Voice recognition method and device, electronic equipment, storage medium and program product

Abstract

The invention provides a sound recognition method and device, electronic equipment, a storage medium and a program product, and the method comprises the steps: carrying out the feature extraction of to-be-processed audio data in response to the obtained to-be-processed audio data, and generating an input vector; inputting the input vector into the trained unsupervised auto-encoder model, so as to determine the abnormal score of the to-be-processed audio data according to the unsupervised auto-encoder model; and determining an identification result of the to-be-processed audio data according to a comparison result of the abnormal score and a preset threshold value.

Description

Technical Field

[0001] The present application relates to the field of sound recognition technology, and in particular to a sound recognition method, device, electronic device, storage medium and program product. Background Art

[0002] In the current field of sound recognition, deep learning networks with supervised learning tasks are generally used for sound recognition tasks. However, when such network models face complex classification problems, it is difficult for a shallow network to achieve the desired effect, and adding a network is prone to overfitting and other problems. For example, if the collected signal is a non-stationary signal, the result obtained by high-resolution estimation is often not accurate; and the algorithm has poor robustness. If there is reflection in the environment, the result obtained by high-resolution estimation will have a large error.

[0003] In the electric power sector, the sound recognition environment for transformers and other devices often has too much interference from environmental noise. At the same time, the vibration signal is a nonlinear and non-stationary signal with weak signals and insufficient samples. Therefore, the sound recognition model in related technologies has low accuracy and poor robustness, making it difficult to achieve an ideal user experience. Summary of the invention

[0004] In view of this, the present application proposes a sound recognition method, device, electronic device, storage medium and program product to solve or partially solve the above problems.

[0005] Based on the above purpose, the present application provides a sound recognition method, comprising: In response to acquiring the audio data to be processed, extracting features from the audio data to be processed and generating an input vector; Inputting the input vector into a trained unsupervised autoencoder model to determine an abnormality score of the audio data to be processed according to the unsupervised autoencoder model; The recognition result of the audio data to be processed is determined according to the comparison result of the abnormality score and the preset threshold.

[0006] In some exemplary embodiments, the training process of the unsupervised autoencoder model includes: Obtaining at least one training input vector corresponding to normal sound data; Inputting the training input vector into the autoencoder model to perform data dimension reduction to obtain a reconstructed training output vector, and taking the reconstruction error between the training output vector and the training input vector as a training score; In response to the training score failing to satisfy a preset loss function, adjusting the model parameters when determining the reconstruction error, and repeating this cycle until a preset termination condition is satisfied to complete the training of the unsupervised autoencoder model; The training score is specifically:

[0007] in, is the training input vector, for The model parameters are The training score at The model parameters to be adjusted to train the model, is the dimension of the training input vector, is the encoding process of the autoencoder, is the decoding process of the autoencoder.

[0008] In some exemplary embodiments, the loss function is specifically:

[0009] in, is the amount of normal sound data.

[0010] In some exemplary embodiments, extracting features from the audio data to be processed and generating an input vector comprises: The logarithmic Mel spectrum is used to extract features of the audio data to be processed, and a corresponding input vector is generated.

[0011] In some exemplary embodiments, the comparison result between the abnormality score and a preset threshold value includes: In response to the abnormality score being less than or equal to the preset threshold, determining that the sound corresponding to the audio data to be processed is a normal sound; In response to the abnormality score being greater than the preset threshold, it is determined that the sound corresponding to the audio data to be processed is an abnormal sound.

[0012] In some exemplary embodiments, in response to obtaining the audio data to be processed, the step includes: In response to detecting a sound signal, the sound signal is converted from analog to digital to obtain the audio data to be processed.

[0013] Based on the same concept, the present application also provides a sound recognition device, including: An acquisition module, configured to extract features of the audio data to be processed and generate an input vector in response to acquiring the audio data to be processed; A processing module, configured to input the input vector into a trained unsupervised autoencoder model, so as to determine an abnormality score of the audio data to be processed according to the unsupervised autoencoder model; The comparison module is used to determine the recognition result of the audio data to be processed according to the comparison result of the abnormality score and the preset threshold.

