Voice recognition method based on flexible optical fiber coupler, controller and storage medium

Abstract

The invention relates to the technical field of voice recognition, in particular to a voice recognition method based on a flexible optical fiber coupler, a controller and a storage medium, and the voice recognition method comprises the steps: obtaining an original voice signal, and carrying out the feature extraction of the original voice signal, and obtaining text data corresponding to the original voice signal; preprocessing the text data to obtain a first spectrum image corresponding to the text data; performing regularization processing on the first frequency spectrum image to obtain a second frequency spectrum image; and inputting the second frequency spectrum image into a Resnet deep learning model for feature extraction and prediction classification processing to obtain a sound analysis result corresponding to the second frequency spectrum image. According to the invention, the sensitivity of acquiring the sound signal is improved, the accuracy of the sound signal recognition result is improved, and the user experience is improved.

Description

Technical Field

[0001] The present application relates to the technical field of sound recognition, and in particular to a sound recognition method, a controller and a storage medium based on a flexible optical fiber coupler. Background Art

[0002] Sound is an important medium for information dissemination in daily life and a very important way of communication in daily life. As people age, their hearing ability will inevitably decline to a certain extent. Currently, hearing aids are mostly used to solve hearing problems of the elderly. However, the size and weight of hearing aids limit the user's usage time. The hearing loss of each user is different, and the damage to the left and right ears is different, so it takes a long time to debug.

[0003] In the related technology, most hearing aids are electronic sensors. In an electromagnetic field environment, traditional electronic sensors may fail due to electromagnetic interference, resulting in low sensitivity of the hearing aid in receiving sound signals and poor performance in processing sound signals. The accuracy of the sound analysis results output by the hearing aid is low, and users cannot accurately and clearly obtain sounds, resulting in a poor user experience. Summary of the invention

[0004] The present application aims to solve at least one of the technical problems existing in the prior art. To this end, the present application embodiment provides a sound recognition method, controller and storage medium based on a flexible optical fiber coupler, which is conducive to improving the sensitivity of obtaining sound signals, improving the accuracy of sound signal recognition results, and improving user experience.

[0005] In a first aspect, an embodiment of the present application provides a sound recognition method based on a flexible optical fiber coupler, comprising:

[0006] Acquire an original sound signal, and perform feature extraction on the original sound signal to obtain text data corresponding to the original sound signal;

[0007] Preprocessing the text data to obtain a first spectrum image corresponding to the text data;

[0008] Performing regularization processing on the first spectrum image to obtain a second spectrum image;

[0009] The second spectrum image is input into the Resnet deep learning model for feature extraction and prediction classification processing to obtain a sound analysis result corresponding to the second spectrum image.

[0010] The technical solution of the first aspect of the present application has at least one of the following advantages or beneficial effects: obtaining the original sound signal collected by the flexible optical fiber coupler, and performing feature extraction on the original sound signal to obtain corresponding text data, thereby screening out effective sound signals; then preprocessing the text data to obtain a first spectrum image corresponding to the text data, and realizing the conversion of the text data into a spectrum image, and then regularizing the first spectrum image to concentrate the informative part of the first spectrum image to obtain a second spectrum image, so that the features in the second spectrum image are centralized, and the weights learned by the Resnet deep learning model are as small as possible, and finally the second spectrum image is input into the Resnet deep learning model for feature extraction and prediction classification processing, and the second spectrum image is subjected to feature extraction and prediction classification by a machine learning algorithm to obtain a sound analysis result corresponding to the second spectrum image. The original sound signal collected by the flexible fiber optic coupler in this application has a fast response time and strong electromagnetic anti-interference ability compared to traditional hearing aids based on electronic sensors. Collecting the original sound signal through the flexible fiber optic coupler can improve the sensitivity of sound signal collection; using the Resnet deep learning model to process and classify the spectrum image corresponding to the original sound signal using a machine learning algorithm can improve the accuracy of the original sound signal analysis results, while improving the user experience.

