A processor 10 is taken as an example.

[0102] The processor 10 may be a central processing unit, a network processor, or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic, or any combination thereof.

[0103] The memory 20 stores instructions that can be executed by at least one processor 10, so as to enable at least one processor 10 to execute the method shown in the above embodiment.

[0104] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created based on the use of a computer device for displaying a small program landing page, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0105] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0106] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0107] The embodiment of the present invention also provides a computer-readable storage medium. The above-mentioned method according to the embodiment of the present invention can be implemented in hardware, firmware, or implemented as a computer code that can be recorded in a storage medium, or implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memory. It can be understood that a computer, a processor, a microprocessor controller or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor or hardware, the method shown in the above embodiment is implemented.

[0108] A portion of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the form in which the computer program instruction exists in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc. Accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium that can be accessed by the computer.

[0109] Although the embodiments of the present invention have been described with reference to the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention. Such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A sagittal plane auditory localization method based on Bayesian reasoning, characterized in that: The method comprises: Obtaining the user's personalized head-related transfer function and positioning response data; Extract the spectrum features of the left and right ears respectively from the binaural input signals; According to the Bayesian criterion, cross-correlation calculation is performed on the spectral features and the feature templates at each polar angle on the target sagittal plane to determine the perceptual likelihood function at each polar angle, and the posterior probability distribution of each polar angle is determined by combining the perceptual likelihood function with the spatial prior distribution, where the feature templates are determined by the personalized head-related transfer function; Selecting a polar angle prediction value of a model from a posterior probability distribution based on a maximum a posteriori estimation criterion, wherein the model is a model constructed based on spectral feature extraction and Bayesian inference; Fitting model parameters to minimize the deviation between model predictions and true positioning responses under a preset threshold; Polar angle positioning is performed on binaural input signals based on a personalized fitting model, where the personalized fitting model is a model after parameter fitting.

2. The method according to claim 1, characterized in that Obtain the user's personalized head-related transfer function and positioning response data, including: Obtaining a user's personalized head-related transfer function based on personalized measurements or physiological parameter matching; A psychoacoustic localization experiment is performed to collect the user's localization response data in all acquisition directions on a spherical grid, and the user's personalized head-related transfer function is resampled to match the spatial resolution of the localization response data.

3. The method according to claim 1, characterized in that Extract the spectrum features of the left and right ears from the binaural input signals respectively, including: The initial feature vectors of binaural input signals are extracted based on a gammatone filter bank that approximates basilar membrane filtering and half-wave rectification of hair cell conduction. determining a spectrum amplitude profile feature based on the root mean square amplitude of the initial feature vector; The spectrum amplitude profile feature and the additive Gaussian noise are combined to determine a noisy feature, and the noisy feature is used as a spectrum feature.

4. The method according to claim 1, wherein According to the Bayesian criterion, the spectral features are cross-correlated with the feature templates of each polar angle direction on the target sagittal plane to determine the perceptual likelihood function in each polar angle direction. The perceptual likelihood function is combined with the spatial prior distribution to determine the posterior probability distribution of each polar angle, including: Calculating a mutual correlation coefficient based on the spectral features and the feature templates of each polar angle direction on the target sagittal plane, retaining the positive correlation coefficient and setting the non-positive correlation coefficient to zero, to obtain a unilateral correlation vector related to each polar angle direction, wherein the unilateral correlation vector includes a left-side correlation vector and a right-side correlation vector; A psychometric function is used to map the unilateral correlation vector into similarity vectors associated with each polar angle direction, wherein the similarity vectors include a left-side similarity vector and a right-side similarity vector, and a perceptual selectivity parameter and a perceptual sensitivity parameter of the psychometric function are fixed values or adaptively adjusted using an adaptive correlation-similarity mapping mechanism; The binaural weighting function is used to weight the sum of the left and right similarity vectors and normalize them to obtain the perceptual likelihood function related to each polar angle direction; The posterior probability distribution of each polar angle is determined by combining the perceptual likelihood function and the spatial prior distribution.