[0014] Based on the same concept, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements any of the methods described above when executing the program.

[0015] Based on the same concept, the present application also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to implement any of the methods described above.

[0016] Based on the same concept, the present application also provides a computer program product, including computer program instructions, which, when executed on a computer, enable the computer to execute any of the above methods.

[0017] As can be seen from the above, the present application provides a method, device, electronic device, storage medium and program product for sound recognition, including: in response to obtaining the audio data to be processed, extracting features from the audio data to be processed and generating an input vector; inputting the input vector into the trained unsupervised autoencoder model to determine the abnormal score of the audio data to be processed according to the unsupervised autoencoder model; and determining the recognition result of the audio data to be processed according to the comparison result of the abnormal score and the preset threshold. The present application uses an autoencoder model trained in an unsupervised manner to recognize sound, and uses an unsupervised autoencoder to cluster normal sounds to determine the approximate range of normal sounds, so that the sound data within the preset threshold range are considered to be normal sounds. Once the sound data exceeding the preset threshold range is detected, it is marked as abnormal sound, and then the model trained in this unsupervised manner does not need to provide a control group, and the amount of normal sound data is extremely large, so it is easier to train an unsupervised model with higher accuracy, and finally improve the accuracy of sound recognition and robustness in an unsupervised manner, and achieve an ideal user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flowchart of an exemplary method provided in an embodiment of the present application.

[0020] Figure 2 A schematic diagram of the structure of an autoencoder model of an exemplary method provided in an embodiment of the present application.

[0021] Figure 3 A schematic diagram of the structure of an exemplary device provided in an embodiment of the present application.

[0022] Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of this specification more clear, this specification is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0024] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should be understood by people with ordinary skills in the field to which the present application belongs. The "first", "second" and similar words used in the embodiments of the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements, objects or method steps appearing before the word cover the elements, objects or method steps listed after the word and their equivalents, without excluding other elements, objects or method steps. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0025] As described in the background technology section, in the current field of sound recognition, most of the neural network training is carried out in a supervised manner, and the supervised method must first be trained with known training samples in the process of training the model to obtain the optimal model. During the training, the input is generally a pair of samples and their labeled comparisons, and the training process is also a process that enables the model to gradually label the samples. Therefore, its core is a classification process, which uses the existing training samples to train an optimal model, and then uses this model to map all inputs to corresponding outputs, and makes simple judgments on the outputs to achieve the purpose of prediction and classification, and also has the ability to predict and classify unknown data. Just like having exercises with standard answers and then taking an exam, the accuracy is higher than that of taking an exam with exercises without answers. It can be seen that the current sound recognition method generally requires a data source with a certain degree of stability, and requires a sufficient number of sample sets and control groups with sufficient depth during training to complete supervised model training.

[0026] However, in some special application scenarios, such as the sound recognition environment for transformers and other devices in the power field mentioned in the background technology, the sound signal obtained by the transformer itself or in combination with the overall environment is generally a nonlinear and non-stationary signal, and the signal itself is weak. More importantly, since the timing of a fault is difficult to predict and the sound data is difficult to collect in real time when a fault occurs (for example, when a short circuit or other fault occurs, the protection device will generally quickly disconnect the circuit as a whole), it is difficult to obtain samples and the number of control groups is seriously insufficient. If an insufficient number of samples is used for supervised model training, the trained network model will be too shallow and it will be difficult to achieve the desired effect, while artificially increasing the network excessively will easily cause problems such as overfitting. This ultimately leads to the low accuracy and poor robustness of the relevant sound recognition model in specific fields, making it difficult to achieve an ideal user experience.

[0027] In combination with the above actual situation, the embodiment of the present application provides a sound recognition solution. The present application uses an autoencoder model trained in an unsupervised manner to recognize sounds, and uses an unsupervised autoencoder to cluster normal sounds to determine the approximate range of normal sounds, so that the sound data within the preset threshold range is considered to be normal sounds. Once the sound data exceeding the preset threshold range is detected, it is marked as an abnormal sound. The model trained in this unsupervised manner does not need to provide a control group, and the amount of normal sound data is extremely large, so it is easier to train an unsupervised model with higher accuracy. Finally, the accuracy of sound recognition is improved and the robustness is improved in an unsupervised manner, achieving an ideal user experience.