[0011] According to some embodiments of the present application, preprocessing the text data to obtain a first spectrum image corresponding to the text data includes:

[0012] The text data is segmented, windowed and fast Fourier transformed by a short-time Fourier transform method to obtain a first spectrum image corresponding to the text data, wherein the first spectrum image includes frequency information corresponding to the text data in different time intervals.

[0013] According to some embodiments of the present application, performing regularization processing on the first spectrum image to obtain the second spectrum image includes:

[0014] The frequency information contained in the first spectrum image is transformed, scaled, rotated, and cropped, and the frequency information contained in the first spectrum image is normalized to obtain a second spectrum image.

[0015] According to some embodiments of the present application, the Resnet deep learning model includes a CNN network model, and the CNN network model includes a convolutional layer, a maximum pooling layer, a global pooling layer, and a fully connected layer.

[0016] According to some embodiments of the present application, the step of inputting the second spectrum image into a Resnet deep learning model for feature extraction and prediction classification processing to obtain a sound analysis result corresponding to the second spectrum image includes:

[0017] Using the second spectrum image as input of the Resnet deep learning model, and transmitting the second spectrum image to the CNN network model;

[0018] Extracting features from the second spectrum image through the plurality of convolutional layers to obtain a feature image;

[0019] The feature image is transmitted to the maximum pooling layer, and the feature image is downsampled by the maximum pooling layer to reduce the number of parameters and the amount of calculation of the feature image;

[0020] The downsampled feature image is transmitted to the global pooling layer, and the spatial size of the feature image is reduced by the global pooling layer, so that the feature image is converted into a vector of a preset size to obtain a feature image vector;

[0021] Inputting the feature image vector into the fully connected layer, classifying the feature image vector through the fully connected layer, and outputting probability distributions of multiple predicted categories corresponding to the feature image vector;

[0022] A sound analysis result corresponding to the second spectrum image is determined according to the probability distribution.

[0023] According to some embodiments of the present application, determining the sound analysis result corresponding to the second spectrum image according to the probability distribution includes:

[0024] A maximum probability distribution is determined among the probability distributions of the plurality of predicted categories, and the predicted category corresponding to the maximum probability distribution is used as a sound analysis result corresponding to the second spectrum image.

[0025] According to some embodiments of the present application, after acquiring the original sound signal, the method further includes:

[0026] Determining whether the original sound signal contains a sound waveform change;

[0027] If the original sound signal has no waveform change, discard the original sound signal;

[0028] If the original sound signal includes waveform changes, feature preprocessing is performed on the original sound signal, and two signals of the original sound signal are converted into a light splitting ratio.

[0029] In a second aspect, an embodiment of the present application provides a controller comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the method described in the technical solution of the first aspect when running the computer program.

[0030] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, characterized in that: computer-executable instructions are stored therein, and the computer-executable instructions are used to execute the method described in the technical solution of the first aspect above.

[0031] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 is a flow chart of a sound recognition method based on a flexible optical fiber coupler provided in an embodiment of the present application;

[0033] Figure 2 is a flow chart of a method for obtaining a sound analysis result corresponding to a second spectrum image provided by an embodiment of the present application;

[0034] Figure 3 is a flow chart of a method for obtaining probability distribution of multiple prediction categories provided in an embodiment of the present application;

[0035] Figure 4 is a flow chart of another sound recognition method based on a flexible optical fiber coupler provided in an embodiment of the present application;

[0036] Figure 5 It is a structural schematic diagram of a controller provided in an embodiment of the present application. DETAILED DESCRIPTION

[0037] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application is further described in detail below in conjunction with the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present application and are not intended to limit the present application. In addition, the characteristics, operations or features described in the specification can be combined in any appropriate manner to form various implementation methods. At the same time, the steps or actions in the method description can also be replaced or adjusted in order in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the accompanying drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a necessary sequence, unless otherwise specified that a certain sequence must be followed.

[0038] In the description of this application, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed", etc. are understood to exclude the number itself, and "above", "below", "within", etc. are understood to include the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.

[0039] The serial numbers of the components in this document, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning. The "connection" and "coupling" mentioned in this application, unless otherwise specified, include direct and indirect connections (couplings).

