A feature processing method and terminal based on speech recognition

Abstract

The application relates to the technical field of machine learning, and discloses a depression feature processing method and terminal based on voice recognition, which comprises the following steps: acquiring voice monitoring data, and extracting information features of the voice monitoring data; inputting the information features into a time sequence feature extraction network, wherein the time sequence feature extraction network outputs time sequence features; inputting the time sequence features into an auxiliary information acquisition network, wherein the auxiliary information acquisition network outputs auxiliary information used for assisting in judging depression. Through continuous voice monitoring of a patient, voice monitoring data of the patient is acquired, information features of the voice monitoring data are extracted, time sequence analysis is performed on the information features of the long-time-acquired voice monitoring data, time sequence features are obtained, and whether the patient has depression can be judged according to the time sequence features, so that the burden of medical personnel is reduced, and more depression patients can be diagnosed and treated in time in the early stage.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a feature processing method, terminal, and readable storage medium based on speech recognition. Background Technology

[0002] Depression is a mental illness that endangers human health. It is usually manifested as symptoms such as low mood and loss of interest. In severe cases, it is accompanied by cognitive impairment. Current technology mainly relies on subjective assessments of scales and interviews for the diagnosis of depression, and the final determination is made by clinical judgment of psychiatry.

[0003] However, due to factors such as lack of awareness, delayed diagnosis, or economic constraints, patients often miss the optimal treatment window. Therefore, early identification, diagnosis, and treatment are crucial for the recovery of depression. However, because information on depression requires the involvement of professional medical personnel, and the number of such personnel is limited, many patients with depression fail to receive an effective diagnosis in the early stages.

[0004] Therefore, there is a need for reliable, user-friendly, and rapid automated feature processing methods to provide reference for medical personnel, support them in timely identification of depression, and continuously monitor treatment progress, so as to reduce the burden on medical personnel and enable more patients with depression to receive early diagnosis and timely treatment. Summary of the Invention

[0005] The purpose of this invention is to provide a feature processing method based on speech recognition, which aims to solve the problem that existing technologies for diagnosing depression lack auxiliary tools and information and rely on manual intervention.

[0006] The technical solution adopted by this invention to solve the technical problem is as follows:

[0007] This invention provides a feature processing method based on speech recognition, the feature processing method based on speech recognition includes:

[0008] Acquire voice monitoring data and extract information features from the voice monitoring data;

[0009] The information features are input into a temporal feature extraction network, and the temporal feature extraction network outputs temporal features.

[0010] The time-series features are input into an auxiliary information acquisition network, which outputs auxiliary information to assist in the diagnosis of depression.

[0011] Furthermore, the extraction of information features from the voice monitoring data specifically includes:

[0012] Perform a Fast Fourier Transform on the speech monitoring data to obtain the speech audio domain signal;

[0013] Mel filtering is applied to the speech audio domain signal to obtain multiple filtered signals;

[0014] Logarithmic processing is performed on each of the filtered signals to obtain multiple logarithmic results;

[0015] The discrete cosine transform is performed on each of the logarithmic results to obtain cepstral coefficients, which are then used as the information features.

[0016] Furthermore, the temporal feature extraction network includes a long short-term memory network and a temporal convolutional network;

[0017] The step of inputting the information features into a temporal feature extraction network, and the temporal feature extraction network outputting temporal features, specifically includes:

[0018] The information features are input into the long short-term memory network to extract temporal relationship features;

[0019] The temporal relationship features are input into the temporal convolutional network to obtain the temporal refinement features;

[0020] The temporal relationship features and the temporal refinement features are used as the temporal sequence features.

[0021] Furthermore, the step of inputting the information features into the Long Short-Term Memory network to extract temporal relationship features specifically includes:

[0022] In each calculation, the information feature X of this calculation is... t And the intermediate quantity H from the previous calculation t-1 Combined into a computational vector [H] t-1 ,X t ];

[0023] Calculate vector [H] t-1 ,X t Each intermediate vector is multiplied by its corresponding weight, and the product is added to the corresponding bias to obtain four intermediate vectors:

[0024] W o ·[H t-1 ,X t ]+b o ;

[0025] W g ·[H t-1 ,X t ]+b g ;

[0026] W i ·[H t-1 ,X t ]+b i ;

[0027] W f ·[H t-1 ,X t ]+b f ;

