Depression state detection method and device based on artificial intelligence, equipment and medium

Abstract

The invention discloses a depression state detection method and device based on artificial intelligence, equipment and a medium, and relates to the technical field of artificial intelligence and medical health, and the method can extract user voice features and depression voice features from user voice, fuse the user voice features and the depression voice features to obtain a fused feature vector, and obtain a depression state detection result. And inputting the fusion feature vector into a trained depression state detection model, fusing voice features of multiple dimensions to carry out depression tendency analysis, and generating a depression state detection result in combination with depression tendency performance of user voice on the features of multiple dimensions. The depressive state detection method based on artificial intelligence is higher in accuracy.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and medical health technology, and in particular to an artificial intelligence-based depression state detection method, device, equipment and medium. Background Art

[0002] With the rise of artificial intelligence technologies related to machine learning and neural networks, artificial intelligence has also been widely used in the field of smart medical care, for example, providing technical support for disease auxiliary diagnosis, health management, remote consultation and other functions.

[0003] Depression is a common mental health problem, usually manifested as persistent low mood, loss of interest and powerlessness. It may be caused by a variety of reasons, such as life stress and genetic factors. Depression not only affects mood, but also interferes with daily life, work and interpersonal relationships. In severe cases, professional treatment is required.

[0004] Although public awareness of mental health is improving, many people still choose to hide or ignore treatment due to traditional beliefs, which exacerbates the problem. Therefore, it is very important to promptly detect people with depression through a variety of different methods.

[0005] At present, most intelligent depression detection models based on deep learning technology perform detection and analysis based on a single feature, and the accuracy of the detection results is poor. Summary of the invention

[0006] The present invention provides a method, device, equipment and medium for detecting depression state based on artificial intelligence to solve the problem of low accuracy of detection results.

[0007] In a first aspect, a method for detecting a depressive state based on artificial intelligence is provided, the method comprising: acquiring user voice, extracting user voice features and depressive voice features from the user voice; fusing the user voice features and the depressive voice features to obtain a fused feature vector; and inputting the fused feature vector into a trained depressive state detection model to obtain a depressive state detection result.

[0008] In a second aspect, a depression state detection device based on artificial intelligence is provided, and the device includes: an acquisition module, used to acquire user voice, and extract user voice features and depression voice features from the user voice; a feature fusion module, used to fuse the user voice features and the depression voice features to obtain a fused feature vector; a detection module, used to input the fused feature vector into a trained depression state detection model to obtain a depression state detection result.

[0009] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned artificial intelligence-based depressive state detection method when executing the computer program.

[0010] In a fourth aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned artificial intelligence-based depression state detection method are implemented.

[0011] In the scheme implemented by the above-mentioned artificial intelligence-based depression state detection method, device, equipment and medium, user voice features and depression voice features can be extracted from the user voice, and the user voice features and depression voice features are fused to obtain a fused feature vector. The fused feature vector is input into a trained depression state detection model, and the voice features of multiple dimensions are fused to carry out depression tendency analysis. The depression tendency performance of the user voice in multiple dimensions is combined to generate a depression state detection result. Compared with depression detection based on a single type of feature, the artificial intelligence-based depression state detection method proposed in this application has higher accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0013] Figure 1 is a schematic diagram of an application environment of a method for detecting depression based on artificial intelligence in one embodiment of the present invention; Figure 2 It is a flow chart of a method for detecting depression based on artificial intelligence in one embodiment of the present invention; Figure 3 is a flow chart of a method for detecting depression based on artificial intelligence in another embodiment of the present invention; Figure 4 is a flow chart of a method for detecting depression based on artificial intelligence in another embodiment of the present invention; Figure 5 is a flow chart of a method for detecting depression based on artificial intelligence in another embodiment of the present invention; Figure 6 is a structural schematic diagram of a depression state detection device based on artificial intelligence in one embodiment of the present invention; Figure 7 It is a structural diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0014] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0015] The artificial intelligence-based depression state detection method provided in the embodiment of the present invention can be applied in Figure 1 In the application environment, the client communicates with the server through the network. The server can obtain the user voice input / uploaded by the user in the client interaction interface, extract the user voice features and depression voice features from the user voice, fuse the user voice features and depression voice features to obtain a fused feature vector, input the fused feature vector into the trained depression state detection model, and obtain the depression state detection result. The server can return the depression state detection result to the client so that the client can display the depression state detection result to the user through SMS push, pop-up reminder, voice alarm, etc. This method detects depression by fusing user voice features and depression voice features, realizes depression state detection with multi-dimensional feature fusion, and can pay attention to the features of user audio in different dimensions when generating detection results, thereby improving detection accuracy.