[0040] Sound is an important medium for information dissemination in daily life and a very important way of communication in daily life. As people age, their hearing ability will inevitably decline to a certain extent. Currently, hearing aids are mostly used to solve hearing problems of the elderly. However, the size and weight of hearing aids limit the user's usage time. The hearing loss of each user is different, and the damage to the left and right ears is different, so it takes a long time to debug.

[0041] In the related technology, most hearing aids are electronic sensors. In an electromagnetic field environment, traditional electronic sensors may fail due to electromagnetic interference, resulting in low sensitivity of the hearing aid in receiving sound signals and poor performance in processing sound signals. The accuracy of the sound analysis results output by the hearing aid is low, and users cannot accurately and clearly obtain sounds, resulting in a poor user experience for the wearer.

[0042] Based on this, the embodiments of the present application provide a sound recognition method, controller and storage medium based on a flexible optical fiber coupler, which is conducive to improving the sensitivity of acquiring sound signals, improving the accuracy of sound signal recognition results, and improving user experience.

[0043] The following is a further description of the sound recognition method based on the flexible optical fiber coupler, the controller and the storage medium provided in the embodiments of the present application in conjunction with the accompanying drawings.

[0044] Reference Figure 1 As shown, Figure 1 is a flow chart of a sound recognition method based on a flexible optical fiber coupler provided in an embodiment of the present application. The sound recognition method based on a flexible optical fiber coupler includes but is not limited to steps S100 to S400. Specifically,

[0045] Step S100: acquiring an original sound signal, and performing feature extraction on the original sound signal to obtain text data corresponding to the original sound signal;

[0046] Step S200: preprocessing the text data to obtain a first spectrum image corresponding to the text data;

[0047] Step S300: performing regularization processing on the first spectrum image to obtain a second spectrum image;

[0048] Step S400: Input the second spectrum image into the Resnet deep learning model for feature extraction and prediction classification processing to obtain a sound analysis result corresponding to the second spectrum image.

[0049] In some embodiments of the present application, a sound recognition method based on a flexible optical fiber coupler includes: first, obtaining an original sound signal collected by a flexible optical fiber coupler, and performing feature extraction on the original sound signal to obtain corresponding text data, thereby screening out effective sound signals; then, preprocessing the text data to obtain a first spectrum image corresponding to the text data, thereby converting the text data into a spectrum image, and then, regularizing the first spectrum image to concentrate the informative part of the first spectrum image to obtain a second spectrum image, so that the features in the second spectrum image are concentrated, causing the weights learned by the Resnet deep learning model to be as small as possible, and finally, inputting the second spectrum image into the Resnet deep learning model for feature extraction and prediction classification processing, performing feature extraction and prediction classification on the second spectrum image through a machine learning algorithm, and obtaining a sound analysis result corresponding to the second spectrum image. The original sound signal collected by the flexible fiber optic coupler in this application has a fast response time and strong electromagnetic anti-interference ability compared to traditional hearing aids based on electronic sensors. Collecting the original sound signal through the flexible fiber optic coupler can improve the sensitivity of sound signal collection; using the Resnet deep learning model to process and classify the spectrum image corresponding to the original sound signal using a machine learning algorithm can improve the accuracy of the original sound signal analysis results, while improving the user experience.

[0050] In some embodiments of the present application, the method of preprocessing text data to obtain a first spectrum image corresponding to the text data includes but is not limited to step S210. Specifically,

[0051] Step S210: Segmenting, windowing and fast Fourier transforming the text data are performed by a short-time Fourier transform method to obtain a first spectrum image corresponding to the text data, wherein the first spectrum image includes frequency information corresponding to the text data in different time intervals.