[0028] Among them, W o W g W i and W f All are weights, b o b g b i and b f All are biased;

[0029] The four intermediate vectors are input into the activation function to obtain four intermediate parameters o. t g t i t and f t ;

[0030] Based on the intermediate parameter o t g t i t and f t And the intermediate quantity C from the previous calculation. t-1 The intermediate values ​​obtained in this calculation are:

[0031] C t =f t ·C t-1 +g t ·i t ;

[0032] H t =o t HardTanh(C t );

[0033] Where HardTanh represents the HardTanh function;

[0034] The intermediate quantity C in this calculation t and H t Save and record the intermediate value H from this calculation. t As a characteristic of time relationship Output

[0035] The HardTanh function is calculated as follows:

[0036]

[0037] Where x represents the input variable.

[0038] Furthermore, the step of inputting the four intermediate vectors into the activation function to obtain four intermediate parameters g is described. t o t f t and i t Specifically, it includes:

[0039] The intermediate vector W g ·[H t-1 ,X t ]+b g and the intermediate vector W f ·[H t-1 ,X t ]+b f Inputting the HardTanh function yields the intermediate parameter g. t and f t :

[0040] g t =HardTanh(W g ·[H t-1 ,X t ]+b g );

[0041] f t =HardTanh(W f ·[H t-1 ,X t ]+b f );

[0042] The intermediate vector W o ·[H t-1 ,X t ]+b o and the intermediate vector W i ·[H t-1 ,X t ]+b i Inputting the HardSigmoid function yields the intermediate parameter o. t and i t :

[0043] o t =HardSigmoid(W o ·[H t-1 ,X t ]+b o );

[0044] i t =HardSigmoid(W i ·[H t-1 ,X t ]+b i );

[0045] Where HardSigmoid represents the HardSigmoid function, and the calculation formula for the HardSigmoid function is as follows:

[0046]

[0047] Furthermore, the auxiliary information acquisition network includes a cross-attention module and a classification network. The step of inputting the temporal features into the auxiliary information acquisition network and outputting auxiliary information for assisting in the diagnosis of depression specifically includes:

[0048] The temporal features are input into the cross-attention module, and the cross-attention module outputs key features;

[0049] The key features are input into the classification network, and the classification network outputs the auxiliary information.

[0050] Furthermore, the step of inputting the temporal features into the cross-attention module and the cross-attention module outputting key features specifically includes:

[0051] The time refinement feature and the time relationship feature are separated into two parts to obtain separate time refinement features and separate time relationship features.

[0052] A similarity matrix is ​​calculated based on the refined features of the split-time period and the features of the relationship between the split-time periods.

[0053] The similarity matrix is ​​normalized using the softmax function to obtain the weight matrix;

[0054] The key features are calculated based on the weight matrix and the time-separated refinement features.

[0055] Furthermore, the classification network includes an average pooling layer and a fully connected layer;

[0056] The step of inputting the key features into the classification network and the classification network outputting the auxiliary information specifically includes:

[0057] The key features are input into the average pooling layer, and the average pooling layer outputs the pooling result.

[0058] The pooling result is input into the fully connected layer, and the fully connected layer outputs the auxiliary information.

[0059] Furthermore, to achieve the above objectives, the present invention also provides a terminal, the terminal comprising: a memory, a processor, and a speech recognition-based feature processing program stored in the memory and executable on the processor, wherein when the speech recognition-based feature processing program is executed by the processor, the terminal controls the terminal to implement the steps of the speech recognition-based feature processing method described above.

[0060] In addition, to achieve the above objectives, the present invention also provides a readable storage medium storing a speech recognition-based feature processing program, which, when executed by a processor, implements the steps of the speech recognition-based feature processing method described above.

[0061] The present invention, by employing the above technical solution, has the following effects:

[0062] This invention continuously monitors patients' voices to obtain their voice monitoring data, extracts information features from the voice monitoring data, and then performs time-series analysis on the information features of the voice monitoring data acquired over a long period to obtain time-series features. Based on these time-series features, it can help determine whether a patient has depression, thereby reducing the burden on medical personnel and enabling more patients with depression to receive early diagnosis and timely treatment. Attached Figure Description

[0063] Figure 1 This is a flowchart of a feature processing method based on speech recognition in a preferred embodiment of the present invention.