[0016] The client and the server can be the front end and the back end of the intelligent consultation platform, wherein the client can include but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server can be implemented with an independent server or a server cluster composed of multiple servers. The present invention is described in detail below through specific embodiments.

[0017] See also Figure 2 As shown, Figure 2 A flowchart of a method for detecting depression based on artificial intelligence provided by an embodiment of the present invention includes the following steps: S201, obtaining user voice, and extracting user voice features and depression voice features from the user voice.

[0018] User voice can be the voice recorded by the user through terminal devices such as mobile phones and voice recorders (i.e. the above-mentioned client), or real-time voice input on the consultation platform, or the user voice obtained by the application software from the user voice stored in the background, such as intelligent voice commands issued by the user, voice call records, etc. The user voice is obtained by the terminal device under the conditions of authorization.

[0019] For example, an application software with health monitoring function installed on the terminal obtains the user voice collected by the terminal in real time, extracts user voice features and depression voice features from the user voice, so as to further determine whether the user has depression tendencies and remind the user in time.

[0020] For another example, a recording device may be used to record the conversation during an interview. During the conversation between the interviewer and the user, voiceprint recognition technology may be used to distinguish the voices of the interviewer and the user, and a speaker audio distinction model may be used to distinguish the interviewer's audio from the user's audio, thereby collecting the user's voice.

[0021] In the embodiment of the present application, the user voice feature is a feature associated with depression tendency extracted from the user voice in a manner that simulates human ear perception; The depressive speech features are statistical features of the depression indicators in the user's speech.

[0022] Depression indicators may include but are not limited to at least one of the following: pitch, volume, speaking speed, rhythm, etc.

[0023] It should be noted that in the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0024] S202, fusing the user's voice features and the depression voice features to obtain a fused feature vector.

[0025] There are many ways to fuse features, such as directly fusing the user voice features and the depression voice features, or multiplying the user voice features and the depression voice features by corresponding weight values ​​respectively and then fusing the features.

[0026] S203, inputting the fused feature vector into the trained depression state detection model to obtain a depression state detection result.

[0027] The fused feature vector obtained in the above step S202 is input into the trained depression state detection model to obtain a depression state detection result.

[0028] The result of the depression state detection may include two situations: the presence of depression tendency or the absence of depression tendency.

[0029] The original depression state detection model can be constructed based on a deep neural network DNN, or can also be constructed based on other types of neural networks, and this application does not impose any restrictions.

[0030] The trained depression state detection model can predict whether the user's voice shows a depressive tendency in human ear perception based on the user's voice features in the fused feature vector. At the same time, it can analyze whether the user's voice shows a depressive tendency in terms of depression indicators based on the depressive voice features in the fused feature vector. Combined with the depressive tendency performance of the user's voice in multiple dimensions, the detection results are comprehensively generated to improve the accuracy of depression state detection.

[0031] In summary, the artificial intelligence-based depression state detection method proposed in the present application extracts user voice features and depression voice features from the user voice, fuses the user voice features and the depression voice features to obtain a fused feature vector, inputs the fused feature vector into a trained depression state detection model, fuses voice features of multiple dimensions to carry out depression tendency analysis, and combines the depression tendency performance of the user voice in multiple dimensions to generate a depression state detection result. Compared with depression detection based on a single type of feature, the artificial intelligence-based depression state detection method proposed in the present application has higher accuracy.