[0028] refer to Figure 1 , which is a flow chart of a sound recognition method provided by an exemplary embodiment of the present application.

[0029] The sound recognition method specifically comprises the following steps: Step 102: In response to obtaining the audio data to be processed, feature extraction is performed on the audio data to be processed and an input vector is generated.

[0030] In this step, the audio data to be processed is the sound data of the data type that needs to be recognized, which can be understood as a section of audio data after completing analog-to-digital conversion.

[0031] In some embodiments, the audio data to be processed may be directly acquired in a data format, such as an operator copying the corresponding data through a removable hard disk or other devices, or directly acquiring data from a third party organization. It may also be data type data obtained by directly acquiring sound in the current scene through a sound capture module, an analog-to-digital conversion module, etc., and converting the sound, such as using a microphone or other structure to capture sound, complete analog-to-digital conversion, and generate the corresponding audio data to be processed.

[0032] In some embodiments, in order to detect and identify the sound in the scene more immediately, the sound signal can be directly captured and converted through the corresponding module structure, so as to quickly and timely obtain the audio data to be processed. That is, in some embodiments, the response to obtaining the audio data to be processed includes: in response to detecting the sound signal, performing analog-to-digital conversion on the sound signal to obtain the audio data to be processed. Among them, the sound signal is the sound emitted by the target device (such as the target transformer), which is generally a waveform signal. These sound signals can be quickly and accurately detected and captured by module structures such as microphones.

[0033] After that, sound feature extraction is an important step in the fields of speech recognition, music information retrieval, speech synthesis, etc. Common sound feature extraction methods include Line Spectrum, Log-mel, and Mel Frequency Cepstral Coefficients (MFCCs). Through the corresponding feature extraction method, the audio data to be processed can be refined into the corresponding vector data, that is, the input vector.

[0034] In some embodiments, the method of converting a sound signal into a spectrogram using a logarithmic Mel spectrum is to divide the sound signal into several time periods, filter the signal in each time period through a Mel filter group, and then take the logarithm to obtain a spectrogram. Since a nonlinear transformation is performed in the process of obtaining the logarithmic Mel spectrum, it can reduce the nonlinear effect on the perception of low-frequency and high-frequency signals, thereby obtaining almost the same perception ability for different frequencies. Furthermore, in the process of sound recognition for a specific field, the use of the logarithmic Mel spectrum can improve the recognition accuracy to a certain extent. Thus, in some embodiments, the feature extraction of the audio data to be processed and the generation of an input vector include: using the logarithmic Mel spectrum to extract features from the audio data to be processed, and generating a corresponding input vector.

[0035] After that, the input vector is constructed according to the logarithmic Mel spectrum. The specific process can be: the horizontal coordinate, vertical coordinate and RGB color value corresponding to each pixel point in the logarithmic Mel spectrum are converted into The input is a Mel spectrum image. Arrange each pixel of the image from left to right and from top to bottom to form a set of input vectors.

[0036] Step 104: input the input vector into the trained unsupervised autoencoder model to determine the abnormality score of the audio data to be processed according to the unsupervised autoencoder model.

[0037] In this step, after the input vector is obtained, it can be transmitted to the unsupervised autoencoder model that has completed the training. Through this model, whether the current input vector is abnormal can be scored, that is, the abnormal score can be determined. Among them, compared with the aforementioned supervised model, the core purpose of the unsupervised model is to perform clustering. At the same time, the unsupervised model often participates in deep learning, performs feature extraction, or simply adopts layer clustering or item clustering to reduce the dimension of data features. The autoencoder itself plays an important role in dimensionality reduction. During the training process, the model only obtains a large amount of normal sound data. In the process of constantly adjusting the model parameters, these normal sound data are close to a range, so as to achieve the clustering effect, and this range is the set threshold range, or it can be simply understood as the threshold range corresponding to the loss function. After completing the training, for the newly obtained input vector, if the result obtained according to the model is within the range, the corresponding data is considered to be normal data, otherwise it is considered to be abnormal data. In some other embodiments, for unsupervised network models, deep networks based on unsupervised learning tasks can be used to extract high-order correlations in observed data, including restricted Boltzmann machines (RBM), deep belief networks (DBN), and denoising autoencoders, etc. These models are usually used to learn features that are more robust and effective than input features when used for unsupervised learning tasks. Therefore, in some embodiments, restricted Boltzmann machines and deep belief networks can also be used to construct unsupervised models.

