CN110457646A - One kind being based on parameter transfer learning low-resource head-position difficult labor personalized method - Google Patents
One kind being based on parameter transfer learning low-resource head-position difficult labor personalized method Download PDFInfo
- Publication number
- CN110457646A CN110457646A CN201910560696.6A CN201910560696A CN110457646A CN 110457646 A CN110457646 A CN 110457646A CN 201910560696 A CN201910560696 A CN 201910560696A CN 110457646 A CN110457646 A CN 110457646A
- Authority
- CN
- China
- Prior art keywords
- head
- transfer function
- related transfer
- module
- personalized
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims description 47
- 238000013526 transfer learning Methods 0.000 title description 9
- 230000006870 function Effects 0.000 claims abstract description 178
- 238000012546 transfer Methods 0.000 claims abstract description 138
- 238000012549 training Methods 0.000 claims abstract description 67
- 238000013508 migration Methods 0.000 claims abstract description 64
- 230000005012 migration Effects 0.000 claims abstract description 64
- 238000013507 mapping Methods 0.000 claims abstract description 4
- 238000005314 correlation function Methods 0.000 claims abstract 3
- 238000007781 pre-processing Methods 0.000 claims description 24
- 238000013461 design Methods 0.000 claims description 11
- 238000002360 preparation method Methods 0.000 claims description 10
- 230000005540 biological transmission Effects 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 abstract description 2
- 230000003044 adaptive effect Effects 0.000 abstract 1
- 238000013459 approach Methods 0.000 abstract 1
- 230000002596 correlated effect Effects 0.000 abstract 1
- 230000000875 corresponding effect Effects 0.000 abstract 1
- 210000003128 head Anatomy 0.000 description 97
- 238000010586 diagram Methods 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 238000000605 extraction Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 210000005069 ears Anatomy 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 238000001228 spectrum Methods 0.000 description 2
- 230000001174 ascending effect Effects 0.000 description 1
- 238000012076 audiometry Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 210000003454 tympanic membrane Anatomy 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/15—Correlation function computation including computation of convolution operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Mathematical Optimization (AREA)
- Computational Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Pure & Applied Mathematics (AREA)
- Computing Systems (AREA)
- Mathematical Analysis (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- Medical Informatics (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Machine Translation (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The present invention relates to field of signal processing, disclose a kind of personalized adaptive approach of low-resource head-position difficult labor, solve the technical issues of accurately obtaining target object personalization HRTF.Feature generation module generates different spatial and is in head-position difficult labor correlated characteristic, with reference to personalized head-position difficult labor model training module based on generating Nonlinear Mapping between feature of the reference head related correlation functions library at different object different spatials and corresponding personalized related transfer function, parameter migration models training module carries out parameter migration to reference personalization related transfer function model based on the low-resource personalization related transfer function database of target object, generate the head-position difficult labor personalized model of target object, the total space personalization head-position difficult labor of personalized correlation function prediction module prediction target object.
Description
Technical Field
The invention relates to the field of signal processing, in particular to a low-resource head-related transfer function personalization method based on parameter migration learning.
Background
The explosion in the field of virtual reality has made virtual hearing more and more interesting. Virtual reality includes virtual vision, virtual hearing, virtual touch, virtual taste, etc., wherein the important issue of virtual hearing technology is to restore the same spatial location features as natural hearing. The human auditory process can be generally considered as a source-channel-receiving model, in which the channel contains the diffraction and interference of the sound source through different parts of the human body and finally reaches the tympanic membrane, and can be regarded as a spatial digital filter called Head-Related Transfer Function (HRTF) which contains all the spectral features caused by the interaction between the sound waves and the body parts. Since the physiological structure of each person is different, HRTF spectral features are extremely personalized, and therefore, it is difficult to measure HRTFs in full space for each individual.