[0032] Based on the above embodiments, Figure 3 As shown, the above step S201 of “extracting user voice features from user voice” may include the following steps: S301, obtaining a mel-spectrogram of the user's speech.

[0033] S302, inputting the mel-spectrogram into a trained convolutional neural network, and outputting the intermediate layer of the trained convolutional neural network as a user voice feature; the trained convolutional neural network is obtained by training the original convolutional neural network based on a voice sample set with a label, and the label is a depression tendency label corresponding to the voice sample.

[0034] Obtain the audio spectrum of the user's voice, perform Mel frequency cepstral coefficient MFCC conversion on the audio spectrum, and obtain a Mel frequency spectrum graph; The mel-spectrogram is input into a trained convolutional neural network for feature extraction, and the intermediate layer of the trained convolutional neural network is output as the user voice feature in the above step S201.

[0035] The trained convolutional neural network can be roughly divided into input layer, intermediate layer and output layer. The intermediate layer may include convolution layer, activation function layer, pooling layer, fully connected layer, etc. Among them, the convolution layer can convolve the mel spectrum map input by the input layer to extract specific features.

[0036] In the embodiment of the present application, the features output by any intermediate layer may be used as user voice features.

[0037] The mel-spectrogram can better simulate the auditory characteristics of the human ear. The features extracted from the mel-spectrogram can, to a certain extent, reflect the characteristic performance of the user's voice perceived by the human ear.

[0038] The trained convolutional neural network used in the embodiments of the present application is obtained by training the original convolutional neural network based on a speech sample set with labels, wherein the labels are depressive tendency labels corresponding to the speech samples in the speech sample set.

[0039] For example, the voices of different users in different scenarios and with different emotions are pre-processed to produce voice samples. In addition, each voice sample can be artificially labeled, and the label can indicate that the voice sample has a depressive tendency, or that the voice sample does not have a depressive tendency, thereby constructing a voice sample set containing a large number of voice samples and the label of each voice sample.

[0040] When constructing speech samples and labels of speech samples, they can be constructed in the following ways: Based on the communication dialogue between the tester and the psychologist, the speaker audio differentiation model is used to distinguish the tester's voice from the conversation voice, and the tester's voice is preprocessed to construct a voice sample.

[0041] A psychologist's diagnosis result on whether the test person has a tendency to depression can be used as a label for the voice sample.

[0042] For example, when performing model training, a mel-spectrogram of a speech sample is first generated, and the mel-spectrogram is input into the original convolutional neural network. The result output by the original convolutional neural network is whether the speech sample shows a depressive tendency; the original convolutional neural network is trained through a loss function between the result output by the original convolutional neural network and the sample label until the loss value corresponding to the trained convolutional neural network is less than a preset loss threshold.

[0043] The trained convolutional neural network is used to learn features related to depressive tendencies from the mel-spectrogram.

[0044] In some embodiments, the original convolutional neural network can be trained by performing model training based on a supervised training method.

[0045] In the embodiment of the present application, the mel-spectrogram of the user's voice is used as input data for extracting user voice features. The mel-spectrogram better simulates the auditory characteristics of the human ear. The features extracted from the mel-spectrogram can reflect the characteristic performance of the user's voice perceived by the human ear to a certain extent. The original convolutional neural network for extracting user voice features is trained using labeled voice samples, so that the trained convolutional neural network can learn features associated with depressive tendencies from the mel-spectrogram, so that the features output by the middle layer of the convolutional neural network, that is, the features used as user voice features, can include the performance of the user's voice in depressive tendencies.

[0046] Based on the above embodiments, Figure 4 As shown, when the middle layer is a fully connected layer, the user voice feature in the above step S202 is represented in the form of a vector, and the "fusing the user voice feature and the depression voice feature to obtain a fused feature vector" in the above step S202 may include the following steps: S401, performing vector representation on the depressive speech feature to obtain a depressive speech feature vector.

