Baby crying recognition method and terminal device

Through fixed-point operation and filtering processing, the baby cry recognition method improves the accuracy of baby cry recognition, and solves the problems of low accuracy and high false alarm rate in the prior art.

CN112382302BActive Publication Date: 2024-11-19ジャン州立達信光電子科技有限公司
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202011391118.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-12-02
Publication Date
2024-11-19
Estimated Expiration
2040-12-02

AI Technical Summary

Technical Problem

The existing infant cry recognition methods have low accuracy and are prone to falsely relay other sounds with larger decibel values.

Method used

Fixed-point operation is used to extract the digital sound signal feature, obtain the fixed-point number feature vector, and then filter the floating-point number feature vector, and input the pre-trained sound recognition neural network model for identification.

Benefits of technology

It improves the accuracy of baby crying recognition, reduces the amount of calculation, and reduces the false alarm rate.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN112382302B_ABST
    Figure CN112382302B_ABST
Patent Text Reader

Abstract

The present invention is applicable to the technical field of sound recognition, and discloses a method for recognizing a baby's cry and a terminal device, the method comprising: obtaining a digitized sound signal; using fixed-point operations to extract features of the digitized sound signal to obtain a fixed-point feature vector; performing filtering operations on the fixed-point feature vector to obtain a floating-point feature vector; inputting the floating-point feature vector into a pre-trained sound recognition neural network model to identify whether the digitized sound signal contains a baby's cry. The present invention can improve the accuracy of baby cry recognition while ensuring a low amount of operations.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of sound recognition, and in particular relates to a method and terminal equipment for recognizing baby crying sounds. Background Art

[0002] At present, most baby cry recognition methods are based on the fact that baby cry has a high decibel value, and are identified based on the decibel value. Although this recognition method has a very low computational complexity, if there are other sounds with a high decibel value, such as impact sounds, calls, etc., they will also be identified as baby cry, resulting in false alarms. Therefore, the accuracy of existing baby cry recognition methods is low. Summary of the invention

[0003] In view of this, an embodiment of the present invention provides a method and a terminal device for recognizing a baby's cry, so as to solve the problem of low accuracy of the prior art method for recognizing a baby's cry.

[0004] A first aspect of an embodiment of the present invention provides a method for recognizing a baby's cry, comprising:

[0005] Acquiring digitized sound signals;

[0006] Using fixed-point operations, the digital sound signal is subjected to feature extraction to obtain a fixed-point feature vector;

[0007] Perform filtering operation on the fixed-point feature vector to obtain a floating-point feature vector;

[0008] The floating-point feature vector is input into a pre-trained sound recognition neural network model to identify whether the digitized sound signal contains a baby crying.

[0009] A second aspect of an embodiment of the present invention provides a baby crying recognition device, comprising:

[0010] An acquisition module, used for acquiring a digitized sound signal;

[0011] The feature extraction module is used to extract features of the digitized sound signal using fixed-point operations to obtain a fixed-point feature vector;

[0012] A filtering module is used to perform filtering operations on fixed-point feature vectors to obtain floating-point feature vectors;

[0013] The recognition module is used to input the floating point feature vector into the pre-trained sound recognition neural network model to identify whether the digitized sound signal contains the baby crying sound.

[0014] A third aspect of an embodiment of the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the baby crying recognition method described in the first aspect are implemented.

[0015] A fourth aspect of an embodiment of the present invention provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by one or more processors, the steps of the baby crying recognition method described in the first aspect are implemented.