At present, there are many HRTF (head related transfer function) personalized methods, and theoretical or mathematical modeling is used for modeling and analyzing a human body, such as a spherical head model, a snowman model, a structural model, a boundary element method, a finite difference time domain method and the like. However, these methods require expensive hardware to perform complex calculations. Therefore, some low complexity methods are proposed. The method for gradually determining parameters to have linear modeling through audiometry experiments based on the perception method needs a large-scale database for matching so as to obtain the HRTF most suitable for the target object, and therefore, the time consumption is long. In consideration of the dependency relationship between the HRTF and the human physiological parameters, the physiological parameter-based regression method is becoming more common in predicting personalized HRTFs, however, most of these methods have the assumption that the physiological parameter weight is equal to the HRTF weight. On the other hand, the HRTF full-space estimation method from a small data measurement set is also a method for HRTF personalization, however, most of the existing methods only obtain coefficients of a linear prediction model from the small data measurement set and then expand the coefficients to the full space, and the method has a large error in a high frequency band (>10 kHz).
Therefore, a method with low complexity and capable of obtaining the personalized HRTF of the target object more accurately is needed.
Disclosure of Invention
In view of the above problems, an object of the present invention is to provide a method for obtaining a target object personalized HRTF with low complexity and accuracy, which has the advantage of obtaining a target object personalized HRTF with more accuracy.
The above object of the present invention is achieved by the following technical solutions:
a low-resource head-related transfer function personalization method based on parameter migration learning is characterized by comprising a feature generation module, a reference personalized related transfer function model training module, a parameter migration model training module and a personalized head-related function prediction module;
the feature generation module generates head-related transfer function related features of different spatial positions, the reference personalized head-related transfer function model training module generates nonlinear mapping between the features of different spatial positions of different objects and corresponding personalized related transfer functions based on a reference head-related function library, the parameter migration model training module performs parameter migration on the reference personalized related transfer function model based on a low-resource personalized related transfer function database of the target object to generate a head-related transfer function personalized model of the target object, and the personalized related function prediction module predicts a full-space personalized head-related transfer function of the target object.
Further, the generation of the feature generation module comprises a direction feature module, a distance feature module, a feature merging module and a feature preprocessing module;
the feature merging module merges the direction features and the distance features and then performs feature preprocessing, wherein the feature preprocessing normalizes the merged features, the mean value is 0, and the variance is 1.
Further, based on a reference head related transfer function database, extracting a log-amplitude minimum phase head related transfer function, then preprocessing the minimum phase head related transfer function, and enabling the preprocessed data to enter a reference personalized related transfer function model training module;
the reference personalized head-related transfer function model training module is connected with the feature generation module, the head-related transfer function preprocessing module and the loss function design module based on expert domain knowledge and used for model training based on a deep neural network to obtain a reference personalized head-related transfer function model;
the loss function design module based on expert domain knowledge obtains a loss function in a training process of a reference personalized related transfer function model.
Further, the parameter migration model training module comprises a feature generation module, a reference personalized head related transfer function model training module, a low resource database preparation module and a model training module based on parameter migration learning;
the parameter migration model training module carries out parameter migration on the reference personalized head-related transfer function model based on the low-resource personalized head-related transfer function database of the target object to generate a head-related transfer function personalized model of the target object;
the low resource data preparation module generates parameter migration model training data;
the model training module based on parameter migration learning is used for matching the reference personalized head-related transfer function model migration with the head-related transfer function data of the target object to obtain a personalized head-related transfer function generation model.
Further, the personalized head-related transfer function prediction module comprises a minimum phase head-related transfer function prediction module and a head-related transfer function reconstruction module based on the spatial position, and predicts the full-space personalized head-related transfer function of the target object;
the space position-based minimum phase head related transfer function prediction module predicts a log-amplitude minimum phase head related transfer function of a target object at a target space position;
the head-related transfer function reconstruction module predicts a log-amplitude minimum phase head-related transfer function to reconstruct an individualized head-related transfer function.
Compared with the prior art, the invention has the beneficial effects that:
(1) the parameter migration learning method provided by the invention can migrate the reference personalized HRTF model to the target object, so that more accurate estimation can be obtained;
(2) the invention obtains more accurate individualized HRTF of the target object by different parameter migration methods;
(3) the individualized HRTF modeling method obtains the individualized HRTF in the whole space by referring to the individualized HRTF model and utilizing the small data test sample of the target object, has high robustness and is convenient to apply in the actual environment.
Drawings
FIG. 1 is a schematic structural diagram of a parameter transfer learning-based low-resource head-related transfer function personalization method according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a feature generation module 1 for implementing a parameter transfer learning-based low-resource head-related transfer function personalization method of parameter transfer learning-based according to the present invention;
FIG. 3 is a schematic structural diagram of a reference personalized head-related transfer function model training module 2 for implementing a parameter transfer learning-based low-resource head-related transfer function personalization method of parameter transfer learning-based according to the present invention;
fig. 4 is a schematic structural diagram of a parameter migration model training module 3 implementing a reference personalized head-related transfer function model training module according to the present invention.