[0047] In the embodiment of the present application, the depression voice feature is a statistical feature of a depression index, for example, a characteristic value of a depression index is extracted from the user's voice using a statistical formula.

[0048] After extracting the characteristic value of the depression index, the characteristic value of the depression index can be encoded using an autoencoder network to generate a characteristic vector of the depression index, that is, a depression speech characteristic vector.

[0049] For example, the user's voice is segmented, and the characteristic value of each depression indicator is calculated from each voice segment. The characteristic values ​​of the same depression indicator are normalized and input into the auto-encoder network. Then the vector of the middle layer of the network is used as the depressive voice feature vector.

[0050] Among them, the autoencoder network can be trained based on an unsupervised training method so that the output of the network can restore the input data.

[0051] S402, concatenating the depression speech feature vector and the user speech feature to obtain the fused feature vector.

[0052] The depressed speech feature vector and the fully connected layer output of the convolutional neural network trained in the above step S301, that is, the user speech feature in vector form, are concatenated to obtain a fused feature vector.

[0053] The embodiment of the present application encodes the characteristic value of the depression index to generate a depression speech feature vector, so as to splice the depression speech feature vector and the user speech feature in vector form to achieve the fusion of features of different dimensions.

[0054] Based on the above embodiments, Figure 5 As shown, the original depression state detection model can be trained in the following manner to obtain the trained depression state detection model used in the above step S203: S501, inputting the fused feature vector of the speech sample into the original depression state detection model to obtain the depression state detection result of the speech sample.

[0055] S502, combining the loss function between the depression state detection result of the speech sample and the depression tendency label corresponding to the speech sample, training the original depression state detection model until the value of the loss function is less than a preset loss threshold, thereby obtaining a trained depression state detection model.

[0056] The original depression state detection model is trained using a labeled training sample set, and the original depression state detection model can be constructed based on a DNN network.

[0057] The training sample set may include speech samples and labels of the speech samples, where the labels are used to indicate that the speech samples show depressive tendencies or are used to indicate that the speech samples do not show depressive tendencies.

[0058] The fused feature vector of the speech sample can be obtained in the following way: A mel-spectrogram of a speech sample is obtained, and the mel-spectrogram is input into a trained convolutional neural network to obtain a fully connected layer output of the trained convolutional neural network. Based on the speech sample, a characteristic value of a depression index is obtained, and the characteristic value of the depression index is input into an autoencoder network to obtain a depressive speech feature vector, and the depressive speech feature vector is concatenated with the fully connected layer output of the trained convolutional neural network to obtain a fused feature vector of the speech sample.

[0059] When training the original depression state detection model, the fused feature vector of the speech sample is input into the original depression state detection model to obtain the depression state detection result of the speech sample, which may include the presence of depression tendency or the absence of depression tendency.

[0060] The loss function between the depression state detection result of the speech sample and the label of the speech sample is calculated, and the original depression state detection model is trained using the loss function until the value of the loss function between the depression state detection result output by the trained depression state detection model and the speech sample label is less than a preset loss threshold.

[0061] It should be noted that the preset loss threshold in this embodiment may be the same as or different from the preset loss threshold mentioned in the above embodiment, and may be set as required, and this application does not impose any limitation thereto.

[0062] Among them, when training the original depression state detection model, the training can be based on a supervised training method.

[0063] The embodiment of the present application trains the original depression state detection model based on a fused feature vector that integrates the user's voice features and the depression voice features and a label indicating whether the voice sample shows a depressive tendency, so that the trained depression state detection model can integrate the multi-dimensional features presented by the user's voice features and the depression voice features, comprehensively analyze whether the user has a depressive tendency, and output an accurate depression state detection result.

[0064] Based on the above embodiments, the depressive speech features may include but are not limited to at least one of the following: tone features, volume features, speech speed features, rhythm features, and the like.