[0052] After obtaining the text data corresponding to the original sound signal, the text data is screened and preprocessed by the short-time Fourier transform method to extract the effective signal. Short-time Fourier transform is a method for analyzing the local time-frequency characteristics of a signal. It divides the signal into short-time frames and applies Fourier transform on each frame to observe the frequency components of the signal at different time points. The screening and preprocessing of text data by the short-time Fourier transform method includes: segmenting, windowing and fast Fourier transforming the text data; segmenting the text data to achieve segmentation of continuous text data into multiple short-time frames, and then multiplying each frame signal by a window function, such as a Hanning window, a Hamming window or a rectangular window, to achieve windowing. The windowing process can reduce the discontinuity effect at the frame boundary; then, fast Fourier transforming the windowed frame to obtain a frequency representation of the frame, generating a time-frequency result through fast Fourier transforming, and obtaining a first spectrum image based on the time-frequency result. It can be understood that the first spectrum image includes frequency information corresponding to text data in different time intervals, and can observe the frequency distribution of text data at each time point, providing a basis for further analysis and processing of text data corresponding to the speech signal.

[0053] In some embodiments of the present application, performing regularization processing on the first spectrum image to obtain the second spectrum image includes but is not limited to step S310. Specifically,

[0054] Step S310: transform, scale, rotate, and crop the frequency information contained in the first spectrum image, and normalize the frequency information contained in the first spectrum image to obtain a second spectrum image.

[0055] After the text data is processed by the short-time Fourier transform method to obtain the first spectrum image, the first spectrum image is regularized, including transforming, scaling, rotating, and cropping the frequency information contained in the first spectrum image to increase the diversity of the data. By increasing the diversity of the data, the generalization ability of the Resnet deep learning model can be improved to avoid overfitting. Then, the frequency information contained in the first spectrum image is normalized to obtain the second spectrum image. After the normalization process, the features of the informative part of the second spectrum image are centralized, so that the weights learned by the Resnet deep learning model are as small as possible, thereby improving the stability of the model.

[0056] In some embodiments of the present application, the Resnet deep learning model includes a CNN network model, and the CNN network model includes a convolutional layer, a maximum pooling layer, a global pooling layer, and a fully connected layer.

[0057] The Resnet deep learning model is a deep learning architecture that solves the degradation problem that occurs as the network depth increases by introducing residual learning. The core of ResNet is the residual module, which allows the gradient to flow directly to the shallower layer through jump connections, thereby alleviating the gradient vanishing problem and making it possible to train deeper networks. The Resnet deep learning model includes a CNN network model, which includes multiple convolutional layers. The convolutional layer is used to perform convolution operations to extract the features of the spectrum graph. The CNN network model also includes a maximum pooling layer, a global pooling layer, and a fully connected layer; the maximum pooling layer is used to retain the input features while reducing the number of parameters, and the global pooling layer is used to change the size of the features extracted by the convolution layer from (H, W) to (1, 1), so that the entire feature map of the second spectrum image is converted into a vector of fixed length, so as to facilitate subsequent connection to the fully connected layer, perform classification tasks through the fully connected layer, and output the sound analysis results corresponding to the second spectrum image.

[0058] Reference Figure 2 As shown, Figure 2 is a flowchart of a method for obtaining a sound analysis result corresponding to a second spectrum image provided by an embodiment of the present application. The method for obtaining a sound analysis result corresponding to the second spectrum image includes but is not limited to steps S410 to S460. Specifically,

[0059] Step S410: using the second spectrum image as an input of the Resnet deep learning model, and transmitting the second spectrum image to the CNN network model;

[0060] Step S420: extracting features from the second spectrum image through multiple convolutional layers to obtain a feature image;

[0061] Step S430: transmitting the feature image to the maximum pooling layer, and downsampling the feature image through the maximum pooling layer to reduce the number of parameters and the amount of calculation of the feature image;

[0062] Step S440: transmitting the downsampled feature image to the global pooling layer, and reducing the spatial size of the feature image through the global pooling layer, so that the feature image is converted into a vector of a preset size, thereby obtaining a feature image vector;

[0063] Step S450: inputting the feature image vector into the fully connected layer, classifying the feature image vector through the fully connected layer, and outputting the probability distribution of multiple prediction categories corresponding to the feature image vector;

[0064] Step S460: determining the sound analysis result corresponding to the second spectrum image according to the probability distribution.