Reference numerals: 1. a feature generation module; 2. a reference individualized head-related transfer function model training module; 3. A parameter migration model training module; 4. a personalized head-related transfer function prediction module; 11. a direction feature generation module; 12. a distance feature generation module; 13. a feature merging module; 14. a feature preprocessing module; 21. a log-amplitude minimum phase HRTF extraction module; 22. a preprocessing module of head-related transfer functions; 23. a loss function design module based on expert domain knowledge; 31. a low resource data preparation module; 32. a model training module based on parameter transfer learning; 41. a minimum phase head related transfer function prediction module based on the spatial position; 42. and a head related transfer function reconstruction module.
Detailed Description
The invention is described in detail below with reference to the figures and examples.
It should be noted that in the drawings or description, the same drawing reference numerals are used for similar or identical parts. And are indicated in the drawings for simplicity and convenience. Furthermore, implementations not shown or described in the drawings are of a form known to those of ordinary skill in the art. Additionally, while exemplifications of parameters including particular values may be provided herein, it is to be understood that the parameters need not be exactly equal to the respective values, but may be approximated to the respective values within acceptable error margins or design constraints.
The invention provides a parameter migration learning-based low-resource head related transfer function personalization method, which is based on parameter migration learning.
The parameter migration learning method can migrate the reference personalized HRTF model to the target object to obtain more accurate estimation, and obtains more accurate personalized HRTF of the target object through different parameter migration methods.
Example one
A low resource head related transfer function personalization method based on parameter migration learning is disclosed, as shown in figure 1, and comprises a feature generation module 1, a reference personalized related transfer function model training module, a parameter migration model training module 3 and a personalized head related function prediction module;
the feature generation module 1 generates head-related transfer function related features of different spatial positions, the reference personalized head-related transfer function model training module 2 generates nonlinear mapping between the features of different spatial positions of different objects and corresponding personalized related transfer functions based on a reference head-related transfer function library, the parameter migration model training module 3 performs parameter migration on the reference personalized related transfer function model based on a low-resource personalized related transfer function database of the target object to generate a head-related transfer function personalized model of the target object, and the personalized related function prediction module predicts a full-space personalized head-related transfer function of the target object.
As shown in fig. 2, further, the generation of the feature generation module 1 includes a direction feature module, a distance feature module, a feature merging module 13, and a feature preprocessing module 14. The feature merging module 13 merges the direction feature and the distance feature and performs feature preprocessing, where the feature preprocessing normalizes the merged features, and the mean value is 0 and the variance is 1.
As shown in fig. 3, further, based on the reference head-related transfer function database, extracting a log-amplitude minimum phase head-related transfer function, then preprocessing the minimum phase head-related transfer function, and entering the preprocessed data into a reference personalized related transfer function model training module;
the reference personalized head related transfer function model training module 2 is connected with the feature generation module 1, the head related transfer function preprocessing module 22 and the loss function design module 23 based on expert domain knowledge, and is used for model training based on a deep neural network to obtain a reference personalized head related transfer function model;
the loss function design module 23 based on expert domain knowledge obtains a loss function in the training process of the reference personalized correlation transfer function model.
As shown in fig. 4, further, the parameter migration model training module 3 includes a feature generation module 1, a reference personalized head-related transfer function model training module 2, a low resource database preparation module, and a model training module 32 based on parameter migration learning;
the parameter migration model training module 3 performs parameter migration on the reference personalized head-related transfer function model based on the low-resource personalized head-related transfer function database of the target object to generate a head-related transfer function personalized model of the target object;
the low-resource data preparation module 31 generates parameter migration model training data;
the model training module 32 based on parameter migration learning performs migration on the reference personalized head-related transfer function model and matches the head-related transfer function data of the target object to obtain a personalized head-related transfer function generation model.