[0064] Figure 2 This is a schematic diagram illustrating the principle of a feature processing method based on speech recognition in a preferred embodiment of the present invention.

[0065] Figure 3 This is a schematic diagram of the structure of the first digital circuit in a preferred embodiment of the present invention;

[0066] Figure 4 This is a schematic diagram of the structure of the second digital circuit in a preferred embodiment of the present invention;

[0067] Figure 5 This is a schematic diagram of the operating environment of a preferred embodiment of the terminal of the present invention. Detailed Implementation

[0068] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0069] Example 1

[0070] Please see Figure 1 and Figure 2 Embodiment 1 of this application is a feature processing method based on speech recognition, which includes the following steps:

[0071] S1. Acquire voice monitoring data and extract information features from the voice monitoring data.

[0072] In traditional testing, depression can be identified using methods such as electroencephalography (EEG), visual expression analysis, and speech analysis. EEG signals objectively reflect the internal state of the brain and are considered to have broad prospects for promoting effective and efficient clinical diagnosis and identification of depression, as well as further improving clinical treatment outcomes. However, EEG requires subjects to wear related collection and detection devices in real time during the testing process, which reduces the convenience of monitoring and may cause inconvenience to patients with depression. This inconvenience may also affect the patient's mood and thus worsen their condition. Furthermore, due to the multi-channel and dense nature of EEG signals, optimizing the spatial information of EEG is challenging, posing a significant challenge to the development of machine learning techniques for depression identification.

[0073] Similarly, visual expression requires a high level of data collection and constant monitoring of subjects under a camera. This greatly limits the monitoring environment and can easily expose the subjects' personal privacy, leading to a worsening of the patient's condition. Therefore, neither of these methods is suitable for developing auxiliary monitoring methods and devices for depression.

[0074] Therefore, the present invention selects the patient's voice monitoring data as input. Voice can reflect a person's emotional state and emotional changes, and can be used to realize the real-time detection of the patient's mental illness.

[0075] After acquiring the patient's voice monitoring data, this embodiment extracts features from the original audio signal using Mel frequency cepstral coefficients (MFCC). Mel frequency cepstral coefficients are a linear transformation of the logarithmic energy spectrum based on a nonlinear Mel scale of sound frequency. By utilizing the nonlinear perception characteristics of sound frequency by the human ear, linear frequencies are converted into Mel frequencies, and then cepstral analysis is performed in the Mel frequency domain to finally obtain Mel frequency cepstral coefficients. This nonlinear representation method can well simulate the human auditory system and therefore has wide applications in the field of sound processing.

[0076] In this embodiment, the voice monitoring data is first preprocessed by pre-emphasis, overlapping framing, and windowing to highlight the high-frequency components in the voice monitoring data and reduce the non-stationarity of the voice monitoring data.

[0077] Next, a Fast Fourier Transform (FFT) is performed on the speech monitoring data to convert it from a time-domain signal to a frequency-domain signal, obtaining the speech audio domain signal. Then, Mel filtering is performed on the speech audio domain signal. Mel filtering uses a series of triangular filters to simulate the auditory characteristics of the human ear. Each filter outputs a filtered signal. The logarithm of all filtered signals is taken to simulate the nonlinear perception of sound intensity by the human ear, obtaining the logarithmic result. Discrete cosine transform is performed on the logarithmic result to extract the cepstral coefficients. The cepstral coefficients describe the envelope of the spectrum and are used as features of the speech monitoring data.

[0078] The features extracted from the speech monitoring data contain most of the effective information for speech recognition analysis, such as the speaker's voice characteristics, speech rate, and emotion. These features can be used as an auxiliary analysis for depression.

[0079] S2. Input the information features into the temporal feature extraction network, and the temporal feature extraction network outputs temporal features.

[0080] In this embodiment, information features are analyzed through a neural network, which includes a temporal feature extraction network and a classification network. The temporal feature extraction network includes a Long Short-Term Memory (LSTM) network and a Temporal Convolutional Network (TCN). The LSTM network is used to extract the temporal relationship features of the information features, and the TCN network is used to refine the temporal relationship features to obtain temporally refined features. The temporal relationship features and the temporally refined features are used as the temporal features.

