Method and device for constructing OAE signal extraction model
By constructing an OAE signal extraction model with a multi-scale noise reduction module and a scaling dot product attention mechanism, the problem of OAE signal extraction under noise influence is solved, and automated and clean OAE signal extraction is achieved.
Patent Information
- Application Number
- CN202411297541.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-18
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-09-18
AI Technical Summary
Existing technologies are easily affected by environmental noise when extracting otoacoustic emissions (OAE) signals, especially under low signal-to-noise ratio conditions where the effectiveness drops significantly, and the operator's skill level affects the purity of the extracted signals.
A multi-scale noise reduction module is constructed to extract different frequency features from the sound signal through convolutional layers of different sizes, and outputs them by combining a scaling dot product attention mechanism to automatically extract the OAE signal.
It effectively suppresses noise, enhances the purity of OAE signals, and enables automated, highly robust signal extraction.
Smart Images

Figure CN118824279B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of signal processing, and in particular to a method and device for constructing an OAE signal extraction model. Background Art
[0002] OAE (otoacoustic emission) signal is a weak sound wave generated by the cochlea, which plays an important role in hearing detection and diagnosis. However, since OAE signal is very weak, it is easily affected by environmental noise. Existing technologies usually use traditional signal processing techniques such as bandpass filtering, waveform matching and correlation analysis to extract OAE signals. Although these methods are effective to a certain extent, they are also sensitive to noise, especially under low signal-to-noise ratio conditions. Therefore, the extraction effect will be significantly reduced when the noise level is high or the OAE signal is weaker than the average level. It is also unable to process some nonlinear signal characteristics. In addition, the existing technology needs to adjust a large number of parameters when in use, so it has high requirements on the operator, and the operator's technical level will also affect the purity of OAE signal extraction.
[0003] In summary, how to effectively suppress noise and automatically extract highly robust OAE signals is a problem that needs to be solved urgently in the existing technology. Summary of the invention
[0004] The embodiment of the present application provides a method for constructing an OAE signal extraction model. By constructing a multi-scale noise reduction module, convolutional layers of different sizes are used to extract and aggregate features in different frequency ranges in the sound signal, thereby achieving the effect of effective denoising and enhancing the OAE signal. The output is combined with a scaled dot product attention mechanism to extract a pure OAE signal. The constructed model does not require human intervention and can automatically extract the OAE signal.
[0005] In a first aspect, an embodiment of the present application provides a method for constructing an OAE signal extraction model, the method comprising:
[0006] Constructing a feature extraction module, wherein the feature extraction module is trained to extract features of a sound source signal marked with OAE signal intensity through a plurality of convolutional layers of the same size to obtain signal features;
[0007] Constructing a multi-scale denoising module, wherein the multi-scale denoising module is trained to respectively extract sub-signal features of different frequencies in the signal features by connecting a plurality of convolutional layers of different sizes in parallel, and aggregating the sub-signal features of different frequencies in a cascade manner to obtain denoising features;
[0008] Constructing a time series capture module, wherein the time series capture module is trained to capture the time dependency in the noise reduction features to obtain the time series features;
[0009] Constructing a result output module, wherein the result output module is trained to output the temporal features as an OAE signal through a scaled dot product attention calculation;
[0010] Among them, the OAE signal extraction model includes a feature extraction module, a multi-scale noise reduction module, a timing capture module and a result output module. The trained OAE signal extraction model is used to extract the OAE signal from the sound source signal to be extracted, and the sound source signal to be extracted is a sound source signal without marking the OAE signal intensity.