Further, the personalized head-related transfer function prediction module 4 includes a minimum phase head-related transfer function prediction module 41 and a head-related transfer function reconstruction module 42 based on spatial positions, and predicts a full-space personalized head-related transfer function of the target object;
the spatial position-based minimum phase head related transfer function prediction module 41 predicts a log-amplitude minimum phase head related transfer function of the target object at the target spatial position;
head-related transfer function reconstruction module 42 predicts a log-amplitude minimum phase head-related transfer function reconstruction personalized head-related transfer function.
Example two
On the basis of the first embodiment, more specifically, the direction feature generation module 11 is a feature generation module 1, and is configured to generate a direction-related feature for a current location. The sound field transmission response from the sound source to both ears is a complex function of frequency, distance, azimuth, elevation, and the sound field can be represented in a specific set of orthogonal sequences. The direction-related features are generated by adopting spherical harmonics, wherein the spherical harmonics are functions of azimuth angles and elevation angles and are defined as formulas (1) and (2).
Wherein N is the degree of the legendre function, and N is 0. m is the order of Legendre function, | m | is less than or equal to n;is a Legendre function with the degree of n and the order of m; theta and phi are the azimuth and elevation angles, respectively, of the measured position.
Distance feature generation module 12 the feature generation module 1 is arranged to generate distance-related features at the current location. The distance-related features are generated by adopting a standard spherical Bessel function and defined as
Wherein jl(x) Is a spherical Bessel function with an order of l,Jl'(x) Is a Bessel function. N is a radical ofnlIn order to normalize the factors, the method comprises the following steps of,xlnis jl(x) 0 n-th ascending positive root. k is a radical ofnl＝xnlWhere a is the wave number and a is the maximum radius. And r is the distance from the current sound source position to the center of the head.
And the feature merging module 13 is connected to the direction feature generating module 11, the feature generating module 1 and the distance feature generating module 12, and is configured to merge features related to direction and distance at any spatial position, and use the merged features as an input of the feature preprocessing module 14. For the position d (r, theta, phi), the direction and distance features are combined to obtain the input feature setAssuming that the degree of the Legendre function is N and the order of the spherical Bessel function is L, the generated feature set contains N in total for each position dt＝[(N+1)2+NL]A characteristic parameter.
And the feature preprocessing module 14 is connected with the feature merging module 13 and the reference personalized head related transfer function model training module 2, and is used for preprocessing the merged features, normalizing the input features within the values of 0 mean value and 1 variance, and using the output of the module as the input of the reference personalized head related transfer function model training. For the ith item in the feature set at the s-th position, the preprocessing procedure is expressed as
Wherein,andrespectively, the mean and standard deviation of the ith feature at all positions. S is the number of spatial measurement positions of the data in the reference HRTF database.
And the log-amplitude minimum-phase HRTF extraction module 21 is connected with the feature generation module 1 and is used for extracting the log-amplitude minimum-phase head-related transfer function and preparing data for the output of the reference personalized head-related transfer function model training module 2. HRTF H for ith bin at s positions(i) The calculation process of the logarithmic amplitude minimum phase HRTF is
The preprocessing module 22 of the head-related transfer function is connected to the log-amplitude minimum-phase HRTF extracting module 21, and is configured to preprocess the minimum-phase HRTF and use the preprocessed minimum-phase HRTF as an output of the reference personalized head-related transfer function model training module 2. The purpose of the HRTF preprocessing is to normalize the input features to a value with a mean of 0 and a variance of 1, reducing the floating range of the data. Log-amplitude minimum phase HRTF for ith bin at s positionThe pretreatment process is expressed as
Wherein,andmean value and standard of ith frequency point respectively representing HRTF at all positionsAnd (4) tolerance. N is a radical offThe number of frequency points used for model training.
And the loss function design module based on expert domain knowledge is used for obtaining a loss function used in the training process of the reference personalized head related transfer function model. The design basis is subjective perception domain knowledge. Since the log-amplitude spectrum retains all perceptually relevant information for the human ear, the loss function is defined based on the log-spectrum distortion criterion as
Wherein k is1And k2The frequency band representing the contrast is from the kth1Band to kth2The frequency band is generally 20Hz to 20kHz according to the hearing range of human ears. N is a radical offIs k1To k2The number of frequency bins in between. S is the number of measurement locations used for model training.And the normalized logarithmic amplitude minimum phase HRTF represents the ith frequency point at the s position predicted by the reference personalized head related transfer function model. By minimizing the loss function, the objective performance of the model can be maximized.