[0016] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: the embodiments of the present invention use fixed-point operations to extract features from digitized sound signals to obtain fixed-point feature vectors, and then perform filtering operations on the fixed-point feature vectors to obtain floating-point feature vectors. The use of fixed-point operations can reduce the amount of calculations; by inputting the floating-point feature vectors into a pre-trained sound recognition neural network model to identify whether the digitized sound signal contains a baby's cry, the accuracy of baby's cry recognition can be improved; the embodiments of the present invention can improve the accuracy of baby's cry recognition while ensuring a low amount of calculations. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0018] Figure 1 It is a schematic diagram of the implementation flow of a method for recognizing a baby's cry provided by an embodiment of the present invention;

[0019] Figure 2 is a structural schematic diagram of a sound recognition neural network model provided by an embodiment of the present invention;

[0020] Figure 3 is a schematic block diagram of a baby crying recognition device provided by an embodiment of the present invention;

[0021] Figure 4 It is a schematic block diagram of a terminal device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0023] In order to illustrate the technical solution of the present invention, a specific embodiment is provided below for illustration.

[0024] Figure 1 FIG. 1 is a schematic diagram of the implementation flow of a method for recognizing a baby cry provided by an embodiment of the present invention. For the sake of convenience, only the part related to the embodiment of the present invention is shown. The execution subject of the embodiment of the present invention may be a processor included in a terminal device. Figure 1 As shown, the method may include the following steps:

[0025] S101: Acquire a digitized sound signal.

[0026] In the embodiment of the present invention, the sound collection module can collect external sound signals, and finally convert the external sound signals into digital sound signals, and send them to the processor. The digital sound signals can refer to sound signals in binary form.

[0027] S102: Using fixed-point operations to extract features from the digitized sound signal to obtain a fixed-point feature vector.

[0028] In the embodiment of the present invention, in order to speed up the instruction cycle and reduce the burden of computing resources, fixed-point computing is used in the process of feature extraction.

[0029] In one embodiment of the present invention, the above S102 may include the following steps:

[0030] Using fixed-point calculation, pre-emphasize the digitized sound signal to obtain a pre-emphasized sound signal;

[0031] Divide the pre-emphasized sound signal into multiple frames of sound frame signals; wherein each frame of the sound frame signal overlaps with the previous frame of the sound frame signal for a preset time;

[0032] Using a Hamming window, each frame of the audio frame signal is windowed to obtain corresponding multi-frame windowed audio frame signals;

[0033] Perform Fourier transform on each frame of the windowed audio frame signal to obtain multiple corresponding Fourier spectra;

[0034] Performing Mel triangle filtering and logarithmic operation on each Fourier spectrum to obtain multiple corresponding Mel cepstrum;

[0035] Each Mel-frequency cepstral spectrum is subjected to discrete cosine transformation to obtain a corresponding plurality of fixed-point MFCC (Mel-frequency Cepstral Coefficient) features, and the plurality of fixed-point MFCC features constitute a fixed-point feature vector.

[0036] Specifically, the digitized sound signal can be processed by pre-emphasis to compensate for the loss of high frequency sound. The digitized sound signal after pre-emphasis is called a pre-emphasized sound signal.

[0037] The pre-emphasized sound signal is divided into a sound frame every 32ms and overlapped for 16ms to obtain a multi-frame sound frame signal. After each frame of the sound frame signal passes through the Hamming window, the distortion of the spectrum and the discontinuity of the sound can be reduced, making the sound more concentrated. The sound frame signal after the Hamming window is called the windowed sound frame signal.

[0038] Perform Fourier transform on the audio frame signal after windowing each frame to obtain multiple Fourier spectra. The energy of each Fourier spectrum is passed through 40 sets of Mel triangle filters and logarithmically calculated to obtain multiple Mel cepstrum, which is the intensity of the auditory frequency response of the human ear. Each Mel cepstrum is reduced in dimension by discrete cosine transform to obtain multiple fixed-point MFCC features, which form a fixed-point feature vector. Among them, the fixed-point MFCC feature refers to the MFCC represented by a fixed-point number.

[0039] S103: Perform filtering operation on the fixed-point feature vector to obtain a floating-point feature vector.

[0040] Each fixed-point MFCC feature in the fixed-point feature vector is filtered to obtain multiple filtered floating-point MFCC features to form a floating-point feature vector. The floating-point MFCC feature refers to the MFCC represented by a floating-point number.