5. The method according to claim 4, characterized in that When the perceptual selectivity parameter and the perceptual sensitivity parameter are adaptively adjusted using an adaptive correlation-similarity mapping mechanism, a psychometric function is used to map the unilateral correlation vector into similarity vectors related to each polar angle direction, including: Arrange the one-sided correlation vectors of each polar angle direction in descending order, and determine the maximum value and the second peak value of the one-sided correlation vector; Performing linear regression on a preset psychometric function based on the maximum value of the unilateral correlation vector and the maximum value of the preset similarity vector, as well as the second peak value of the unilateral correlation vector and the minimum value of the preset similarity vector, to determine adaptive parameters, wherein the adaptive parameters include a perceptual selectivity parameter and a perceptual sensitivity parameter; The unilateral correlation vector is mapped into a similarity vector associated with each polar angle direction based on a psychometric function determined by an adaptive parameter.

6. The method according to claim 1, wherein The polar angle prediction value of the model is selected from the posterior probability distribution based on the maximum a posteriori estimation criterion, including: According to the maximum a posteriori estimation criterion, the direction with the maximum probability in the posterior distribution is selected as the polar angle prediction value of the target sound source, and response noise is introduced, and the response noise obeys the von Mises-Fisher distribution.

7. The method according to claim 6, characterized in that Fitting model parameters to minimize the deviation between model predictions and true positioning responses under a preset threshold includes: The polar angle prediction including the response noise is taken as the predicted value, and the polar angle direction in the positioning response data is taken as the true value; The preset parameters are adjusted so that the deviation between the predicted value and the true value meets the preset requirements. The preset parameters include the spectrum feature extraction stage, the Bayesian inference stage, and the free parameters in the response noise.

8. The method according to claim 7, characterized in that Adjust the preset parameters so that the deviation between the predicted value and the true value meets the preset requirements, including: Calculating polarity positioning performance indicators corresponding to the predicted value and the true value respectively, wherein the polarity positioning performance indicators include local polarity error and quadrant error rate, and the predicted value is a model prediction value corresponding to the preset parameters based on the current model parameters; Determining a prediction deviation between a predicted value and a true value based on the polarity positioning performance indicator, the prediction deviation including prediction deviations corresponding to a local polarity error and a quadrant error rate respectively; Determining a joint prediction deviation based on the prediction deviations corresponding to the local polarity error and the quadrant error rate; The initial parameters are adjusted based on the relationship between the joint prediction deviation and the joint prediction deviation threshold, and the polarity positioning performance index, prediction deviation, determination of the joint prediction deviation and parameter adjustment process are repeated until the relationship between the joint prediction deviation and the joint prediction deviation threshold meets the preset requirements. The joint prediction deviation threshold is determined based on the threshold of the polarity positioning performance index.

9. A sagittal plane auditory localization device based on Bayesian reasoning, characterized in that: The device comprises: A data acquisition module, used to obtain the user's personalized head-related transfer function and positioning response data; A feature extraction module is used to extract the spectrum features of the left and right ears from the binaural input signals respectively; a Bayesian inference module, configured to perform a cross-correlation calculation on the spectral features and a feature template at each polar angle on the target sagittal plane according to the Bayesian criterion, determine a perceptual likelihood function at each polar angle, and determine a posterior probability distribution at each polar angle by combining the perceptual likelihood function with a spatial prior distribution, wherein the feature template is determined by the personalized head-related transfer function; A perception decision module, configured to select a polar angle prediction value of a model from a posterior probability distribution based on a maximum a posteriori estimation criterion, wherein the model is constructed based on spectral feature extraction and Bayesian inference; A parameter fitting module, used to fit the model parameters to minimize the deviation between the model prediction and the true positioning response under a preset threshold; The direction prediction module is used to perform polar angle positioning on the binaural input signal based on a personalized fitting model, wherein the personalized fitting model is a model after parameter fitting.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the sagittal plane auditory localization method based on Bayesian reasoning according to any one of claims 1 to 8.

Images