[0065] The pitch feature is the statistical feature of the pitch in the user's voice.

[0066] The tone feature may include a statistical characteristic value of tone intensity, and the statistical characteristic value may include but is not limited to at least one of the following: extreme value, mean, and variance; The volume feature is a statistical feature of the volume in the user's voice, and may include a statistical feature value of the volume; the statistical feature value may include but is not limited to at least one of the following: an extreme value, a mean, and a variance.

[0067] The speech rate feature is the statistical feature of the speech rate in the user's voice, which may include the statistical feature value of the speech rate, the statistical feature value of the pause duration, and the statistical feature value of the pause frequency; the statistical feature value may include the extreme value, the mean, and the variance.

[0068] The musicality features are statistical features of musicality in the user's voice, and may include statistical feature values ​​of fluency and statistical feature values ​​of clarity.

[0069] The above-mentioned extreme values ​​may include maximum values ​​and / or minimum values.

[0070] For example, when the artificial intelligence-based depression state detection model proposed in this application is applied to an intelligent diagnosis and treatment platform, the patient can make voice responses to the diagnosis questions raised by the platform on the intelligent diagnosis and treatment platform. The intelligent diagnosis and treatment platform can obtain the voice audio input by the patient in real time, and call the trained convolutional neural network to extract features related to depression tendency from the mel spectrum corresponding to the voice audio as the user's voice features; At the same time, the extreme values ​​and changes of the depression index of the voice audio at different stages are obtained. The depression index can be determined by analyzing the clinical manifestations of a large number of patients with depression, such as analyzing the objective impact of depression on the user's language expression, and constructing a depression index; the statistical characteristics of the depression index in the voice audio are used as depressive voice features.

[0071] For example, the characteristics of depressive speech may include the extreme values ​​and changes of volume at different stages, the extreme values ​​and changes of pitch at different stages, the speed of speech at different stages, the length and frequency of pauses in speech at different stages, and the fluency and clarity of the patient's speech at different stages, etc.

[0072] In some embodiments, the pitch intensity can be represented by the energy intensity of the user's voice at different frequency components. The pitch feature can be obtained by: The user's voice is segmented to obtain multiple audio segments, and the energy intensity extreme values ​​of each audio segment on different frequency components are obtained respectively; further, the mean of the energy intensity extreme values ​​of each audio segment on different frequency components can be calculated, and / or the variance of the energy intensity extreme values ​​of each audio segment on different frequency components can be calculated.

[0073] In some embodiments, the volume feature may be obtained in the following manner: The volume extreme value of each audio segment of the user's voice is obtained respectively, and the mean of the volume extreme values ​​of all the audio segments can be further calculated, and / or the variance of the volume extreme values ​​of all the audio segments can be calculated.

[0074] In some embodiments, the speech rate feature may be obtained in the following manner: Obtaining the speech rate extreme value of each audio segment of the user's voice respectively, and / or obtaining the variance of the speech rate extreme values ​​of all audio segments, and / or obtaining the mean of the speech rate extreme values ​​of all audio segments; Obtaining the extreme value of the pause duration of each audio segment of the user's voice respectively, and / or obtaining the variance of the extreme values ​​of the pause duration of all audio segments, and / or obtaining the mean of the extreme values ​​of the pause duration of all audio segments; Obtain the pause frequency extreme value of each audio segment of the user's voice respectively, and / or obtain the variance of the pause frequency extreme values ​​of all audio segments, and / or obtain the mean of the pause frequency extreme values ​​of all audio segments.

[0075] In one embodiment, the musical characteristics are obtained by: The fluency of each audio segment of the user's speech is obtained respectively, and the mean and variance of the fluency of all audio segments can be further obtained; The clarity of each audio segment of the user's voice is obtained respectively, and the mean and variance of the clarity of all audio segments are further obtained.