[0041] In one embodiment of the present invention, the above S103 may include:

[0042] S104: Inputting the floating-point feature vector into a pre-trained sound recognition neural network model to identify whether the digitized sound signal contains a baby crying sound.

[0043] In the embodiment of the present invention, the crying sound of a baby is recognized by a sound recognition neural network model. The sound recognition neural network model can recognize whether the sound signal contains the crying sound of a baby.

[0044] From the above description, it can be seen that the embodiment of the present invention uses fixed-point operations to extract features from digitized sound signals to obtain fixed-point feature vectors, and then performs filtering operations on the fixed-point feature vectors to obtain floating-point feature vectors. The use of fixed-point operations can reduce the amount of calculations; by inputting the floating-point feature vectors into a pre-trained sound recognition neural network model to identify whether the digitized sound signal contains a baby's cry, the accuracy of baby's cry recognition can be improved; the embodiment of the present invention can improve the accuracy of baby's cry recognition while ensuring a low amount of calculations.

[0045] In one embodiment of the present invention, the sound recognition neural network model includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer connected in sequence;

[0046] The dimension of the input layer is the same as the dimension of the floating-point feature vector;

[0047] The first hidden layer includes 64 GRU (Gated Recurrent Unit) neurons, the second hidden layer includes 32 DNN (Deep Neural Network) neurons, and the third hidden layer includes 16 DNN neurons;

[0048] The output layer contains 18 neurons.

[0049] See also Figure 2 , the sound recognition neural network model may include an input layer, a hidden layer and an output layer, and the hidden layer may include a first hidden layer, a second hidden layer and a third hidden layer. Among them, the dimension of the input layer is the same as the dimension of the floating-point feature vector, both of which are 12. The hidden layer uses a smaller and shallower three-layer architecture, and the number of parameters can be greatly reduced. The 18 neurons in the output layer can represent 18 common sound event categories, including the category of baby crying.

[0050] The sound recognition neural network model used in the embodiment of the present invention has a high degree of abstractness for data and can have a good effect on nonlinear data operations. In addition, the model is robust and has strong noise resistance.

[0051] In one embodiment of the present invention, before the above S104, the above method for recognizing a baby's cry may further include:

[0052] Obtain an initial training set, and perform data augmentation on the initial training set to obtain a target training set;

[0053] Perform feature extraction on each training sample in the target training set to obtain a floating-point training feature vector corresponding to each training sample;

[0054] Perform Z-score standardization on each floating-point training feature vector to obtain corresponding multiple standard training feature vectors;

[0055] Based on multiple standard training feature vectors, a preset sound recognition neural network model is trained to obtain a pre-trained sound recognition neural network model.

[0056] The initial training set may include sound data of different categories, that is, training samples of different categories, and each sound data has been labeled with a category. The sound data of different categories include data of the category of baby crying.

[0057] In order to make up for the lack of real data and enhance the model's anti-noise ability, data enhancement methods are used to enrich the sound data in the initial training set to obtain the target training set. Data enhancement methods may include EMDA (Equalized Mixture Data Augmentation) and VTLP (Vocal Tract Length Perturbation). Among them, EMDA generates new sound data by adding white noise or other noise to clean sound data, such as baby crying data; VTLP generates new sound data by slightly perturbing the spectrum. The generated new sound data and the original sound data constitute the target training set.

[0058] The above-mentioned feature extraction method is used for each training sample in the target training set to extract the corresponding floating-point training feature vector. The floating-point training feature vector is the floating-point feature vector corresponding to the training sample.

[0059] Z-score Standardization is a standardization technique. After Z-score standardization, the data can conform to the normal distribution (mean value is 0, standard deviation is 1). Its advantage is that it can reduce the impact of outliers on the overall data and also reduce the impact on the model.