[0011] In a second aspect, an embodiment of the present application provides a device for constructing an OAE signal extraction model, comprising:
[0012] A feature extraction module, wherein the feature extraction module is trained to extract features of a sound source signal marked with OAE signal strength through a plurality of convolutional layers of the same size to obtain signal features;
[0013] A multi-scale noise reduction module, wherein the multi-scale noise reduction module is trained to respectively extract sub-signal features of different frequencies in the signal features by connecting a plurality of convolutional layers of different sizes in parallel, and aggregate the sub-signal features of different frequencies in a cascade manner to obtain noise reduction features;
[0014] A time series capture module, wherein the time series capture module is trained to capture the time dependency in the noise reduction features to obtain time series features;
[0015] A result output module is provided, wherein the result output module is trained to output the time series features as OAE signals through scaled dot product attention calculation, wherein the OAE signal extraction model includes a feature extraction module, a multi-scale noise reduction module, a time series capture module and a result output module, and the trained OAE signal extraction model is used to extract the OAE signal from the sound source signal to be extracted, wherein the sound source signal to be extracted is a sound source signal without labeled OAE signal intensity.
[0016] In a third aspect, an embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a method for constructing an OAE signal extraction model.
[0017] In a fourth aspect, an embodiment of the present application provides a readable storage medium, in which a computer program is stored. The computer program includes a program code for controlling a process to execute a process, and the process includes a method for constructing an OAE signal extraction model.
[0018] The main contributions and innovations of the present invention are as follows:
[0019] The embodiment of the present application uses a larger convolution layer in the feature extraction module to perform feature extraction, which can better capture the continuous long-term information in the sound source signal, make the feature extraction more accurate, and help the model learn useful information from the extracted features; this solution uses convolution layers of different sizes to extract and aggregate features in different frequency ranges in the sound signal by constructing a multi-scale denoising module, thereby achieving the effect of effective denoising and enhancing the OAE signal; and in order to avoid the gradient explosion caused by the multi-scale denoising module, this solution sets a residual block and a depth-separable convolution layer after the multi-scale denoising module to ensure the normal amount of model parameters; the output module in this solution outputs through the scaled dot product attention mechanism to extract a pure OAE signal, and the constructed model does not require human intervention, and the OAE signal can be automatically extracted.
[0020] Details of one or more embodiments of the present application are set forth in the following drawings and description to make other features, objects, and advantages of the present application more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0022] Figure 1 is a flow chart of a method for constructing an OAE signal extraction model according to an embodiment of the present application;
[0023] Figure 2 is a structural principle diagram of an OAE signal extraction model according to an embodiment of the present application;
[0024] Figure 3 is a structural diagram of a residual block implemented according to the present application;
[0025] Figure 4 It is a structural block diagram of a device for constructing an OAE signal extraction model according to an embodiment of the present application;
[0026] Figure 5 It is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION
[0027] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with one or more embodiments of this specification. Instead, they are merely examples of devices and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.
[0028] It should be noted that: in other embodiments, the steps of the corresponding method are not necessarily performed in the order shown and described in this specification. In some other embodiments, the steps included in the method may be more or less than those described in this specification. In addition, a single step described in this specification may be decomposed into multiple steps for description in other embodiments; and multiple steps described in this specification may be combined into a single step for description in other embodiments.
[0029] Embodiment 1:
[0030] The embodiment of the present application provides a method for constructing an OAE signal extraction model, which uses convolutional layers of different sizes to extract features in different frequency ranges in sound signals and aggregates them by constructing a multi-scale noise reduction module, thereby achieving the effect of effectively denoising and enhancing OAE signals, and combining the scaled dot product attention mechanism for output to extract pure OAE signals. The constructed model does not require manual participation, and can automatically extract OAE signals. For details, refer to Figure 1 as well as Figure 2 , the method comprising:
[0031] Constructing a feature extraction module, wherein the feature extraction module is trained to extract features of a sound source signal marked with OAE signal intensity through a plurality of convolutional layers of the same size to obtain signal features;
[0032] Constructing a multi-scale denoising module, wherein the multi-scale denoising module is trained to respectively extract sub-signal features of different frequencies in the signal features by connecting a plurality of convolutional layers of different sizes in parallel, and aggregating the sub-signal features of different frequencies in a cascade manner to obtain denoising features;
[0033] Constructing a time series capture module, wherein the time series capture module is trained to capture the time dependency in the noise reduction features to obtain the time series features;
[0034] Constructing a result output module, wherein the result output module is trained to output the temporal features as an OAE signal through a scaled dot product attention calculation;
[0035] Among them, the OAE signal extraction model includes a feature extraction module, a multi-scale noise reduction module, a timing capture module and a result output module. The trained OAE signal extraction model is used to extract the OAE signal from the sound source signal to be extracted, and the sound source signal to be extracted is a sound source signal without marking the OAE signal intensity.