[0076] In the embodiment of the present application, the complete user voice is segmented, and the depression index performance in each audio segment is analyzed; then, starting from the overall situation, the fluctuation of the depression index in the entire user voice is analyzed. In this way, when the depression state detection model performs depression tendency analysis, it considers the depression tendency shown in the user voice characteristics, combines the performance of the depression index in the user voice, and integrates multi-dimensional features to perform depression state detection.

[0077] In the embodiment of the present application, by analyzing the voice extraction features of a user, a depression state detection model is established and trained based on deep learning technology to determine whether the user has a tendency to depression. This model has many advantages. First, the model integrates multi-dimensional voice features, that is, based on user voice features and depression voice features, it is not limited to a single type of feature analysis, so as to capture emotional changes more comprehensively; through deep learning technology, the model can automatically learn and optimize the feature extraction and classification process, improve detection accuracy and robustness; the model supports real-time analysis, and can continuously monitor depression tendencies in daily conversations, which is more flexible and applicable than traditional methods; the intelligent depression detection system established based on this method can provide early warning by analyzing voice features, helping individuals take intervention measures before depressive symptoms worsen.

[0078] 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.

[0079] In one embodiment, a depression state detection device based on artificial intelligence is provided, and the depression state detection device based on artificial intelligence corresponds one-to-one to the depression state detection method based on artificial intelligence in the above-mentioned embodiment.

[0080] like Figure 6 As shown, the artificial intelligence-based depression state detection device 600 includes an acquisition module 601, a feature fusion module 602 and a detection module 603. The detailed description of each functional module is as follows: An acquisition module 601 is used to acquire user voice and extract user voice features and depression voice features from the user voice; A feature fusion module 602, used to fuse the user voice feature and the depression voice feature to obtain a fused feature vector; The detection module 603 is used to input the fused feature vector into a trained depression state detection model to obtain a depression state detection result.

[0081] In some embodiments, an acquisition module is used to obtain a mel-spectrogram of the user's voice; the mel-spectrogram is input into a trained convolutional neural network, and an intermediate layer of the trained convolutional neural network is output as the user's voice feature; the trained convolutional neural network is obtained by training the original convolutional neural network based on a set of labeled voice samples, and the label is a depressive tendency label corresponding to the voice sample.

[0082] In some embodiments, when the middle layer is a fully connected layer, the user voice feature is represented in vector form, and the acquisition module is used to: vectorize the depression voice feature to obtain a depression voice feature vector; and vector concatenate the depression voice feature vector and the user voice feature to obtain the fused feature vector.

[0083] In some embodiments, an artificial intelligence-based depression state detection device trains an original depression state detection model in the following manner: inputting a fused feature vector of a speech sample into the original depression state detection model to obtain a depression state detection result of the speech sample; and training the depression state detection model by combining a loss function between the depression state detection result of the speech sample and a depression tendency label corresponding to the speech sample, until the value of the loss function is less than a preset loss threshold, thereby obtaining the trained depression state detection model.

[0084] In some embodiments, the depressive speech feature includes at least one of the following: a pitch feature, a volume feature, a speech speed feature, and a rhythm feature.

[0085] In some embodiments, the pitch feature includes a statistical feature value of pitch intensity; the volume feature includes a statistical feature value of volume; the speech rate feature includes a statistical feature value of speech rate, a statistical feature value of pause duration, and a statistical feature value of pause frequency; the rhythm feature includes a statistical feature value of fluency and a statistical feature value of clarity. The statistical feature value includes at least one of the following: extreme value, mean value, and variance.

[0086] In some embodiments, the pitch feature is obtained by: segmenting the user voice to obtain audio segments; obtaining the pitch intensity extreme values ​​of each of the audio segments at different frequency components, and determining the mean and variance of the pitch intensity extreme values ​​corresponding to each of the audio segments.