And the reference personalized head-related transfer function model training module is connected with the feature generation module, the head-related transfer function preprocessing module and the loss function design module based on expert domain knowledge and is used for model training based on the deep neural network to obtain a reference personalized head-related transfer function model.
And the low-resource data preparation module is connected with the feature generation module and the reference personalized head related transfer function model training module and is used for generating data for parameter migration model training. The parameter migration model training adopts a low-resource personalized head related transfer function database of a target object, and the database comprises a total S measured on the target objecttHRTF at each position, wherein St< S. For measured StA spaceAnd (4) firstly, generating the characteristics of the parameter migration model training module through the characteristic generation module, wherein the data is used as the input of the parameter migration model. Then, for the HRTF corresponding to the target function at the s-th position, obtaining the HRTF after the target object is preprocessed by adopting a log-amplitude minimum-phase HRTF extraction module in a reference personalized head-related transfer function model training module and a preprocessing module of the head-related transfer function, and taking the data as the output of a parameter migration model.
And the model training module based on parameter migration learning is connected with the data preparation module and the reference personalized head related transfer function model training module and is used for carrying out model migration on the reference personalized head related transfer function model so as to enable the model migration to be matched with HRTF data of a target object. Because the HRTF measuring positions of the target object are less, three parameter migration methods are adopted, which respectively correspond to the following steps: the method comprises the following steps of first hidden layer parameter migration, middle hidden layer parameter migration and last hidden layer parameter migration. And in the model training process based on parameter migration learning, all node parameters except the first hidden layer of the fixed reference personalized HRTF model are unchanged, and the first hidden layer node parameters in the reference personalized HRTF model are updated by using input features and output data obtained by the low-resource data preparation module. Similarly, the intermediate hidden layer parameter migration and the last hidden layer parameter migration only update the intermediate hidden layer node parameter and the last hidden layer node parameter, respectively.
And the head-related transfer function reconstruction module is connected with the minimum phase head-related transfer function prediction module based on the space position and used for reconstructing the personalized head-related transfer function through the predicted log amplitude minimum phase head-related transfer function. For target position dsAnd performing log-amplitude minimum phase HRTF de-normalization on the output of the minimum phase head-related transfer function prediction module based on the spatial position, wherein the log-amplitude minimum phase HRTF de-normalization is calculated as follows:
then, after the logarithm is changed into linear and inverse Hilbert transform, the reconstructed HRTF is obtained.
The parameter transfer learning-based low-resource head-related transfer function personalization method based on parameter transfer learning can be written by Matlab and c languages or any other programming languages. Furthermore, the present invention may be applied to a computer terminal, a handheld mobile device, or other forms of mobile devices.
The above description is only a preferred embodiment of the present invention, and the protection scope of the present invention is not limited to the above embodiments, and all technical solutions belonging to the idea of the present invention belong to the protection scope of the present invention. It should be noted that modifications and embellishments within the scope of the invention may occur to those skilled in the art without departing from the principle of the invention, and are considered to be within the scope of the invention.
Claims (5)
1. A low-resource head-related transfer function personalization method based on parameter migration learning is characterized by comprising a feature generation module (1), a reference personalized related transfer function model training module, a parameter migration model training module (3) and a personalized head-related function prediction module;
the characteristic generation module (1) generates head-related transmission function related characteristics at different spatial positions, the reference personalized head-related transmission function model training module (2) generates nonlinear mapping between the characteristics at different spatial positions of different objects and corresponding personalized related transmission functions based on a reference head-related correlation function library, the parameter migration model training module (3) performs parameter migration on the reference personalized related transmission function model based on a low-resource personalized related transmission function database of the target object to generate a head-related transmission function personalized model of the target object, and the personalized related function prediction module predicts a full-space personalized head-related transmission function of the target object.
2. The method for personalizing a low-resource head-related transfer function based on parameter migration learning according to claim 1, wherein the generation of the feature generation module (1) comprises a direction feature module, a distance feature module, a feature merging module (13) and a feature preprocessing module (14);
the characteristic merging module (13) merges the direction characteristic and the distance characteristic and then carries out characteristic preprocessing, the characteristic preprocessing normalizes the merged characteristic, the mean value is 0, and the variance is 1.