[0036] In some specific embodiments, the original signal with a signal-to-noise ratio ranging from 0 to 50 db is obtained by using a professional OAE signal acquisition device to collect signals from multiple subjects in a quiet environment and artificially adding random white noise.
[0037] Furthermore, a bandpass filter is used to filter out irrelevant frequency components in the original signal, and normalized to have a uniform amplitude range to obtain a sound source signal, and then the OAE signal intensity in the sound source signal is labeled to obtain a sound source signal marked with the OAE signal intensity as training data.
[0038] In this scheme, in order to improve the generalization ability of the OAE signal extraction model, the training data is enhanced by using time expansion, frequency transformation and other methods, and the enhanced training data is used to perform supervised learning on each module in the OAE signal extraction model.
[0039] In this solution, the size of each convolutional layer in the feature extraction module is 16×1 and the step size is 2.
[0040] Furthermore, the number of convolutional layers in the feature extraction module is 32, that is, this solution uses 32 convolutional layers of size 16×1 and step size 2 in series to extract signal features.
[0041] Specifically, in a conventional feature extraction module, the size of the convolution layer is generally selected to be 3×3 or 5×5. This is because in the feature extraction process, it is necessary to ensure faster computing efficiency and obtain more information. However, since this scheme performs feature extraction on the sound source signal including the OAE signal, and the time continuity of the sound source signal in this scheme is very long, if the conventional 3×3 or 5×5 convolution is used for feature extraction, the continuous long-term information in the sound source signal cannot be captured. Therefore, this scheme uses a 16×1 convolution layer to capture features, thereby better capturing the continuous long-term information in the sound source signal.
[0042] Moreover, since this scheme uses a 16×1 convolutional layer, the amount of calculation will be very large when too many convolutional layers are stacked, so this scheme sets the convolution step size to 2 to reduce the input size of subsequent convolutional layers, thereby reducing the amount of calculation of the feature extraction module.
[0043] In summary, this scheme sets the convolution layer to 16×1 size and a step size of 2 to help the feature extraction module establish a feature hierarchy from local to global.
[0044] In this scheme, the signal features are linearly activated and then input into the multi-scale denoising module. The multi-scale denoising module is composed of a first convolutional layer, a second convolutional layer, and a third convolutional layer in parallel. The first convolutional layer, the second convolutional layer, and the third convolutional layer are respectively used to extract sub-signal features of different frequencies in the signal features, wherein the first convolutional layer is a 1×1 convolution, the second convolutional layer is a 3×1 convolution, and the third convolutional layer is a 5×1 convolution.
[0045] Specifically, since the signal characteristics include different frequency components such as noise and OAE signals, and the OAE signal is a weak sound signal, the multi-scale denoising module in this scheme extracts these weak signals in different frequencies through 1×1 convolutional layers, 3×1 convolutional layers, and 5×1 convolutional layers, and composes them in a cascade manner to highlight the OAE signals in the signal characteristics and reduce the influence of background noise signals on the OAE signals.
[0046] Specifically, the number of filters in the first convolutional layer, the second convolutional layer, and the third convolutional layer is 32.
[0047] In this scheme, sub-signal features of different frequencies are cascaded in the order of the first convolution layer, the second convolution layer, and the third convolution layer.
[0048] Furthermore, a maximum pooling layer is additionally added to the multi-scale denoising module, and the maximum pooling layer is connected in parallel with the first convolution layer, the second convolution layer and the third convolution layer. The maximum pooling layer is used to perform maximum pooling on the signal features to obtain a pooling result, and the pooling result is aggregated with the sub-signal features of different frequencies in a cascade manner to obtain the denoising feature.
[0049] Specifically, adding an additional maximum pooling layer in the multi-scale denoising module can increase the robustness of the OAE signal extraction model to information of different scales.
[0050] Specifically, the first convolution layer, the second convolution layer, the third convolution layer, and the maximum pooling layer are cascaded in this order.
[0051] Specifically, the pooling kernel size of the maximum pooling layer is 3×1, and the pooling stride is 1.
[0052] In this solution, a residual block is constructed after the multi-scale denoising module to process the denoising features to obtain residual features, and then a depth-wise separable convolution is used to process the residual features and input them into the timing capture module.
[0053] In this solution, the structure of the residual block is as follows: Figure 3 As shown, the residual block is connected in series by a first residual convolution layer, a first residual activation function layer, a second residual convolution layer and a second residual activation function layer. That is to say, the denoising features are sequentially processed by the first residual convolution layer, the first residual activation function layer, the second residual convolution layer and the second residual activation function layer, and then feature splicing with the denoising features to complete the residual connection.
[0054] Specifically, constructing a residual block after the multi-scale denoising module to process the denoising features can alleviate the problem of gradient vanishing or gradient exploding that may be caused by the depth of the network layer, thereby improving the stability of the entire OAE signal extraction model.
[0055] In some specific embodiments, the first residual convolution layer and the second residual convolution layer use 3×1 convolution, the number of filters is 16, and the first residual activation function layer and the second residual activation function layer use ReLU activation function.
[0056] Furthermore, a pooling layer is used to process the residual features before inputting them into the depthwise separable convolution.
[0057] Specifically, the depthwise separable convolution processes the input features in a 1×1 point-by-point convolution manner, and the number of filters of the depthwise separable convolution is 64.
[0058] Specifically, this scheme reduces the number of feature parameters through depthwise separable convolution, and because point-by-point convolution allows the network to recombine feature maps between multiple channels, it also enhances the expressiveness of the features, allowing the model to learn features more flexibly for better OAE signal extraction.
[0059] In this solution, the timing capture module is a bidirectional LSTM layer.
[0060] Furthermore, the bidirectional LSTM layer in this scheme adopts a bidirectional LSTM with 64 input channels, 64 hidden units, and 64 channels of output features.
[0061] Specifically, the timing capture module can capture time series information in two directions, from front to back and from back to front, respectively, thereby helping the OAE signal extraction model to understand the time dependency in the sound source signal and improve the accuracy of OAE signal extraction.
[0062] In this scheme, the output module includes a scaled dot product attention calculation unit, a deconvolution unit and an output unit connected in series. The scaled dot product attention calculation unit is used to perform a scaled dot product attention calculation on the time series feature to obtain an attention result. The deconvolution unit is used to adjust the dimension of the attention result to be the same as the dimension of the input sound source signal. The output unit outputs the attention result after dimension adjustment to obtain an OAE signal.
[0063] Furthermore, the scaled dot product attention is achieved by calculating the dot product between the query and the key, applying a scaling factor, converting these values into probabilities using the softmax function, and finally multiplying these probabilities with the value to obtain a weighted sum. The specific steps are as follows:
[0064] 1. Calculate dot product: Given a query matrix `Q` and a key matrix `K`, calculate their dot product. If the dimensions of `Q` and `K` are `[batch_size,seq_len_q,d]` and `[batch_size,seq_len_k,d]` respectively, the result of the dot product will be a matrix of `[batch_size,seq_len_q,seq_len_k]`.
[0065] 2. Scaling: Divide the dot product result by the square root of the dimension of the key (`d`). This is to prevent the dot product from becoming too large at large `d` values, causing the gradient of the softmax function to disappear.
[0066] 3. Softmax: Apply the softmax function to the scaled dot product so that the result at each query position will be a probability distribution over the key position.
[0067] 4. Weighted sum: Multiply the output of softmax by the value matrix `V` to get the result of weighted sum. The dimension of the value matrix `V` is usually `[batch_size, seq_len_v, d]`.
[0068] Specifically, the number of input layer channels d_model of the scaled dot product attention used in this scheme is 64, and the total number of parameters including the bias is 3*d_model^2+3*d_model, which is 12480.
[0069] Furthermore, the deconvolution unit used in this scheme is a 16×1 convolution layer with a step size of 2.
[0070] Furthermore, the output unit is a Sigmoid activation function, which integrates all features and predicts the OAE signal through a fully connected layer.
[0071] In summary, the output dimension and parameter quantity of each layer in the OAE signal extraction model constructed in this scheme are shown in Table 1.
[0072] Table 1 Output dimensions and parameter quantities of each layer in the OAE signal extraction model:
[0073]
[0074] In this scheme, the collected sound source signals are divided into training set, validation set and test set to complete the training of the OAE signal extraction model, where the ratio of training set, validation set and test set is 6:2:2.
[0075] In some specific embodiments, a deep learning framework such as TensorFlow or PyTorch is used to construct an OAE signal extraction model. During the training process, any combination of one or more loss functions can be selected, and parameters such as learning rate and weight decay can be set during the training process, and optimization can be performed using optimizers such as Adam or AdamW.
[0076] In some specific embodiments, when training the OAE signal extraction model, after obtaining the OAE signal output by the model through forward propagation, the extracted OAE signal is compared with the labeled OAE signal intensity to calculate the loss, and the gradient of the OAE signal extraction model is calculated through back propagation to update the network weights. When the loss function meets the set conditions, the performance of the temporarily trained OAE signal extraction model is evaluated on the validation set. When the performance of the OAE signal extraction model on the validation set no longer improves, the training is stopped to prevent the OAE signal extraction model from overfitting, or the training of the OAE signal extraction model is stopped when the number of iterative training meets the set number.
[0077] In some specific embodiments, when the training of the OAE signal extraction model is completed, the OAE signal extraction model is exported to a format that can be used for reasoning, such as SavedModel of TensorFlow or torchscript of PyTorch.
[0078] Embodiment 2:
[0079] Based on the same idea, refer to Figure 4 , the present application also proposes a construction device for an OAE signal extraction model, comprising:
[0080] A feature extraction module, wherein the feature extraction module is trained to extract features of a sound source signal marked with OAE signal strength through a plurality of convolutional layers of the same size to obtain signal features;
[0081] A multi-scale noise reduction module, wherein the multi-scale noise reduction module is trained to respectively extract sub-signal features of different frequencies in the signal features by connecting a plurality of convolutional layers of different sizes in parallel, and aggregate the sub-signal features of different frequencies in a cascade manner to obtain noise reduction features;
[0082] A time series capture module, wherein the time series capture module is trained to capture the time dependency in the noise reduction features to obtain time series features;
[0083] A result output module is provided, wherein the result output module is trained to output the time series features as OAE signals through scaled dot product attention calculation, wherein the OAE signal extraction model includes a feature extraction module, a multi-scale noise reduction module, a time series capture module and a result output module, and the trained OAE signal extraction model is used to extract the OAE signal from the sound source signal to be extracted, wherein the sound source signal to be extracted is a sound source signal without labeled OAE signal intensity.
[0084] Embodiment three:
[0085] This embodiment also provides an electronic device, referring to Figure 5 , comprises a memory 404 and a processor 402, wherein the memory 404 stores a computer program, and the processor 402 is configured to run the computer program to execute the steps in any of the above method embodiments.
[0086] Specifically, the processor 402 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.
[0087] Among them, the memory 404 may include a large capacity memory 404 for data or instructions. For example, but not limitation, the memory 404 may include a hard disk drive (HDD), a floppy disk drive, a solid state drive (SSD), a flash memory, an optical disk, a magneto-optical disk, a tape, or a universal serial bus (USB) drive, or a combination of two or more of these. In appropriate cases, the memory 404 may include a removable or non-removable (or fixed) medium. In appropriate cases, the memory 404 may be inside or outside the data processing device. In a specific embodiment, the memory 404 is a non-volatile memory. In a specific embodiment, the memory 404 includes a read-only memory (ROM) and a random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (Programmable Read-Only Memory, PROM for short), an erasable PROM (Erasable Programmable Read-Only Memory, EPROM for short), an electrically erasable PROM (Electrically Erasable Programmable Read-Only Memory, EEPROM for short), an electrically alterable ROM (Electrically Alterable Read-Only Memory, EAROM for short) or a flash memory (FLASH) or a combination of two or more of these. In appropriate circumstances, the RAM may be a static random access memory (SRAM) or a dynamic random access memory (DRAM), wherein the DRAM may be a fast page mode dynamic random access memory 404 (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.
[0088] The memory 404 may be used to store or cache various data files required for processing and / or communication, as well as possible computer program instructions executed by the processor 402 .
[0089] The processor 402 reads and executes the computer program instructions stored in the memory 404 to implement any one of the methods for constructing the OAE signal extraction model in the above embodiments.
[0090] Optionally, the electronic device may further include a transmission device 406 and an input / output device 408 , wherein the transmission device 406 is connected to the processor 402 , and the input / output device 408 is connected to the processor 402 .
[0091] The transmission device 406 can be used to receive or send data via a network. The specific examples of the above-mentioned network may include a wired or wireless network provided by a communication provider of the electronic device. In one example, the transmission device includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 406 can be a radio frequency (Radio Frequency, referred to as RF) module, which is used to communicate with the Internet wirelessly.
[0092] The input / output device 408 is used to input or output information. In this embodiment, the input information may be a sound source signal, etc., and the output information may be an OAE signal, etc.
[0093] Optionally, in this embodiment, the processor 402 may be configured to perform the following steps through a computer program:
[0094] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementation modes, and this embodiment will not be described in detail here.
[0095] In general, various embodiments may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. Some aspects of the invention may be implemented in hardware, while other aspects may be implemented in firmware or software that may be executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flow charts, or using some other graphical representation, it should be understood that, as non-limiting examples, the boxes, devices, systems, techniques, or methods described herein may be implemented in hardware, software, firmware, dedicated circuits or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.
[0096] Embodiments of the present invention can be implemented by computer software, which is executable by a data processor of a mobile device, such as in a processor entity, or implemented by hardware, or implemented by a combination of software and hardware. Computer software or programs (also referred to as program products) including software routines, applets and / or macros can be stored in any device readable data storage medium, and they include program instructions for performing specific tasks. Computer program products can include one or more computer executable components configured to perform embodiments when the program is running. One or more computer executable components can be at least one software code or a part thereof. In addition, at this point, it should be noted that any box of the logic flow in the figure can represent a program step, or interconnected logic circuits, boxes and functions, or a combination of program steps and logic circuits, boxes and functions. Software can be stored in physical media such as memory chips or storage blocks implemented in processors, magnetic media such as hard disks or floppy disks, and optical media such as, for example, DVDs and data variants thereof, CDs. Physical media are non-transient media.
[0097] Those skilled in the art should understand that the technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0098] The above embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. A method for constructing an OAE signal extraction model, characterized in that: The following steps are involved: Constructing a feature extraction module, wherein the feature extraction module is trained to extract features of a sound source signal marked with OAE signal intensity through multiple convolutional layers of the same size to obtain signal features; Constructing a multi-scale noise reduction module, wherein the multi-scale noise reduction module is trained to extract sub-signal features of different frequencies from the signal features by connecting multiple convolutional layers of different sizes in parallel, and aggregating the sub-signal features of different frequencies in a cascade manner to obtain noise reduction features; Constructing a time series capture module, wherein the time series capture module is trained to capture the time dependency in the noise reduction features to obtain time series features; Constructing a result output module, wherein the result output module is trained to output the temporal features as an OAE signal through scaled dot product attention calculation; Among them, the OAE signal extraction model includes a feature extraction module, a multi-scale noise reduction module, a time series capture module and a result output module. The trained OAE signal extraction model is used to extract the OAE signal from the sound source signal to be extracted, and the sound source signal to be extracted is a sound source signal without labeled OAE signal intensity.
2. The method for constructing an OAE signal extraction model according to claim 1, wherein: The size of each convolutional layer in the feature extraction module is 16×1, and the stride is 2.
3. The method for constructing an OAE signal extraction model according to claim 1, wherein: The multi-scale noise reduction module is composed of a first convolutional layer, a second convolutional layer, and a third convolutional layer connected in parallel. The first convolutional layer, the second convolutional layer, and the third convolutional layer are respectively used to extract sub-signal features of different frequencies in the signal features, wherein the first convolutional layer is a 1×1 convolution, the second convolutional layer is a 3×1 convolution, and the third convolutional layer is a 5×1 convolution.
4. The method for constructing an OAE signal extraction model according to claim 3, wherein: An additional maximum pooling layer is added to the multi-scale denoising module. The maximum pooling layer is connected in parallel with the first convolutional layer, the second convolutional layer, and the third convolutional layer. The maximum pooling layer is used to perform maximum pooling on the signal features to obtain a pooling result, and the pooling result is aggregated with the sub-signal features of different frequencies in a cascade manner to obtain the denoising feature.
5. The method for constructing an OAE signal extraction model according to claim 1, wherein: A residual block is constructed after the multi-scale denoising module to process the denoising features to obtain residual features, and then a depth-wise separable convolution is used to process the residual features and input them into the timing capture module.
6. The method for constructing an OAE signal extraction model according to claim 1, wherein: The timing capture module is a bidirectional LSTM layer.
7. The method for constructing an OAE signal extraction model according to claim 1, characterized in that: The output module includes a scaled dot product attention calculation unit, a deconvolution unit and an output unit. The scaled dot product attention calculation unit is used to perform scaled dot product attention calculation on the time series feature to obtain an attention result. The deconvolution unit is used to adjust the dimension of the attention result to be the same as the dimension of the input sound source signal. The output unit uses an activation function for output.
8. A device for constructing an OAE signal extraction model, characterized in that: include: A feature extraction module is trained to extract features of a sound source signal marked with OAE signal strength through multiple convolutional layers of the same size to obtain signal features; A multi-scale noise reduction module, which is trained to extract sub-signal features of different frequencies from the signal features by connecting multiple convolutional layers of different sizes in parallel, and aggregate the sub-signal features of different frequencies in a cascade manner to obtain noise reduction features; A time series capture module is trained to capture the time dependency in the noise reduction features to obtain time series features; A result output module is trained to output the time series features as OAE signals through scaled dot product attention calculation, wherein the OAE signal extraction model includes a feature extraction module, a multi-scale noise reduction module, a time series capture module and a result output module. The trained OAE signal extraction model is used to extract the OAE signal from the sound source signal to be extracted, and the sound source signal to be extracted is a sound source signal without labeled OAE signal intensity.
9. An electronic device comprising a memory and a processor, characterized in that: The memory stores a computer program, and the processor is configured to run the computer program to execute the method for constructing an OAE signal extraction model according to any one of claims 1 to 7.
10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which includes a program code for controlling a process to execute a process, wherein the process includes a method for constructing an OAE signal extraction model according to any one of claims 1 to 7.