[0060] In one embodiment of the present invention, after obtaining the pre-trained sound recognition neural network model, the method for recognizing a baby's cry further includes:

[0061] Obtain a test sample, perform feature extraction on the test sample, and obtain a floating-point test feature vector corresponding to the test sample;

[0062] Perform Z-score normalization on the floating-point test feature vector to obtain a standard test feature vector;

[0063] Input the standard test feature vector into the pre-trained sound recognition neural network model to obtain the recognition result of the test sample;

[0064] If the recognition result of the test sample is different from the pre-labeled result of the test sample, a new initial training set is obtained, and the sound recognition neural network model is continued to be trained based on the new initial training set until the recognition result of the test sample is the same as the pre-labeled result of the test sample.

[0065] Among them, the floating-point test feature vector is the floating-point feature vector corresponding to the test sample. Using the same feature extraction method, the floating-point test feature vector corresponding to the test sample can be obtained. The pre-labeled result is the category to which the test sample has been labeled.

[0066] The pre-trained sound recognition neural network model is tested with the test sample. If the recognition result of the test sample is the same as the pre-labeled result of the test sample, the training is completed. If the recognition result of the test sample is different from the pre-labeled result of the test sample, a new initial training set is obtained, the above training process is re-executed, and the sound recognition neural network model is continuously trained until the recognition result of the test sample is the same as the pre-labeled result of the test sample.

[0067] In one embodiment of the present invention, after the above S104, the above method for recognizing a baby's cry may further include:

[0068] If the digitized sound signal contains the crying sound of a baby, an alarm signal is sent to the alarm module, and the alarm signal is used to instruct the alarm module to sound an alarm.

[0069] If the baby's crying sound is recognized, an alarm will be sounded to alert the caregiver.

[0070] From the above description, it can be seen that the training process optimizes the MFCC sound features with Z-score standardization, so that the sound recognition neural network model converges better. Fixed-point operations and small neural networks greatly reduce the amount of calculation, increase the possibility of transplanting low-level platforms, and can effectively detect baby crying and reduce the false alarm rate. It can solve the common problems of baby crying detection devices on the market.

[0071] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0072] Corresponding to the above-mentioned baby crying recognition method, an embodiment of the present invention further provides a baby crying recognition device, which has the same beneficial effects as the above-mentioned baby crying recognition method. Figure 3Detailed description is a schematic block diagram of a baby crying recognition device provided by an embodiment of the present invention. For ease of description, only the parts related to the embodiment of the present invention are shown.

[0073] In the embodiment of the present invention, the baby crying recognition device 30 may include an acquisition module 301 , a feature extraction module 302 , a filtering module 303 and a recognition module 304 .

[0074] Wherein, the acquisition module 301 is used to acquire a digitized sound signal;

[0075] The feature extraction module 302 is used to extract features from the digitized sound signal using fixed-point operations to obtain a fixed-point feature vector;

[0076] A filtering module 303 is used to perform filtering operation on the fixed-point feature vector to obtain a floating-point feature vector;

[0077] The recognition module 304 is used to input the floating point feature vector into the pre-trained sound recognition neural network model to identify whether the digitized sound signal contains the crying sound of a baby.

[0078] In one embodiment of the present invention, the sound recognition neural network model in the recognition module 304 includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer connected in sequence;

[0079] The dimension of the input layer is the same as the dimension of the floating-point feature vector;

[0080] The first hidden layer includes 64 GRU neurons, the second hidden layer includes 32 DNN neurons, and the third hidden layer includes 16 DNN neurons;

[0081] The output layer contains 18 neurons.

[0082] In one embodiment of the present invention, the feature extraction module 302 is specifically used to:

[0083] Using fixed-point calculation, pre-emphasize the digitized sound signal to obtain a pre-emphasized sound signal;

[0084] Divide the pre-emphasized sound signal into multiple frames of sound frame signals; wherein each frame of the sound frame signal overlaps with the previous frame of the sound frame signal for a preset time;

[0085] Using a Hamming window, each frame of the audio frame signal is windowed to obtain corresponding multi-frame windowed audio frame signals;

[0086] Perform Fourier transform on each frame of the windowed audio frame signal to obtain multiple corresponding Fourier spectra;

[0087] Performing Mel triangle filtering and logarithmic operation on each Fourier spectrum to obtain multiple corresponding Mel cepstrum;

[0088] Each Mel-frequency cepstrum is subjected to discrete cosine transformation to obtain corresponding multiple fixed-point MFCC features, and multiple fixed-point MFCC features constitute a fixed-point feature vector.