3. The parameter migration learning-based low-resource head-related transfer function personalization method according to claim 2, characterized in that a log-amplitude minimum-phase head-related transfer function is extracted based on a reference head-related transfer function database, then the minimum-phase head-related transfer function is preprocessed, and the preprocessed data enter a reference personalization-related transfer function model training module;
a reference personalized head-related transfer function model training module (2) which is connected with the feature generation module (1), the head-related transfer function preprocessing module (22) and the loss function design module (23) based on expert domain knowledge and is used for model training based on a deep neural network to obtain a reference personalized head-related transfer function model;
the expert knowledge based loss function design module (23) obtains a loss function in a reference personalized correlation transfer function model training process.
4. The method for personalizing a low-resource head-related transfer function based on parameter migration learning according to claim 3, wherein the parameter migration model training module (3) comprises a feature generation module (1), a reference personalized head-related transfer function model training module (2), a low-resource database preparation module and a model training module (32) based on parameter migration learning;
the parameter migration model training module (3) performs parameter migration on the reference personalized head-related transfer function model based on the low-resource personalized head-related transfer function database of the target object to generate a head-related transfer function personalized model of the target object;
the low-resource data preparation module (31) generates parameter migration model training data;
the model training module (32) based on parameter migration learning is used for matching the reference personalized head-related transfer function model migration with the head-related transfer function data of the target object to obtain a personalized head-related transfer function generation model.
5. The method for personalizing a low-resource head-related transfer function based on parameter migration learning according to claim 4, wherein the personalized head-related transfer function prediction module (4) comprises a minimum phase head-related transfer function prediction module (41) based on spatial position and a head-related transfer function reconstruction module (42) for predicting a full-space personalized head-related transfer function of the target object;
the spatial position-based minimum phase head related transfer function prediction module (41) predicts a log-amplitude minimum phase head related transfer function of the target object at the target spatial position;
the head-related transfer function reconstruction module (42) predicts a log-amplitude minimum-phase head-related transfer function to reconstruct an individualized head-related transfer function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910560696.6A CN110457646B (en) | 2019-06-26 | 2019-06-26 | Low-resource head-related transfer function personalization method based on parameter migration learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910560696.6A CN110457646B (en) | 2019-06-26 | 2019-06-26 | Low-resource head-related transfer function personalization method based on parameter migration learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110457646A true CN110457646A (en) | 2019-11-15 |
| CN110457646B CN110457646B (en) | 2022-12-13 |
Family
ID=68481073
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910560696.6A Active CN110457646B (en) | 2019-06-26 | 2019-06-26 | Low-resource head-related transfer function personalization method based on parameter migration learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110457646B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111246363A (en) * | 2020-01-08 | 2020-06-05 | 华南理工大学 | A method and device for virtual sound customization based on auditory matching |
| CN112328676A (en) * | 2020-11-27 | 2021-02-05 | 江汉大学 | Method for estimating personalized head-related transfer function and related equipment |
| CN113806679A (en) * | 2021-09-13 | 2021-12-17 | 中国政法大学 | Head-related transfer function personalization method based on pre-training model |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9648438B1 (en) * | 2015-12-16 | 2017-05-09 | Oculus Vr, Llc | Head-related transfer function recording using positional tracking |
| CN107480100A (en) * | 2017-07-04 | 2017-12-15 | 中国科学院自动化研究所 | Head-position difficult labor modeling based on deep-neural-network intermediate layer feature |
| CN108038291A (en) * | 2017-12-05 | 2018-05-15 | 武汉大学 | A kind of personalized head related transfer function generation system and method based on human parameters adaptation algorithm |
| US10028070B1 (en) * | 2017-03-06 | 2018-07-17 | Microsoft Technology Licensing, Llc | Systems and methods for HRTF personalization |