[0038] In some embodiments, the specific training process of the unsupervised autoencoder model may be: obtaining a training input vector corresponding to at least one normal sound data; inputting the training input vector into the autoencoder model for data dimension reduction to obtain a reconstructed training output vector, and using the reconstruction error between the training output vector and the training input vector as a training score; in response to the training score failing to satisfy a preset loss function, adjusting the model parameters when determining the reconstruction error, and repeating this cycle until a preset termination condition is satisfied to complete the training of the unsupervised autoencoder model; wherein the training score is specifically:

[0039] in, is the training input vector, for The model parameters are The training score at The model parameters to be adjusted to train the model, is the dimension of the training input vector, is the encoding process of the autoencoder, is the decoding process of the autoencoder.

[0040] In this embodiment, the autoencoder is a multi-layer feedforward neural network whose expected output is the input itself. Because there is no reconstruction error in the classification task, the autoencoder is not used in most classification scenarios (that is, the autoencoder cannot be used in most supervised models). However, the autoencoder plays an important role in dimensionality reduction. For example, in many fields such as pattern recognition and data mining, the problem of "dimensionality disaster" is often faced, and dimensionality reduction methods are needed to map high-dimensional features to low-dimensional space for processing. Figure 2 As shown in the figure, the structure of the autoencoder is divided into an encoder E and a decoder D. The input vector is encoded into a hidden vector through a feedforward neural network and then decoded into an output vector. Through this process, an effective representation of the input vector can be learned.

[0041]

[0042] in, is the input vector, of course, it is the training input vector during training; is the vector reconstructed by the model, that is, the output vector or the training output vector during training; is the encoding process of the autoencoder; is the decoding process of the autoencoder. After that, for training, its purpose is also to minimize the difference between the input vector and the output vector. That is, the optimization process of the autoencoder is as follows:

[0043] in, Usually used Of course, if it contains multiple hidden layers, it is a deep autoencoder. Through multiple layers of encoders and decoders, the deep autoencoder can effectively represent the complex distribution of input X.

[0044] Specifically in this embodiment, when training the model, a large amount of normal sound data will be obtained. For the normal sound data, the corresponding training input vector can be obtained by the method of step 102, wherein the training input vector can be obtained by the method of logarithmic Mel spectrum. Then, according to the above, the input vector obtained according to the logarithmic Mel spectrum can be specifically in the form of a set, that is, a set of vectors can be obtained. After that, the reconstruction error is the autoencoder first models the normal sound according to the above, then adjusts the model parameters in the reconstruction error of the model according to the normal sound, and finally uses the reconstruction error as the abnormality score. , the formula for training reconstruction error is:

[0045] in, is the training output vector, and The purpose of training is to adjust , so that the model can finally achieve the purpose of unsupervised clustering and achieve the desired effect of the model.

[0046] Afterwards, the conditions for completing the training, or the termination conditions of the training, can be determined by setting the loss function. According to the above-mentioned purpose of the autoencoder, the core is to , applied to the specific scenario of this embodiment, the loss function can be:

[0047] in, is the number of normal sound data. The process of model fitting is the process of minimizing the abnormal score of normal sound.

[0048] Of course, in some embodiments, the termination condition of training may also be set to completing a set number of cycles (eg, 10,000 cycles), etc. Of course, the two termination conditions may also be combined.

[0049] Finally, the trained unsupervised autoencoder model can be used to determine the abnormality score of the input vector. The determination process can actually be considered as going through the training process of the unsupervised autoencoder model again. That is, through the above formula conduct The difference is that when it is formally applied These are the values ​​that have been trained and do not need to be adjusted.