[0065] In some embodiments of the present application, after performing a short-time Fourier transform on the text data corresponding to the original speech signal to obtain a first spectrum image, the first spectrum image is regularized to obtain a second spectrum image, and the second spectrum image is used as the input of the Resnet deep learning model, and the second spectrum image is transmitted to the CNN network model, and the CNN network model includes a convolution layer, a maximum pooling layer, a global pooling layer, and a fully connected layer. The second spectrum image is subjected to feature extraction through multiple convolution layers to screen out effective features to obtain a feature image, and then the feature image is transmitted to the maximum pooling layer. The purpose of the maximum pooling operation is to reduce the number of parameters while retaining the input features. The features after completing multiple convolution operations are sent to the global pooling layer, and the feature image is downsampled through the maximum pooling layer to reduce the number of parameters and the amount of calculation of the feature image. After that, the feature image vector is input to the fully connected layer, and the feature image vector is classified through the fully connected layer, and the probability distribution of multiple prediction categories corresponding to the feature image vector is output, and the sound analysis result corresponding to the second spectrum image is determined according to the probability distribution.

[0066] The second spectrum image is input into the Resnet deep learning model for feature extraction and prediction classification, and the second spectrum image is subjected to feature extraction and prediction classification by a machine learning algorithm to obtain a sound analysis result corresponding to the second spectrum image. The original sound signal collected by the flexible fiber optic coupler in this application has a fast response time and strong electromagnetic anti-interference ability compared to traditional hearing aids based on electronic sensors. Collecting the original sound signal by the flexible fiber optic coupler can improve the sensitivity of sound signal collection; processing and classifying the spectrum image corresponding to the original sound signal by the Resnet deep learning model using a machine learning algorithm can improve the accuracy of the original sound signal analysis results, while improving the user experience.

[0067] Reference Figure 3 As shown, Figure 3 is a flowchart of a method for obtaining a probability distribution of multiple prediction categories provided in an embodiment of the present application. The method for obtaining a probability distribution of multiple prediction categories includes but is not limited to steps S451 to S452. Specifically,

[0068] Step S451: determining the number of feature channels of the feature image vector through a fully connected layer, and transforming the number of feature channels into the predicted number of categories;

[0069] Step S452: Convert the predicted number of categories into probability distributions of multiple predicted categories corresponding to the feature image vector through the Softmax function.

[0070] When the CNN network model processes image classification tasks, the fully connected layer and the Softmax function are usually used in the final stage of the network to complete the transformation from feature extraction to final classification. The fully connected layer is a dense layer in CNN. In the fully connected layer, the input data is usually a one-dimensional feature vector, which is extracted from the previous convolutional layer and pooling layer. The purpose of the fully connected layer is to transform the number of feature channels of the feature image vector into the predicted number of categories. The Softmax function is usually used in multi-classification problems to convert the predicted number of categories output by the fully connected layer into a probability distribution, where each category has a probability value between 0 and 1, and the sum of the probabilities of all categories is 1. In this way, each feature image vector corresponds to a predicted category probability distribution, which indicates the confidence of the Resnet deep learning model that the spectrum image belongs to each category.

[0071] By inputting the feature image vector into the fully connected layer, the fully connected layer and the Softmax function cooperate with each other to complete the conversion from the number of feature channels of the feature image vector to the final classification corresponding to the feature image vector, thereby obtaining the probability distribution of multiple prediction categories.

[0072] In some embodiments of the present application, the method for determining the sound analysis result corresponding to the second spectrum image according to the probability distribution includes but is not limited to step S461. Specifically,

[0073] Step S461: determining a maximum probability distribution among probability distributions of a plurality of prediction categories, and taking the prediction category corresponding to the maximum probability distribution as the sound analysis result corresponding to the second spectrum image.

[0074] The feature image vector is classified through a fully connected layer, and probability scores of multiple prediction categories corresponding to the feature image vector are output. Then, the maximum probability distribution is determined among the probability distributions of the multiple prediction categories, and the prediction category corresponding to the maximum probability distribution is used as the sound analysis result corresponding to the second spectrum image.

[0075] In some embodiments of the present application, after the sound analysis result corresponding to the second spectrum image is output, the sound analysis result is tested for accuracy, and when the accuracy test of the sound analysis result is greater than 97%, the sound analysis result is output. By performing an accuracy test on the sound analysis result, the accuracy of the original sound signal analysis result can be improved, and the user experience can be improved.