| CN108549907A (en) * | 2018-04-11 | 2018-09-18 | 武汉大学 | A kind of data verification method based on multi-source transfer learning |
| CN109145360A (en) * | 2018-06-29 | 2019-01-04 | 中国科学院自动化研究所 | Head-position difficult labor personalization modeling based on sparse constraint |
-
2019
- 2019-06-26 CN CN201910560696.6A patent/CN110457646B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9648438B1 (en) * | 2015-12-16 | 2017-05-09 | Oculus Vr, Llc | Head-related transfer function recording using positional tracking |
| US10028070B1 (en) * | 2017-03-06 | 2018-07-17 | Microsoft Technology Licensing, Llc | Systems and methods for HRTF personalization |
| CN107480100A (en) * | 2017-07-04 | 2017-12-15 | 中国科学院自动化研究所 | Head-position difficult labor modeling based on deep-neural-network intermediate layer feature |
| CN108038291A (en) * | 2017-12-05 | 2018-05-15 | 武汉大学 | A kind of personalized head related transfer function generation system and method based on human parameters adaptation algorithm |
| CN108549907A (en) * | 2018-04-11 | 2018-09-18 | 武汉大学 | A kind of data verification method based on multi-source transfer learning |
| CN109145360A (en) * | 2018-06-29 | 2019-01-04 | 中国科学院自动化研究所 | Head-position difficult labor personalization modeling based on sparse constraint |
Non-Patent Citations (1)
| Title |
|---|
| 黄婉秋等: "基于多维生理参数的头相关传递函数个人化方法", 《西北工业大学学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111246363A (en) * | 2020-01-08 | 2020-06-05 | 华南理工大学 | A method and device for virtual sound customization based on auditory matching |
| CN111246363B (en) * | 2020-01-08 | 2021-07-20 | 华南理工大学 | A method and device for virtual sound customization based on auditory matching |
| CN112328676A (en) * | 2020-11-27 | 2021-02-05 | 江汉大学 | Method for estimating personalized head-related transfer function and related equipment |
| CN113806679A (en) * | 2021-09-13 | 2021-12-17 | 中国政法大学 | Head-related transfer function personalization method based on pre-training model |
| CN113806679B (en) * | 2021-09-13 | 2024-05-28 | 中国政法大学 | Head related transfer function individuation method based on pre-training model |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110457646B (en) | 2022-12-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10607358B2 (en) | Ear shape analysis method, ear shape analysis device, and ear shape model generation method | |
| CN107480100B (en) | A head-related transfer function modeling system based on features of the middle layers of deep neural networks | |
| Komorowski et al. | The use of continuous wavelet transform based on the fast fourier transform in the analysis of multi-channel electrogastrography recordings | |
| CN110457646B (en) | Low-resource head-related transfer function personalization method based on parameter migration learning | |
| Ahrens et al. | HRTF magnitude modeling using a non-regularized least-squares fit of spherical harmonics coefficients on incomplete data | |
| CN106412793B (en) | The sparse modeling method and system of head-position difficult labor based on spheric harmonic function | |
| US11863959B2 (en) | Personalized three-dimensional audio | |
| US20230336936A1 (en) | Modeling of the head-related impulse responses | |
| Tashev | HRTF phase synthesis via sparse representation of anthropometric features | |
| CN109145360B (en) | A personalized modeling system for head-related transfer functions based on sparsity constraints | |
| Zhang et al. | Efficient continuous HRTF model using data independent basis functions: Experimentally guided approach | |
| Lu et al. | Personalization of head-related transfer function based on sparse principle component analysis and sparse representation of 3D anthropometric parameters | |
| Xi et al. | Magnitude modelling of individualized HRTFs using DNN based spherical harmonic analysis | |
| EP4327569A1 (en) | Error correction of head-related filters | |
| Zhao et al. | Efficient prediction of individual head-related transfer functions based on 3D meshes | |
| CN113806679B (en) | Head related transfer function individuation method based on pre-training model | |
| CN113038356A (en) | Personalized HRTF rapid modeling acquisition method | |
| CN114067832B (en) | Prediction method and device of head related transfer function and electronic equipment | |
| CN107301153B (en) | A Head-Related Transfer Function Modeling Method Based on Adaptive Fourier Decomposition | |
| Wang et al. | Prediction of head-related transfer function based on tensor completion | |
| EP3352481B1 (en) | Ear shape analysis device and ear shape analysis method | |
| CN117409799B (en) | Audio signal processing system and method | |
| Rönkkö | Measuring acoustic intensity field in upscaled physical model of ear | |
| Qi et al. | Parameter-Transfer Learning for Low-Resource Individualization of Head-Related Transfer Functions. | |
| Qi et al. | Sparsity-Constrained Weight Mapping for Head-Related Transfer Functions Individualization from Anthropometric Features. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |