Digital audio electrical signal optimization processing method

Abstract

The invention provides a digital audio electric signal optimization processing method, and relates to the technical field of signal processing. The method comprises the following specific steps: collecting and manufacturing a digital sound electric signal data set and carrying out preprocessing, constructing a space processing module, constructing a time sequence processing module, constructing an output processing module, constructing a digital sound electric signal processing model, and training and detecting the digital sound electric signal processing model. Meanwhile, a digital audio electrical signal processing model is provided, a space processing module adopts depth separable convolution to extract channel features and reduce calculation overhead, and an important area in an enhanced signal is adaptively selected in combination with a space attention mechanism; the time sequence processing module focuses on extracting the time sequence characteristics of the digital audio electric signals through a plurality of time sequence processing blocks, and the time sequence characteristic characterization capability is improved; and the output processing module combines a self-attention mechanism and a full connection layer, dynamically pays attention to key features and carries out final signal reconstruction, so that the quality of the output digital audio electrical signal is improved.

Description

Technical Field

[0001] The invention belongs to the technical field of signal processing, and in particular relates to a method for optimizing processing of digital audio electrical signals. Background Art

[0002] With the rapid development of digital audio technology, the application scenarios of digital audio signals are becoming more and more extensive, including music production, communication, voice recognition, smart home and other fields; however, audio signals are often affected by various noises and distortions during the collection, transmission and processing, resulting in reduced signal clarity and quality; for example, background noise, electromagnetic interference and equipment noise can all interfere with audio signals, especially in complex environments, the impact of noise is more obvious; the above problems will not only affect the auditory experience, but also interfere with the accuracy of intelligent systems such as voice recognition.

[0003] At present, commonly used audio signal processing methods mainly rely on traditional filtering, frequency analysis, and linear prediction technologies. The above methods are effective in removing specific types of noise, but they have poor adaptability to complex noise environments and are difficult to effectively solve the interference of non-stationary noise or multi-source noise. In addition, traditional signal processing methods are also insufficient in maintaining the original quality of the signal, which can easily lead to audio distortion or loss of details and cannot meet the needs of high-fidelity audio processing.

[0004] With the development of deep learning technology, audio signal processing methods based on neural networks have gradually attracted attention. Neural networks can automatically identify and remove noise by automatically learning the complex features in audio signals, while enhancing the clarity of signals. However, current deep learning models still have problems such as complex model structure, high computational overhead, and insufficient real-time processing capabilities, which limit their promotion in practical applications. Therefore, how to design an efficient digital audio electrical signal processing method that can not only improve signal quality in complex noise environments, but also process efficiently under limited computing resources is still a difficult problem that needs to be solved urgently. Summary of the invention

[0005] The present invention provides a digital audio electrical signal optimization processing method, aiming to propose a digital audio electrical signal processing model, wherein a spatial processing module introduces a deep separable convolution to extract local features of each channel, significantly reduces the computational overhead while maintaining the high efficiency of the model, and combines a spatial attention mechanism to adaptively select and strengthen important areas in the signal and ignore unimportant information; a timing processing module introduces a gated recurrent unit GRU to capture the long-term dependency and dynamic changes of timing signals and accurately process long timing signals, a timing processing block focuses on extracting timing features of digital audio electrical signals, and average pooling compresses redundant information by downsampling to improve the timing feature representation capability; an output processing module combines a self-attention mechanism and a fully connected layer to dynamically focus on key parts of the input signal, extract the global features of the signal and perform final signal reconstruction to ensure the quality of the output digital audio electrical signal.

[0006] In order to achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for optimizing processing of digital audio electrical signals, the specific steps of which are as follows.

[0007] S1. Collect and preprocess digital audio signal data sets.

[0008] S2, build a spatial processing module, including depth-wise separable convolution, batch normalization, spatial attention and Activation function.

[0009] S3. Build a timing processing module, including a timing processing block.

[0010] S4. Build the output processing module, including the self-attention mechanism, the fully connected layer and the Softmax function.

[0011] S5. Construct a digital audio electrical signal processing model, including input, spatial processing module, timing processing module, output processing module and output.

[0012] S6. Training and testing of digital audio and electrical signal processing model: using the digital audio and electrical signal data set to train the digital audio and electrical signal processing model. After the training is completed, the digital audio and electrical signal that needs to be denoised and optimized is obtained and input into the digital audio and electrical signal processing model to obtain the processing result of the input signal.

[0013] Preferably, in step S1, digital audio electrical signal data is collected to make a data set, which is divided into a training set and a test set in a ratio of 8 to 2, and a filter is used to preprocess the digital audio electrical signal data set to preliminarily reduce noise interference in the signal.

[0014] Preferably, in step S2, for the spatial processing module, the input is a digital audio electrical signal ,in, represents the number of samples, Indicates the number of channels. The digital audio signal is first processed by depthwise separable convolution and then by batch normalization to obtain the intermediate output , ,in, stands for depthwise separable convolution, Represents batch normalization, and the intermediate output is then passed through the spatial attention mechanism to obtain the output of spatial attention , ,in, represents the convolutional layer, represent Activation function, Represents a splicing operation, represents average pooling, Represents the maximum pooling, and the final output is passed through Activation function processing to obtain the final output of the spatial processing module , ,in, represent Activation function.

[0015] Preferably, in step S2, for the spatial processing module, spatial features are extracted from the input digital audio electrical signal to help the model identify important information in the digital audio electrical signal and suppress noise: first, local features of each channel are extracted through deep separable convolution, which significantly reduces the computational overhead while maintaining the high efficiency of the model; then, the batch normalization operation standardizes the convolved data to eliminate the deviation between different batches of data, stabilize the training process and accelerate model convergence; then, through the spatial attention mechanism, the model adaptively selects and strengthens important areas in the signal and ignores unimportant information and noise; the spatial processing module makes feature extraction more efficient and targeted, which helps to achieve higher quality digital audio electrical signal optimization.

[0016] Preferably, in step S3, for the temporal processing module, the input is the output of the spatial processing module , the input is passed through Gated Recurrent Unit , random dropout layer and the time series processing block composed of residual connection to obtain signal characteristics , , , where the random inactivation layer randomly inactivates neurons at a ratio of 0.5, and then Signal characteristics of a timing processing block Then downsample through average pooling to get the final output of the timing processing module , ,in, represents average pooling.

[0017] Preferably, in step S3, for the timing processing module, the time features of the digital audio electrical signal are deeply extracted and analyzed, and the long-short time dependencies in the digital audio electrical signal are captured through a series of timing processing blocks composed of gated recurrent units, random inactivation layers and residual connections, so as to effectively process the dynamic changes in the digital audio electrical signal; GRU selectively remembers or forgets the timing information through the gate mechanism to ensure that the model can accurately process the long time series digital audio electrical signal; the random inactivation layer avoids overfitting of the model by randomly inhibiting a part of the neurons, while improving the generalization ability of the model; the residual connection helps to avoid information loss, so that key information remains intact in the deep network; the timing processing module can fully capture the time dependency in the digital audio electrical signal, make the model sensitive to dynamic changes, deeply mine and optimize the time features, and at the same time ensure that the model has good robustness and generalization ability when facing complex data, thereby enhancing the quality of the digital audio electrical signal.

[0018] Preferably, in step S4, for the output processing module, the input is the signal feature after being processed by the timing processing module. The input signal features are first calculated through the self-attention mechanism to obtain the weighted features , , , , , ,in, Is generated The weight matrix of Is generated The weight matrix of Is generated The weight matrix of the attention weight score is then calculated. , ,in, is the scaling factor, Express Transpose the matrix, yes Function, and finally get the output of self-attention , The features output by the self-attention mechanism are then passed through two fully connected layers to extract the global features of the digital audio electrical signal. The output of the fully connected layer is passed through Function to get the final output result , ,in, It is a fully connected layer.

[0019] Preferably, in step S4, for the output processing module, the signal features from the timing processing module are further processed and optimized to generate the final digital audio electrical signal output; the self-attention mechanism is used to emphasize the key information in the digital audio electrical signal by calculating the weighted features, and reduce the influence of noise and irrelevant features; the self-attention mechanism enables the model to dynamically focus on different parts of the input features and assign corresponding weights according to their importance; through two fully connected layers, the global features of the digital audio electrical signal are extracted, and the final signal reconstruction is performed to ensure the quality of the output digital audio electrical signal; the design of the output processing module can reduce information loss and ensure the high fidelity of the final output processing result.

[0020] Preferably, in step S5, for the digital audio electrical signal processing model, the digital audio electrical signal data is first input into the model, and the input digital audio electrical signal first passes through a spatial processing module to extract the spatial features of the signal, separate the useful features and noise in the signal while selectively suppressing the noise, and then passes through a timing processing module to extract the time features of the digital audio electrical signal, enhance and retain useful time information, and finally passes through an output processing module for calculation and processing to enhance the overall expression ability of the signal, while filtering out redundant information, and after the processing is completed, the final digital audio electrical signal processing model is obtained.

[0021] Compared with the prior art, the present invention has the following technical effects: the technical solution provided by the present invention proposes a digital audio electrical signal processing model, wherein the spatial processing module introduces a deep separable convolution to extract the local features of each channel, significantly reducing the computational overhead while maintaining the high efficiency of the model, and combining the spatial attention mechanism to adaptively select and strengthen important areas in the signal and ignore unimportant information; the timing processing module introduces a gated recurrent unit GRU to capture the long-term dependency and dynamic changes of the timing signal and accurately process the long timing signal. The timing processing block focuses on extracting the timing features of the digital audio electrical signal, and average pooling compresses redundant information by downsampling, thereby improving the timing feature representation capability; the output processing module combines the self-attention mechanism and the fully connected layer, dynamically focuses on the key parts of the input signal, extracts the global features of the signal and performs the final signal reconstruction to ensure the quality of the output digital audio electrical signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The present invention provides a flowchart of a digital audio electrical signal optimization processing method.

[0023] Figure 2 It is a structural diagram of the space processing module provided by the present invention.

[0024] Figure 3 It is a structural diagram of the timing processing module provided by the present invention.

[0025] Figure 4 It is a structural diagram of the output processing module provided by the present invention.

[0026] Figure 5 It is an effect diagram of the training and testing process of the digital audio electrical signal processing model provided by the present invention. DETAILED DESCRIPTION

[0027] The present invention aims to propose a method for optimizing the processing of digital audio electrical signals, and proposes a digital audio electrical signal processing model, wherein the spatial processing module introduces a deep separable convolution to extract the local features of each channel, significantly reducing the computational overhead while maintaining the high efficiency of the model, and combining the spatial attention mechanism to adaptively select and strengthen important areas in the signal and ignore unimportant information; the timing processing module introduces a gated recurrent unit GRU to capture the long-term dependency and dynamic changes of the timing signal, and accurately processes the long timing signal. The timing processing block focuses on extracting the timing features of the digital audio electrical signal, and the average pooling compresses the redundant information by downsampling, thereby improving the timing feature representation capability; the output processing module combines the self-attention mechanism and the fully connected layer, dynamically pays attention to the key parts of the input signal, extracts the global features of the signal, and performs the final signal reconstruction to ensure the quality of the output digital audio electrical signal.

[0028] See also Figure 1 As shown, a digital audio electrical signal optimization processing method in an embodiment of the present application has the following specific steps.

[0029] S1. Collect and preprocess digital audio signal data sets.

[0030] Furthermore, in step S1, digital audio electrical signal data is collected to make a data set, which is divided into a training set and a test set in a ratio of 8 to 2, and a filter is used to preprocess the digital audio electrical signal data set to preliminarily reduce noise interference in the signal.

[0031] S2, build a spatial processing module, including depth-wise separable convolution, batch normalization, spatial attention and Activation function.

[0032] Furthermore, in step S2, for the spatial processing module, its structure is as follows: Figure 2 As shown, the input is a digital audio signal ,in, represents the number of samples, Indicates the number of channels. The digital audio signal is first processed by depthwise separable convolution and then by batch normalization to obtain the intermediate output , ,in, stands for depthwise separable convolution, Represents batch normalization, and the intermediate output is then passed through the spatial attention mechanism to obtain the output of spatial attention , ,in, represents the convolutional layer, represent Activation function, Represents a splicing operation, represents average pooling, Represents the maximum pooling, and the final output is passed through Activation function processing to obtain the final output of the spatial processing module , ,in, represent Activation function.

[0033] S3. Build a timing processing module, including a timing processing block.

[0034] Furthermore, in step S3, for the timing processing module, its structure is as follows: Figure 3 As shown, the input is the output of the spatial processing module , the input is passed through Gated Recurrent Unit , random dropout layer and the time series processing block composed of residual connection to obtain signal characteristics , , , where the random inactivation layer randomly inactivates neurons at a ratio of 0.5, and then Signal characteristics of a timing processing block Then downsample through average pooling to get the final output of the timing processing module , ,in, represents average pooling. In this embodiment, the number of timing processing blocks is set to 4.

[0035] S4. Build the output processing module, including the self-attention mechanism, the fully connected layer and the Softmax function.

[0036] Further, in step S4, for the output processing module, its structure is as follows: Figure 4 As shown, the input is the signal characteristics after being processed by the timing processing module The input signal features are first calculated through the self-attention mechanism to obtain the weighted features , , , , , ,in, Is generated The weight matrix of Is generated The weight matrix of Is generated The weight matrix of the attention weight score is then calculated. , ,in, is the scaling factor, Express Transpose the matrix, yes Function, and finally get the output of self-attention , The features output by the self-attention mechanism are then passed through two fully connected layers to extract the global features of the digital audio electrical signal. The output of the fully connected layer is passed through Function to get the final output result , ,in, It is a fully connected layer.

[0037] S5. Construct a digital audio electrical signal processing model, including input, spatial processing module, timing processing module, output processing module and output.

[0038] Further, in step S5, in the step S5, for the digital audio electrical signal processing model, the digital audio electrical signal data is first input into the model, and the input digital audio electrical signal first passes through the spatial processing module to extract the spatial characteristics of the signal, separate the useful characteristics and noise in the signal while selectively suppressing the noise, and then passes through the timing processing module to extract the time characteristics of the digital audio electrical signal, enhance and retain useful time information, and finally passes through the output processing module for calculation and processing to enhance the overall expression ability of the signal, and filter out redundant information, and after the processing is completed, the final digital audio electrical signal processing model is obtained.

[0039] S6. Training and testing of digital audio electrical signal processing model. Use the digital audio electrical signal data set to train the digital audio electrical signal processing model. After the training is completed, obtain the digital audio electrical signal that needs to be denoised and optimized, input it into the digital audio electrical signal processing model, and obtain the processing optimization result of the input signal.

[0040] Furthermore, in step S5, for the digital audio electrical signal processing model, code is written in the application Pycharm based on the Python language, and the Pytorch framework is used to input the digital audio electrical signal data into the digital audio electrical signal processing model. The Adam optimizer is used, the learning rate is set to 0.001, the training batch size is 128, the model is trained 100 times from scratch, the input digital audio electrical signal is trained, the training loss rate is output, and the final trained digital audio electrical signal processing model is obtained.

[0041] Furthermore, in step S6, the effect diagram of the training and testing process of the digital audio electrical signal processing model is as follows: Figure 5 As shown, Figure 5 It shows a significant decrease in the loss rate during training, up to 0.05. Although the initial loss rate of the test process is higher than the training loss rate, the test loss rate gradually decreases with the increase in the number of iterations, and finally drops below 0.1. The changes in training and testing further prove that the proposed digital acoustic electrical signal processing model has efficient processing capabilities and superior performance. Overall, the change in loss rate reflects the high efficiency of the model in optimizing digital acoustic electrical signal processing, and effectively realizes the optimization of digital acoustic electrical signals.

[0042] The above are only preferred embodiments of the present invention. It should be pointed out that a person skilled in the art can make several modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A method for optimizing processing of digital audio electrical signals, characterized in that: The specific steps include: S1, collect and prepare digital audio signal data set and perform preprocessing; S2, build a spatial processing module, including depth-wise separable convolution, batch normalization, spatial attention, and activation functions; S3, constructing a timing processing module, including a timing processing block; S4, construct the output processing module, including the self-attention mechanism, the fully connected layer and the Softmax function; S5, constructing a digital audio electrical signal processing model, including input, spatial processing module, timing processing module, output processing module and output; S6. Training and testing of digital audio and electrical signal processing model: using the digital audio and electrical signal data set to train the digital audio and electrical signal processing model. After the training is completed, the digital audio and electrical signal that needs to be denoised and optimized is obtained and input into the digital audio and electrical signal processing model to obtain the processing result of the input signal.

2. A digital audio electrical signal optimization processing method according to claim 1, characterized in that: In the step S1, digital audio electrical signal data is collected to make a data set, which is divided into a training set and a test set in a ratio of 8 to 2, and a filter is used to preprocess the digital audio electrical signal data set to preliminarily reduce noise interference in the signal.

3. A digital audio electrical signal optimization processing method according to claim 2, characterized in that: In the step S2, for the spatial processing module, the input is a digital audio electrical signal ,in, represents the number of samples, Indicates the number of channels. The digital audio signal is first processed by depthwise separable convolution and then by batch normalization to obtain the intermediate output , ,in, stands for depthwise separable convolution, Represents batch normalization, and the intermediate output is then passed through the spatial attention mechanism to obtain the output of spatial attention , ,in, represents the convolutional layer, represent Activation function, Represents a splicing operation, represents average pooling, Represents the maximum pooling, and the final output is passed through Activation function processing to obtain the final output of the spatial processing module , ,in, represent Activation function.

4. A digital audio electrical signal optimization processing method according to claim 3, characterized in that: In step S3, for the temporal processing module, the input is the output of the spatial processing module. , the input is passed through Gated Recurrent Unit , random dropout layer and the time series processing block composed of residual connection to obtain signal characteristics , , , where the random inactivation layer randomly inactivates neurons at a ratio of 0.5, and then Signal characteristics of a timing processing block Then downsample through average pooling to get the final output of the timing processing module , ,in, represents average pooling.

5. A digital audio electrical signal optimization processing method according to claim 4, characterized in that: In the step S4, for the output processing module, the input is the signal feature after being processed by the timing processing module. The input signal features are first calculated through the self-attention mechanism to obtain the weighted features , , , , , ,in, Is generated The weight matrix of Is generated The weight matrix of Is generated The weight matrix of the attention weight score is then calculated. , ,in, is the scaling factor, Express Transpose the matrix, yes Function, and finally get the output of self-attention , The features output by the self-attention mechanism are then passed through two fully connected layers to extract the global features of the digital audio electrical signal. The output of the fully connected layer is passed through Function to get the final output result , ,in, It is a fully connected layer.

6. A digital audio electrical signal optimization processing method according to claim 5, characterized in that: In the step S5, for the digital audio electrical signal processing model, the digital audio electrical signal data is first input into the model, and the input digital audio electrical signal first passes through the spatial processing module to extract the spatial characteristics of the signal, separate the useful characteristics and noise in the signal, and selectively suppress the noise, and then passes through the timing processing module to extract the time characteristics of the digital audio electrical signal, enhance and retain the useful time information, and finally passes through the output processing module for calculation and processing to enhance the overall expression ability of the signal, and filter out redundant information, and after the processing is completed, the final digital audio electrical signal processing model is obtained.

Images