[0076] Reference Figure 4 As shown, Figure 4 is a flow chart of another sound recognition method based on a flexible optical fiber coupler provided in an embodiment of the present application. The sound recognition method based on a flexible optical fiber coupler includes but is not limited to steps S110 to S130. Specifically,

[0077] Step S110: determining whether the original sound signal contains sound waveform changes;

[0078] Step S120: If the original sound signal has no waveform change, discard the original sound signal;

[0079] Step S130: If the original sound signal includes waveform changes, feature preprocessing is performed on the original sound signal, and two signals of the original sound signal are converted into a light splitting ratio.

[0080] In some embodiments of the present application, after acquiring the original sound signal, it is determined whether the original sound signal contains sound waveform changes. If the original sound signal has no waveform changes, it indicates that the original sound signal is unqualified and the original sound signal is discarded. When the original sound signal includes waveform changes, it indicates that the original sound signal is qualified, and feature preprocessing is performed on the original sound signal, and the two signals of the original sound signal are converted into optical power ratios, in which the real-time optical power ratio of one signal to the sum of the two signals at this time is P1 / (P1+P2).

[0081] By determining whether the original sound signal contains sound waveform changes after acquiring the original sound signal and screening out qualified original sound signals, the quality of the original sound signal can be improved, thereby improving the accuracy of the sound signal recognition results and improving the user experience.

[0082] In some embodiments of the present application, after the original sound signal is feature preprocessed and the two signals of the original sound signal are converted into a light-to-noise ratio, the processed original sound signal is feature extracted, and the original sound signal is converted into a spectrum using the Fourier transform method for feature extraction. The color depth on the spectrum image is the logarithm of the power density, and the colors on the spectrum image represent the different sizes of the frequencies here. The brighter the color, the greater the power of the frequency component; the darker the color, the smaller the power.

[0083] Afterwards, the spectrum image is put into the data set, and is put into the training set and the test set respectively in a ratio of 8:2, and the data set is trained and learned using the Resnet deep learning model. The above spectrum image is subjected to feature extraction through convolution operation. This method effectively reduces the number of model parameters and computer memory without affecting training efficiency and accuracy. The sound analysis results corresponding to the spectrum image are classified and identified, and the accuracy is judged. When the accuracy test of the sound analysis result is greater than 97%, the sound analysis result is output. By performing an accuracy test on the sound analysis result, the accuracy of the original sound signal analysis result can be improved, and the user experience can be improved.

[0084] The sound recognition method based on flexible optical fiber coupler is applied to the daily communication of the hearing-impaired people. When different sound signals are received based on the flexible optical fiber coupler, different waveform signals will be generated, and the waveform signals will be converted into text data. The text data will be subjected to Fourier transform and regularization processing to obtain the spectrum image corresponding to the sound signal. The spectrum image will be input into the Resnet deep learning model for feature extraction and prediction classification processing to obtain the sound analysis result corresponding to the second spectrum image. Users can obtain the content corresponding to the sound signal based on the sound analysis results, which is convenient for users' daily communication.

[0085] Reference Figure 5 As shown, Figure 5 It is a structural diagram of a controller 1000 provided in an embodiment of the present application, including a processor 1001, which can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the method provided in the embodiment of the present application; a memory 1002, which can be implemented in the form of a read-only memory 1002 (Read Only Memory, ROM), a static storage device, a dynamic storage device or a random access memory 1002 (Random Access Memory, RAM), etc. The memory 1002 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 1002 and are called by the processor 1001 to execute the embodiments of this application; the input / output interface 1003 is used to implement information input and output; the communication interface 1004 is used to implement communication interaction between this device and other devices, and communication can be achieved through wired methods (such as USB, network cables, etc.) or wireless methods (such as mobile networks, WIFI, Bluetooth, etc.); a bus transmits information between various components of the device (such as the processor 1001, the memory 1002, the input / output interface 1003 and the communication interface 1004); wherein the processor 1001, the memory 1002, the input / output interface 1003 and the communication interface 1004 are connected to each other within the device through the bus.