[0089] In one embodiment of the present invention, the baby cry recognition device 30 may further include a training module.

[0090] Training modules can be used to:

[0091] Obtain an initial training set, and perform data augmentation on the initial training set to obtain a target training set;

[0092] Perform feature extraction on each training sample in the target training set to obtain a floating-point training feature vector corresponding to each training sample;

[0093] Perform Z-score standardization on each floating-point training feature vector to obtain corresponding multiple standard training feature vectors;

[0094] Based on multiple standard training feature vectors, a preset sound recognition neural network model is trained to obtain a pre-trained sound recognition neural network model.

[0095] In one embodiment of the present invention, the baby crying recognition device 30 may further include a testing module.

[0096] Test modules can be used to:

[0097] Obtain a test sample, perform feature extraction on the test sample, and obtain a floating-point test feature vector corresponding to the test sample;

[0098] Perform Z-score normalization on the floating-point test feature vector to obtain a standard test feature vector;

[0099] Input the standard test feature vector into the pre-trained sound recognition neural network model to obtain the recognition result of the test sample;

[0100] If the recognition result of the test sample is different from the pre-labeled result of the test sample, a new initial training set is obtained, and the sound recognition neural network model is continued to be trained based on the new initial training set until the recognition result of the test sample is the same as the pre-labeled result of the test sample.

[0101] In one embodiment of the present invention, the baby crying recognition device 30 may further include an alarm signal sending module.

[0102] The alarm signal sending module can be used for:

[0103] If the digitized sound signal contains the crying sound of a baby, an alarm signal is sent to the alarm module, and the alarm signal is used to instruct the alarm module to sound an alarm.

[0104] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the baby cry recognition device is divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into a processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned device can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0105] Figure 4 is a schematic block diagram of a terminal device provided by an embodiment of the present invention. Figure 4 As shown, the terminal device 40 of this embodiment includes: one or more processors 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, the steps in the above-mentioned various embodiments of the method for recognizing the crying sound of a baby are implemented, for example Figure 1 Alternatively, when the processor 401 executes the computer program 403, the functions of each module / unit in the above-mentioned baby crying recognition device embodiment are realized, for example Figure 3 Functions of modules 301 to 304 are shown.

[0106] Exemplarily, the computer program 403 may be divided into one or more modules / units, which are stored in the memory 402 and executed by the processor 401 to complete the present application. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 403 in the terminal device 40. For example, the computer program 403 may be divided into an acquisition module, a feature extraction module, a filtering module, and a recognition module, and the specific functions of each module are as follows:

[0107] An acquisition module, used for acquiring a digitized sound signal;

[0108] The feature extraction module is used to extract features of the digitized sound signal using fixed-point operations to obtain a fixed-point feature vector;

[0109] A filtering module is used to perform filtering operations on fixed-point feature vectors to obtain floating-point feature vectors;

[0110] The recognition module is used to input the floating point feature vector into the pre-trained sound recognition neural network model to identify whether the digitized sound signal contains the baby crying sound.

[0111] Other modules or units can refer to Figure 3 The description in the illustrated embodiment will not be repeated here.

[0112] The terminal device 40 may be a computing device such as a desktop computer, a notebook, a PDA, a cloud server, or a speech recognition device. The terminal device 40 includes but is not limited to a processor 401 and a memory 402. Those skilled in the art will appreciate that Figure 4 It is only an example of the terminal device 40 and does not constitute a limitation of the terminal device 40. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the terminal device 40 may also include an input device, an output device, a network access device, a bus, etc.