[0050] Step 106: Determine the recognition result of the audio data to be processed according to the comparison result between the abnormality score and the preset threshold.

[0051] In this step, after the abnormality score is obtained in step 104, it can be compared with a preset threshold to determine whether the sound corresponding to the audio data to be processed is abnormal. The preset threshold here can be a threshold actively input or set by the operator, or a threshold determined during the unsupervised autoencoder training process.

[0052] Afterwards, in some embodiments, the comparison results can generally be divided into normal sounds and abnormal sounds. The comparison method can be that when the obtained abnormality score is less than or equal to the preset threshold, it is considered to be a normal sound, that is, the sound corresponding to the audio data to be processed is within the clustering range of the unsupervised autoencoder. When the obtained abnormality score is greater than the preset threshold, it is considered to be an abnormal sound, that is, the sound corresponding to the audio data to be processed is outside the clustering range of the unsupervised autoencoder. That is, in some embodiments, the comparison result based on the abnormality score and the preset threshold includes: in response to the abnormality score being less than or equal to the preset threshold, it is determined that the sound corresponding to the audio data to be processed is a normal sound; in response to the abnormality score being greater than the preset threshold, it is determined that the sound corresponding to the audio data to be processed is an abnormal sound.

[0053] Finally, the recognition result can be displayed and output on the corresponding device to give the operator corresponding feedback. Of course, in other embodiments, the output method of the recognition result may not be limited to output display, and it can also be used to store, display, use or reprocess the recognition result. According to different application scenarios and implementation needs, the specific output method of the recognition result can be flexibly selected.

[0054] For example, for an application scenario in which the method of this embodiment is executed on a single device, the recognition result can be directly output in a displayed manner on a display component (display, projector, etc.) of the current device, so that the operator of the current device can directly see the content of the recognition result from the display component.

[0055] For another example, for an application scenario where the method of this embodiment is executed on a system composed of multiple devices, the recognition result can be sent to other preset devices as receivers in the system, that is, the synchronization terminal, through any data communication method (wired connection, NFC, Bluetooth, wifi, cellular mobile network, etc.), so that the synchronization terminal can perform subsequent processing on it. Optionally, the synchronization terminal can be a preset server, which is generally set up in the cloud as a data processing and storage center, which can store and distribute the recognition results; wherein the recipient of the distribution is the terminal device, and the holder or operator of these terminal devices can be the superior supervisor, maintenance personnel, etc. of the sound generating device.

[0056] For another example, in an application scenario where the method of this embodiment is executed on a system composed of multiple devices, the recognition result can be sent directly to a preset terminal device through any data communication method, and the terminal device can be one or more of the ones listed in the preceding paragraphs.

[0057] In a specific embodiment, taking the sound recognition of a transformer as an example, the specific sound recognition process can be: first, a sound collection device such as a voiceprint sensor is deployed near the transformer to collect the voiceprint signal when the transformer is running, and the collected equipment vibration acoustic signal is segmented and labeled in the time and frequency domain. The annotation content includes the equipment model, operating status parameters, operating environment influencing factors, etc. In order to ensure the recognition accuracy of deep learning, the annotated audio signal is preprocessed, and the information unrelated to the equipment voiceprint is separated and filtered. Construct the selected unsupervised autoencoder model network, input the voiceprint signal into the neural network model, and finally obtain an unsupervised autoencoder model with a high success rate of fault diagnosis and identification through iterative processes such as adjusting the parameters of the neural network model, training, and testing. Finally, the trained unsupervised autoencoder model can be transplanted into the relevant power grid main equipment fault diagnosis and early warning software with good human-computer interaction.