[0081] Please refer to Figure 3 Long Short-Term Memory (LSTM) networks are used to handle the variability of the length of information feature inputs and to coarsely extract time relationship features from signals. Specifically, in this embodiment, the LSM network is implemented through a first digital circuit structure, which includes a control unit, a storage unit, and a computing unit. The control unit is used to control the operation of the storage unit and the computing unit, the storage unit is used to temporarily store data, and the computing unit is used for calculation.

[0082] After the information features are input into the Long Short-Term Memory (LSTM) network, they are first temporarily stored in the buffer of the storage unit. During t computations, the information features X input into the LTM network are... t And the intermediate quantity H from the previous calculation t-1 and C t-1 The weights and biases are input into the computation unit, and the computation unit processes the information feature X. t and intermediate quantity H t-1 Calculations are performed on information feature X. t and intermediate quantity H t-1 Merge into a single vector [H] t-1 ,X t ], vector [H t-1 ,Xt ] and the corresponding weight W o W g W i and W f After multiplication, the product and the corresponding bias b o b g b i and b f Adding them together yields four processed vectors W. o ·[H t-1 ,X t ]+b o W g ·[H t-1 ,X t ]+b g W i ·[H t-1 ,X t ]+b i and W f ·[H t-1 ,X t ]+b f The four processed vectors are then processed by four activation functions to obtain four intermediate parameters.

[0083] Specifically, considering that the traditional activation functions Sigmoid and Tanh are both nonlinear operators, and that nonlinear operators are difficult to support in hardware, this embodiment simplifies the LSTM gating mechanism and state update calculations by using HardTanh and HardSigmoid functions instead of Sigmoid and Tanh functions as activation functions. The formulas for HardTanh and HardSigmoid functions are as follows:

[0084]

[0085] Where x represents the variable input to the activation function.

[0086] Based on the activation functions described above, calculate the intermediate parameters of the four long short-term memory networks:

[0087] g t =HardTanh(W g ·[H t-1 X t ]+b g );

[0088] o t =HardSigmoid(W o ·[H t-1 ,X t ]+b o );

[0089] f t =HardTanh(W f ·[H t-1 ,X t ]+b f );

[0090] i t =HardSigmoid(W i ·[H t-1 ,X t ]+b i );

[0091] Then, based on the results of the four intermediate parameters, the new intermediate quantity H is calculated. t and C t The specific calculation formula is as follows:

[0092] C t =f t ·C t-1 +g t ·i t ;

[0093] H t =o t HardTanh(C t );

[0094] The new intermediate quantity H will be output. t and C t Store the new intermediate quantity H in a memory location for the next calculation. t It is also a temporal relationship characteristic of the output of the Long Short-Term Memory network.

[0095] Next, the output of the Long Short-Term Memory (LSTM) network is input into the temporal convolutional network. It's worth noting that this invention employs a dual-clock mechanism. Because the computational requirements of the LSM network are significantly higher than other modules, it becomes the bottleneck of the entire system, causing the temporal convolutional network to be idle while waiting for the LSM network's result. Therefore, a dual-clock mechanism is introduced and resource allocation is optimized. The LSM network operates using a high-speed clock, while the temporal convolutional network operates using a slower clock. Thanks to the dual-clock mechanism, the temporal convolutional network module using the slower clock can effectively reduce power consumption by lowering the voltage conversion rate. This, through reasonable resource allocation and adaptive switching circuitry, reduces circuit power consumption and improves circuit parallelism.

[0096] However, due to the different system clock frequencies of the Long Short-Term Memory (LSTM) network and the temporal convolutional network, a handshake signal is needed to securely and stably transmit data to the temporal convolutional network. This is achieved through a Clock Domain Crossing (CDC) module, which coordinates the LSM and temporal convolutional networks to ensure the secure and stable transmission of data. The temporal convolutional network uses one-dimensional convolutional layers, applying unidirectional convolution to obtain temporally refined features based on temporal relationships.

[0097] S3. Input the time-series features into the auxiliary information acquisition network, and the auxiliary information acquisition network outputs auxiliary information to assist in the judgment of depression.

[0098] Specifically, in this embodiment, the auxiliary information acquisition network includes a cross-attention module and a classification network.

[0099] First, a cross-attention module is used to identify key features in the temporal features. In this embodiment, the cross-attention module uses the temporal relationship features extracted by the Long Short-Term Memory network. As the query (Q), the temporally refined features extracted by the temporal convolutional network. Key features are identified using a cross-attention approach, with keys (K) and values ​​(V) as both.