[0113] The processor 401 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0114] The memory 402 may be an internal storage unit of the terminal device 40, such as a hard disk or memory of the terminal device 40. The memory 402 may also be an external storage device of the terminal device 40, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash memory card (FlasT Card), etc. equipped on the terminal device 40. Further, the memory 402 may also include both an internal storage unit of the terminal device 40 and an external storage device. The memory 402 is used to store the computer program 403 and other programs and data required by the terminal device 40. The memory 402 may also be used to temporarily store data that has been output or is to be output.

[0115] In one embodiment of the present invention, the terminal device 40 further includes a sound collection module and an alarm module;

[0116] The sound collection module and the alarm module are both connected to the processor;

[0117] The sound acquisition module is used to collect external sound signals, convert the external sound signals into analog current signals, and respectively perform amplification processing, filtering processing and analog-to-digital conversion processing on the analog current signals to obtain digital sound signals, and send the digital sound signals to the processor;

[0118] The alarm module is used to sound an alarm when receiving an alarm signal sent by the processor.

[0119] Optionally, the sound collection module may include a microphone sound receiving unit, an amplifier, a filter and an analog-to-digital converter connected in sequence.

[0120] The microphone receiving unit is used to collect external sound signals and convert the external sound signals into analog current signals.

[0121] The amplifier is used to amplify the analog current signal. The amplifier can pre-set multiple sensitivities according to user needs, and select the appropriate sensitivity to amplify the analog current signal according to actual needs.

[0122] The filter is used to filter the amplified analog current signal. In the filtering process, the spectrum response of the signal can be adjusted to perform sound enhancement, equalization, noise filtering, etc.

[0123] The analog-to-digital converter is used to convert the filtered current signal into a digital sound signal and send it to the processor. The analog-to-digital converter can set the appropriate sampling frequency, bit number, etc. to convert the analog signal into a digital signal.

[0124] After receiving the alarm signal, the alarm module can sound an alarm or send relevant information to the mobile terminal of the relevant personnel.

[0125] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0126] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0127] In the embodiments provided in the present application, it should be understood that the disclosed baby cry recognition device and method can be implemented in other ways. For example, the baby cry recognition device embodiments described above are merely schematic. For example, the division of the modules or units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0128] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0129] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0130] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0131] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for identifying a baby's cry, characterized in that: include: Acquiring digitized sound signals; Using fixed-point calculation to extract features from the digitized sound signal to obtain a fixed-point feature vector; Performing filtering operation on the fixed-point feature vector to obtain a floating-point feature vector; Inputting the floating-point feature vector into a pre-trained sound recognition neural network model to identify whether the digitized sound signal contains a baby crying sound; The sound recognition neural network model comprises an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer connected in sequence; The dimension of the input layer is the same as the dimension of the floating point feature vector; The first hidden layer includes 64 GRU neurons, the second hidden layer includes 32 DNN neurons, and the third hidden layer includes 16 DNN neurons; The output layer comprises 18 neurons; the 18 neurons of the output layer represent 18 sound event categories, and the 18 sound event categories include a baby crying category; Before inputting the floating-point feature vector into a pre-trained sound recognition neural network model to identify whether the digitized sound signal contains a baby crying sound, the baby crying sound recognition method further includes: Acquire an initial training set, and perform data enhancement processing on the initial training set to obtain a target training set; the data enhancement method includes EMDA and VTLP; Perform feature extraction on each training sample in the target training set to obtain a floating-point training feature vector corresponding to each training sample; Perform Z-score standardization on each floating-point training feature vector to obtain corresponding multiple standard training feature vectors; Based on the training feature vectors of the multiple standards, the preset sound recognition neural network model is trained to obtain the pre-trained sound recognition neural network model.