[0058] It can be seen from the above embodiments that a sound recognition method provided by the embodiment of the present application includes: in response to obtaining the audio data to be processed, extracting features of the audio data to be processed and generating an input vector; inputting the input vector into the unsupervised autoencoder model that has completed the training, so as to determine the abnormal score of the audio data to be processed according to the unsupervised autoencoder model; and determining the recognition result of the audio data to be processed according to the comparison result of the abnormal score and the preset threshold. The present application uses the autoencoder model trained in an unsupervised manner to recognize the sound, and uses the unsupervised autoencoder to cluster the normal sound to determine the approximate range of the normal sound, so that the sound data within the preset threshold range is considered to be normal sound. Once the sound data exceeding the preset threshold range is detected, it is marked as abnormal sound, and then the model trained in this unsupervised manner does not need to provide a control group, and the amount of normal sound data is extremely large, so it is easier to train an unsupervised model with higher accuracy, and finally improve the accuracy of sound recognition and robustness through an unsupervised manner, and achieve an ideal user experience.

[0059] It should be noted that the method of the embodiment of the present application can be performed by a single device, such as a computer or server. The method of the embodiment of the present application can also be applied to a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present application, and the multiple devices will interact with each other to complete the described method.

[0060] It should be noted that the above describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0061] Based on the same concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a sound recognition device.

[0062] refer to Figure 3 , the voice recognition device comprises: The acquisition module 210 is used for extracting features of the audio data to be processed and generating an input vector in response to acquiring the audio data to be processed.

[0063] The processing module 220 is used to input the input vector into the trained unsupervised autoencoder model to determine the abnormality score of the audio data to be processed according to the unsupervised autoencoder model.

[0064] The comparison module 230 is used to determine the recognition result of the audio data to be processed according to the comparison result of the abnormality score and the preset threshold.

[0065] In some exemplary embodiments, the training process of the unsupervised autoencoder model includes: Obtaining at least one training input vector corresponding to normal sound data; Inputting the training input vector into the autoencoder model to perform data dimension reduction to obtain a reconstructed training output vector, and taking the reconstruction error between the training output vector and the training input vector as a training score; In response to the training score failing to satisfy a preset loss function, adjusting the model parameters when determining the reconstruction error, and repeating this cycle until a preset termination condition is satisfied to complete the training of the unsupervised autoencoder model; The training score is specifically:

[0066] in, is the training input vector, for The model parameters are The training score at The model parameters to be adjusted to train the model, is the dimension of the training input vector, is the encoding process of the autoencoder, is the decoding process of the autoencoder.

[0067] In some exemplary embodiments, the loss function is specifically:

[0068] in, is the amount of normal sound data.

[0069] In some exemplary embodiments, the acquisition module 210 is further configured to: The logarithmic Mel spectrum is used to extract features of the audio data to be processed, and a corresponding input vector is generated.

[0070] In some exemplary embodiments, the comparison module 230 is further configured to: In response to the abnormality score being less than or equal to the preset threshold, determining that the sound corresponding to the audio data to be processed is a normal sound; In response to the abnormality score being greater than the preset threshold, it is determined that the sound corresponding to the audio data to be processed is an abnormal sound.

[0071] In some exemplary embodiments, the acquisition module 210 is further configured to: In response to detecting a sound signal, the sound signal is converted from analog to digital to obtain the audio data to be processed.

[0072] For the convenience of description, the above devices are described in terms of functions divided into various modules. Of course, when implementing the embodiments of the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0073] The device of the above embodiment is used to implement the corresponding sound recognition method in the above embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0074] Based on the same concept, corresponding to any of the above-mentioned embodiments, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the sound recognition method as described in any of the above embodiments is implemented.

[0075] Figure 4 A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 1010, a memory 1020, an input / output interface 1030, a communication interface 1040, and a bus 1050. The processor 1010, the memory 1020, the input / output interface 1030, and the communication interface 1040 are connected to each other through the bus 1050 in the device.

[0076] The processor 1010 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0077] The memory 1020 may be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 1020 may 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 1020 and are called and executed by the processor 1010.

[0078] The input / output interface 1030 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device may include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device may include a display, a speaker, a vibrator, an indicator light, etc.

[0079] The communication interface 1040 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0080] The bus 1050 includes a path that transmits information between the various components of the device (eg, the processor 1010 , the memory 1020 , the input / output interface 1030 , and the communication interface 1040 ).