[0087] In summary, the artificial intelligence-based depression state detection device proposed in the present application can extract user voice features and depression voice features from the user voice, fuse the user voice features and the depression voice features to obtain a fused feature vector, input the fused feature vector into a trained depression state detection model, fuse the voice features of multiple dimensions to carry out depression tendency analysis, and combine the depression tendency performance of the user voice in multiple dimensions to generate a depression state detection result. Compared with depression detection based on a single type of feature, the artificial intelligence-based depression state detection method proposed in the present application has higher accuracy.

[0088] In one embodiment, Figure 7 As shown, a computer device 700 is provided, including a memory 701, a processor 702, and a computer program stored in the memory 701 and executable on the processor 702. When the processor executes the computer program, the steps of the artificial intelligence-based depression state detection method described in any of the above embodiments are implemented.

[0089] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the artificial intelligence-based depressive state detection method described in any of the above embodiments are implemented.

[0090] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant description in the aforementioned method embodiment. In order to avoid repetition, they will not be described one by one here.

[0091] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0092] Those skilled in the art 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. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0093] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those 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 invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for detecting depression based on artificial intelligence, characterized in that: The method comprises: Acquire user voice, and extract user voice features and depression voice features from the user voice; fusing the user voice feature and the depression voice feature to obtain a fused feature vector; The fused feature vector is input into a trained depression state detection model to obtain a depression state detection result.

2. The method according to claim 1, characterized in that The extracting the user voice feature from the user voice comprises: Obtaining a Mel-spectrogram of the user's speech; Inputting the Mel-spectrogram into a trained convolutional neural network, and outputting the intermediate layer of the trained convolutional neural network as the user voice feature; The trained convolutional neural network is obtained by training the original convolutional neural network according to a speech sample set with labels, wherein the labels are depression tendency labels corresponding to the speech samples.

3. The method according to claim 2, characterized in that When the intermediate layer is a fully connected layer, the user voice feature is represented in a vector form, and the fusing of the user voice feature and the depression voice feature to obtain a fused feature vector includes: Performing vector representation on the depression speech feature to obtain a depression speech feature vector; The depressed speech feature vector and the user speech feature are concatenated to obtain the fused feature vector.

4. The method according to claim 1, characterized in that The original depression state detection model was trained in the following way: Inputting the fused feature vector of the speech sample into the original depression state detection model to obtain the depression state detection result of the speech sample; The original depression state detection model is trained by combining the loss function between the depression state detection result of the speech sample and the depression tendency label corresponding to the speech sample, until the value of the loss function is less than a preset loss threshold, thereby obtaining the trained depression state detection model.

5. The method according to claim 1, characterized in that The depressive speech feature includes at least one of the following: a pitch feature, a volume feature, a speech speed feature, and a rhythm feature.

6. The method according to claim 5, characterized in that The tone feature includes a statistical characteristic value of tone intensity; The volume feature includes a statistical feature value of the volume; The speech rate features include statistical characteristic values ​​of speech rate, statistical characteristic values ​​of pause duration, and statistical characteristic values ​​of pause frequency; The musical characteristics include statistical characteristic values ​​of fluency and statistical characteristic values ​​of clarity; The statistical characteristic value includes at least one of the following: extreme value, mean value and variance.

7. The method according to claim 6, characterized in that The pitch feature is obtained by: Segmenting the user's voice to obtain audio segments; The extreme values ​​of the pitch intensity of the single audio segment at different frequency components are respectively obtained, and the mean and variance of the extreme values ​​of the pitch intensity corresponding to the single audio segment are determined.

8. A depression state detection device based on artificial intelligence, characterized in that: The device comprises: An acquisition module, used for acquiring user voice, and extracting user voice features and depression voice features from the user voice; A feature fusion module, used for fusing the user voice feature and the depression voice feature to obtain a fused feature vector; The detection module is used to input the fused feature vector into a trained depression state detection model to obtain a depression state detection result.

9. A computer 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 artificial intelligence-based depressive state detection method as claimed in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence-based depressive state detection method as claimed in any one of claims 1 to 7 are implemented.

Images