[0086] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer-readable storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer-readable storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer-readable storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically embodies computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0087] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the above-mentioned implementation mode. Technical personnel familiar with the field can also make various equivalent deformations or substitutions without violating the spirit of the present application. These equivalent deformations or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A sound recognition method based on a flexible optical fiber coupler, characterized in that: include: Acquire an original sound signal, and perform feature extraction on the original sound signal to obtain text data corresponding to the original sound signal; Preprocessing the text data to obtain a first spectrum image corresponding to the text data; Performing regularization processing on the first spectrum image to obtain a second spectrum image; The second spectrum image is input into the Resnet deep learning model for feature extraction and prediction classification processing to obtain a sound analysis result corresponding to the second spectrum image.

2. The sound recognition method based on the flexible optical fiber coupler according to claim 1, characterized in that: The preprocessing of the text data to obtain a first spectrum image corresponding to the text data comprises: The text data is segmented, windowed and fast Fourier transformed by a short-time Fourier transform method to obtain a first spectrum image corresponding to the text data, wherein the first spectrum image includes frequency information corresponding to the text data in different time intervals.

3. The sound recognition method based on flexible optical fiber coupler according to claim 2, characterized in that: The performing regularization processing on the first spectrum image to obtain a second spectrum image comprises: The frequency information contained in the first spectrum image is transformed, scaled, rotated, and cropped, and the frequency information contained in the first spectrum image is normalized to obtain a second spectrum image.

4. The sound recognition method based on flexible optical fiber coupler according to claim 1, characterized in that: The Resnet deep learning model includes a CNN network model, and the CNN network model includes a convolutional layer, a maximum pooling layer, a global pooling layer, and a fully connected layer.

5. The sound recognition method based on flexible optical fiber coupler according to claim 4, characterized in that: The step of inputting the second spectrum image into a Resnet deep learning model for feature extraction and prediction classification processing to obtain a sound analysis result corresponding to the second spectrum image includes: Using the second spectrum image as input of the Resnet deep learning model, and transmitting the second spectrum image to the CNN network model; Extracting features from the second spectrum image through the plurality of convolutional layers to obtain a feature image; The feature image is transmitted to the maximum pooling layer, and the feature image is downsampled by the maximum pooling layer to reduce the number of parameters and the amount of calculation of the feature image; The downsampled feature image is transmitted to the global pooling layer, and the spatial size of the feature image is reduced by the global pooling layer, so that the feature image is converted into a vector of a preset size to obtain a feature image vector; Inputting the feature image vector into the fully connected layer, classifying the feature image vector through the fully connected layer, and outputting probability distributions of multiple predicted categories corresponding to the feature image vector; A sound analysis result corresponding to the second spectrum image is determined according to the probability distribution.

6. The sound recognition method based on flexible optical fiber coupler according to claim 5, characterized in that: After classifying the feature image vector through the fully connected layer, the method further includes: Determining the number of feature channels of the feature image vector through the fully connected layer, and transforming the number of feature channels into a predicted number of classifications; The predicted classification quantity is converted into the probability distribution of the multiple predicted categories corresponding to the feature image vector through the Softmax function.

7. The sound recognition method based on flexible optical fiber coupler according to claim 5, characterized in that: The determining, according to the probability distribution, a sound analysis result corresponding to the second spectrum image comprises: A maximum probability distribution is determined among the probability distributions of the plurality of predicted categories, and the predicted category corresponding to the maximum probability distribution is used as a sound analysis result corresponding to the second spectrum image.

8. The sound recognition method based on flexible optical fiber coupler according to claim 1, characterized in that: After obtaining the original sound signal, the method further includes: Determining whether the original sound signal contains a sound waveform change; If the original sound signal has no waveform change, discard the original sound signal; If the original sound signal includes waveform changes, feature preprocessing is performed on the original sound signal, and two signals of the original sound signal are converted into a light splitting ratio.

9. A controller, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the method according to any one of claims 1 to 8 when executing the computer program.

10. A computer-readable storage medium, characterized in that: Computer executable instructions are stored, and the computer executable instructions are used to execute the method according to any one of claims 1 to 8.

Images