[0081] It should be noted that, although the above device only shows the processor 1010, the memory 1020, the input / output interface 1030, the communication interface 1040 and the bus 1050, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.

[0082] The electronic device of the above embodiment is used to implement the corresponding sound recognition method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0083] Based on the same concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the sound recognition method described in any of the above embodiments.

[0084] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0085] The computer instructions stored in the storage medium of the above embodiments are used to enable the computer to execute the sound recognition method described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0086] Based on the same concept, corresponding to any of the above-mentioned embodiments, the present application also provides a computer program product, which includes computer program instructions. In some embodiments, the computer program instructions can be executed by one or more processors of a computer so that the computer and / or the processor execute the voice recognition method. Corresponding to the execution subject corresponding to each step in each embodiment of the voice recognition method, the processor that executes the corresponding step can belong to the corresponding execution subject.

[0087] The computer program product of the above embodiment is used to enable the computer and / or the processor to execute the sound recognition method described in any of the above embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0088] A person skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present application (including the claims) is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0089] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present application difficult to understand, the known power / ground connections to the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. In addition, the device may be shown in the form of a block diagram to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform on which the embodiments of the present application are to be implemented (that is, these details should be fully within the scope of understanding of those skilled in the art). Where specific details (e.g., circuits) are set forth to describe exemplary embodiments of the present application, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details or with changes in these specific details. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0090] Although the present application has been described in conjunction with specific embodiments of the present application, many alternatives, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the discussed embodiments.

[0091] The embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of the present application.

Claims

1. A sound recognition method, characterized in that: include: In response to acquiring the audio data to be processed, extracting features from the audio data to be processed and generating an input vector; Inputting the input vector into a trained unsupervised autoencoder model to determine anomaly scores of the audio data to be processed according to the unsupervised autoencoder model; The recognition result of the audio data to be processed is determined according to the comparison result of the abnormality score and the preset threshold.

2. The method according to claim 1, characterized in that The training process of the unsupervised autoencoder model includes: Obtaining at least one training input vector corresponding to normal sound data; Inputting the training input vector into the autoencoder model to perform data dimension reduction to obtain a reconstructed training output vector, and taking the reconstruction error between the training output vector and the training input vector as a training score; In response to the training score failing to satisfy a preset loss function, adjusting the model parameters when determining the reconstruction error, and repeating this cycle until a preset termination condition is satisfied to complete the training of the unsupervised autoencoder model; The training score is specifically: in, is the training input vector, for The model parameters are The training score at The model parameters to be adjusted to train the model, is the dimension of the training input vector, is the encoding process of the autoencoder, is the decoding process of the autoencoder.

3. The method according to claim 2, characterized in that The loss function is specifically: in, is the amount of normal sound data.

4. The method according to claim 1, characterized in that: The step of extracting features from the audio data to be processed and generating an input vector comprises: The logarithmic Mel spectrum is used to extract features of the audio data to be processed, and a corresponding input vector is generated.

5. The method according to claim 1, characterized in that: The comparison result between the abnormality score and the preset threshold value includes: In response to the abnormality score being less than or equal to the preset threshold, determining that the sound corresponding to the audio data to be processed is a normal sound; In response to the abnormality score being greater than the preset threshold, it is determined that the sound corresponding to the audio data to be processed is an abnormal sound.

6. The method according to claim 1, characterized in that The step of responding to obtaining the audio data to be processed includes: In response to detecting a sound signal, the sound signal is converted from analog to digital to obtain the audio data to be processed.

7. A voice recognition device, characterized in that: include: An acquisition module, configured to extract features of the audio data to be processed and generate an input vector in response to acquiring the audio data to be processed; A processing module, configured to input the input vector into a trained unsupervised autoencoder model, so as to determine an abnormality score of the audio data to be processed according to the unsupervised autoencoder model; The comparison module is used to determine the recognition result of the audio data to be processed according to the comparison result of the abnormality score and the preset threshold.

8. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to implement the method according to any one of claims 1 to 6.

10. A computer program product, characterized in that The method comprises computer program instructions, which, when executed on a computer, cause the computer to execute the method according to any one of claims 1 to 6.

Images