2. The method for identifying a baby's cry according to claim 1, characterized in that: The method of using fixed-point operation to extract features from the digitized sound signal to obtain a fixed-point feature vector includes: Using fixed-point calculation, pre-emphasize the digitized sound signal to obtain a pre-emphasized sound signal; Dividing the pre-emphasized sound signal into multiple frames of sound frame signals; wherein each frame of the sound frame signal overlaps with the previous frame of the sound frame signal for a preset time; Using a Hamming window, each frame of the audio frame signal is windowed to obtain corresponding multi-frame windowed audio frame signals; Perform Fourier transform on each frame of the windowed audio frame signal to obtain multiple corresponding Fourier spectra; Performing Mel triangle filtering and logarithmic operation on each Fourier spectrum to obtain multiple corresponding Mel cepstrum; Discrete cosine transformation is performed on each Mel-frequency cepstrum to obtain a corresponding plurality of fixed-point MFCC features, and the plurality of fixed-point MFCC features constitute the fixed-point feature vector.

3. The method for recognizing a baby's cry according to claim 1, characterized in that: After obtaining the pre-trained sound recognition neural network model, the method for recognizing a baby's cry further includes: Acquire a test sample, and perform feature extraction on the test sample to obtain a floating-point test feature vector corresponding to the test sample; Performing Z-score normalization processing on the floating point test feature vector to obtain a standard test feature vector; Inputting the standard test feature vector into the pre-trained sound recognition neural network model to obtain the recognition result of the test sample; If the recognition result of the test sample is different from the pre-labeled result of the test sample, a new initial training set is obtained, and the sound recognition neural network model is continued to be trained based on the new initial training set until the recognition result of the test sample is the same as the pre-labeled result of the test sample.

4. The method for recognizing a baby's cry according to any one of claims 1 to 3, characterized in that: After identifying whether the digitized sound signal contains a baby crying sound, the method further includes: If the digitized sound signal contains the crying sound of a baby, an alarm signal is sent to the alarm module, and the alarm signal is used to instruct the alarm module to sound an alarm.

5. A baby crying recognition device, characterized in that: include: An acquisition module, used for acquiring a digitized sound signal; A feature extraction module, used to extract features from the digitized sound signal using fixed-point operations to obtain a fixed-point feature vector; A filtering module, used for performing a filtering operation on the fixed-point feature vector to obtain a floating-point feature vector; A recognition module, used for inputting the floating-point feature vector into a pre-trained sound recognition neural network model to identify whether the digitized sound signal contains a baby crying sound; The sound recognition neural network model includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer and an output layer connected in sequence; The dimension of the input layer is the same as the dimension of the floating point feature vector; The first hidden layer includes 64 GRU neurons, the second hidden layer includes 32 DNN neurons, and the third hidden layer includes 16 DNN neurons; The output layer comprises 18 neurons; the 18 neurons of the output layer represent 18 sound event categories, and the 18 sound event categories include a baby crying category; The baby crying recognition device also includes a training module; The training module is used to: obtain an initial training set, and perform data enhancement processing on the initial training set to obtain a target training set; the data enhancement method includes EMDA and VTLP; perform feature extraction on each training sample in the target training set to obtain a floating-point training feature vector corresponding to each training sample; perform Z-score standardization processing on each floating-point training feature vector to obtain corresponding multiple standard training feature vectors; based on the multiple standard training feature vectors, train a preset sound recognition neural network model to obtain the pre-trained sound recognition neural network model.

6. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method for recognizing the crying sound of a baby according to any one of claims 1 to 4 are implemented.

7. The terminal device according to claim 6, characterized in that: The terminal device also includes a sound collection module and an alarm module; The sound collection module and the alarm module are both connected to the processor; The sound collection module is used to collect external sound signals, convert the external sound signals into analog current signals, and respectively perform amplification processing, filtering processing and analog-to-digital conversion processing on the analog current signals to obtain digital sound signals, and send the digital sound signals to the processor; The alarm module is used to issue an alarm when receiving an alarm signal sent by the processor.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by one or more processors, the steps of the method for recognizing the crying sound of a baby according to any one of claims 1 to 4 are implemented.