[0100] Specifically, let's start with the characteristics of time relationships. As the query (Q), the temporally refined features extracted by the temporal convolutional network. As the key (K), the similarity matrix is ​​obtained. The attention matrix is ​​then input into the softmax function to calculate the attention weight of each head. Since exponential calculation is required, the computational cost is very high. Therefore, unlike the previous method of approximating the softmax function, this embodiment adopts an optimized calculation method to speed up the calculation of attention weights in the cross-attention module while maintaining the minimum calculation error. Based on the characteristic that each attention head in the cross-attention module only processes two features, an approximate calculation method is specially designed to directly provide the softmax function output, thereby reducing division operations and improving accuracy.

[0101] First, the query (Q), key (K), and value (V) need to be mapped through a linear layer. Then, the data is divided into sub-attention points based on the number of attention points, resulting in sub-attention time refinement features and sub-attention time relationship features. Next, a similarity matrix is ​​calculated based on these sub-attention time refinement features and sub-attention time relationship features, and then normalized using the softmax function.

[0102]

[0103] Where x1 is the first feature of the input softmax function, and x2 is the second feature of the input softmax function.

[0104] After normalization, a weight matrix is ​​obtained. The product of the weight matrix and the time-refinement feature as the value (V) is calculated, and the product is concatenated as the key feature.

[0105] Finally, the key features are input into the classification network to provide auxiliary information. The classification network includes an average pooling layer and a fully connected layer (Classification).

[0106] In this embodiment, the cross-attention module and the temporal convolutional network belong to the same digital circuit structure, namely the second digital circuit structure. The second digital circuit structure also includes a control unit, a storage unit, and a computation unit. The computation unit of the second digital circuit structure consists of a multiplication matrix, an adder, and an activation function. In the second digital circuit structure, the multiplication operation of the computation unit mainly relies on the reuse of the multiplication matrix, and the addition operation relies on the reuse of the adder. Even though this will take more time, this time is still much less than the computation time of the long short-term memory network. Therefore, it will not cause a significant slowdown in output in practical applications.

[0107] Example 2

[0108] Please see Figure 5 Based on the above method, the present invention also provides a terminal, the terminal including a processor 10, a memory 20 and a display 30. Figure 5 Only some of the terminal components are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0109] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc. Further, the memory 20 may include both internal and external storage devices. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a speech recognition-based feature processing program 40, which can be executed by the processor 10 to realize the terminal of this application.

[0110] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing related programs of the feature processing method based on speech recognition.

[0111] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display 30 is used to display information on the terminal and to display a visual user interface. The components 10-30 of the terminal communicate with each other via a system bus.

[0112] In one embodiment, when the processor 10 executes a speech recognition-based feature processing program 40 in the memory 20, it implements the steps of a speech recognition-based feature processing method as described above.

[0113] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory. In other embodiments, the memory 20 may be an external storage device of the terminal, such as a plug-in hard disk, SmartMediaCard (SMC), SecureDigital (SD) card, or FlashCard. Furthermore, the memory 20 may include both internal and external storage units. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code installed on the terminal. The memory 20 can also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 20 stores a speech recognition-based feature processing program, which can be executed by the processor 10 to implement a speech recognition-based feature processing method as described in this application.

[0114] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in the memory 20 or process data, such as executing a feature processing method based on speech recognition.

[0115] Example 3

[0116] This embodiment provides a storage medium that stores a speech recognition-based feature processing program. When the speech recognition-based feature processing program is executed by a processor, it implements the steps of the speech recognition-based feature processing method described above.

[0117] In summary, this invention continuously monitors patients' voices to obtain their voice monitoring data, extracts information features from the voice monitoring data, and then performs time-series analysis on the information features of the voice monitoring data acquired over a long period to obtain time-series features. Based on these time-series features, it can help determine whether a patient has depression, thereby reducing the burden on medical personnel and enabling more patients with depression to receive early diagnosis and timely treatment.

[0118] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal that includes that element.

[0119] Of course, those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware (such as a processor, controller, etc.). The program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The storage medium can be a memory, magnetic disk, optical disk, etc.

[0120] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A feature processing method based on speech recognition, characterized in that, The feature processing method based on speech recognition includes: Acquire voice monitoring data and extract information features from the voice monitoring data; The information features are input into a temporal feature extraction network, and the temporal feature extraction network outputs temporal features. The time-series features are input into the auxiliary information acquisition network, and the auxiliary information acquisition network outputs auxiliary information to assist in the diagnosis of depression. The temporal feature extraction network includes a long short-term memory network and a temporal convolutional network; The step of inputting the information features into a temporal feature extraction network, and the temporal feature extraction network outputting temporal features, specifically includes: The information features are input into the long short-term memory network to extract temporal relationship features; The temporal relationship features are input into the temporal convolutional network to obtain the temporal refinement features; The temporal relationship features and the temporal refinement features are used as the temporal sequence features; The auxiliary information acquisition network includes a cross-attention module and a classification network. The temporal features are input into the auxiliary information acquisition network, and the network outputs auxiliary information for assisting in the diagnosis of depression, specifically including: The temporal features are input into the cross-attention module, and the cross-attention module outputs key features; The key features are input into the classification network, and the classification network outputs the auxiliary information. The step of inputting the temporal features into the cross-attention module and outputting key features by the cross-attention module specifically includes: The time refinement feature and the time relationship feature are separated into two parts to obtain separate time refinement features and separate time relationship features. A similarity matrix is ​​calculated based on the refined features of the split-time period and the features of the relationship between the split-time periods. The similarity matrix is ​​normalized using the softmax function to obtain the weight matrix; The key features are calculated based on the weight matrix and the time-separated refinement features.

2. The feature processing method based on speech recognition according to claim 1, characterized in that, The extraction of information features from the voice monitoring data specifically includes: Perform a Fast Fourier Transform on the speech monitoring data to obtain the speech audio domain signal; Mel filtering is applied to the speech audio domain signal to obtain multiple filtered signals; Logarithmic processing is performed on each of the filtered signals to obtain multiple logarithmic results; The discrete cosine transform is performed on each of the logarithmic results to obtain cepstral coefficients, which are then used as the information features.

3. The method for processing depression features based on speech recognition according to claim 1, characterized in that, The step of inputting the information features into the long short-term memory network to extract temporal relationship features specifically includes: In each calculation, the information characteristics of this calculation are... Intermediate value from the previous calculation Combined into a computational vector ; Calculate the vector Multiply each of these vectors by its corresponding weight, and then add the product to the corresponding bias to obtain four intermediate vectors: ； ； ； ； in, , , and All are weights. , , and All are biased; The four intermediate vectors are input into the activation function to obtain four intermediate parameters. , , and ; Based on intermediate parameters , , and And the intermediate values ​​from the previous calculation. The intermediate values ​​obtained in this calculation are: ； ； in, This refers to the HardTanh function; The intermediate values ​​in this calculation and Save and include the intermediate values ​​from this calculation. As a characteristic of time relationship Output The HardTanh function is calculated as follows: ； in, This represents the input variable.

4. The feature processing method based on speech recognition according to claim 3, characterized in that, The four intermediate vectors are input into the activation function to obtain four intermediate parameters. , , and Specifically, it includes: intermediate vector and intermediate vector Input the HardTanh function to obtain intermediate parameters and : ； ； intermediate vector and intermediate vector Input the HardSigmoid function to obtain intermediate parameters and : ； ； in, This represents the HardSigmoid function, and the formula for calculating the HardSigmoid function is as follows: 。 5. The feature processing method based on speech recognition according to claim 1, characterized in that, The classification network includes an average pooling layer and a fully connected layer; The step of inputting the key features into the classification network and the classification network outputting the auxiliary information specifically includes: The key features are input into the average pooling layer, and the average pooling layer outputs the pooling result. The pooling result is input into the fully connected layer, and the fully connected layer outputs the auxiliary information.

6. A terminal, characterized in that, The terminal includes: a memory, a processor, and a speech recognition-based feature processing program stored in the memory and executable on the processor. When the speech recognition-based feature processing program is executed by the processor, it controls the terminal to implement the steps of the speech recognition-based feature processing method as described in any one of claims 1-5.

7. A readable storage medium, characterized in that, The readable storage medium stores a speech recognition-based feature processing program, which, when executed by a processor, implements the steps of a speech recognition-based feature processing method as described in any one of claims 1-5.

Images