Speech emotion recognition method and system based on artificial intelligence

Abstract

The present invention relates to the field of speech recognition technology, and specifically to a speech emotion recognition method and system based on artificial intelligence. The present invention analyzes the change of the speech signal of the target object to obtain the characteristic value of the fluctuation of the target object's speech signal that may have emotional changes; then analyzes the characteristic value of the fluctuation to obtain various characteristic curves; by numerically analyzing the characteristic curve corresponding to the known emotional state, obtains the weight of the influence of each emotional type characteristic curve on each emotional state; finally, analyzes the characteristic curve corresponding to the target object's speech signal, combines the weight, obtains the confidence of the target object showing each emotional state, and completes the recognition of the target object's emotion. The method and system provided by the present invention can realize all-round language monitoring of the target object in the process of psychological counseling, identify the emotions expressed in the target object's speech signal, and thus improve the efficiency and accuracy of psychological counseling judgment.

Description

Technical Field

[0001] The present invention relates to the field of speech recognition technology, and in particular to a speech emotion recognition method and system based on artificial intelligence. Background Art

[0002] In the process of psychological consultation, a series of communication and exchanges are often conducted with the target object, i.e. the consulting object. The psychologist will observe the target object's emotions during the communication and exchanges with the target object to analyze and judge the target object's answers. Traditional methods generally identify the emotions that the target object may express based on the communication and exchanges between the psychologist and the target object. This method is prone to miss some minor emotional changes of the target object, resulting in the loss of some important information.

[0003] Speech Emotion Recognition (SER) refers to the technology of identifying the emotional state of a speaker through the acoustic features of a piece of speech. As one of the most natural ways of communication for humans, the emotional information contained in speech is of great significance for improving the human-computer interaction experience. Any event that affects the speaker's thinking or inner consciousness will also find corresponding signs in the sound. Through the sound of a person's speech, we can not only obtain the emotional state such as joy, anger, sorrow, and happiness in the words, but also perceive the speaker's psychological fluctuations, thinking consciousness, physical and mental stress, and health level. Through speech emotion recognition, smart devices can understand the user's emotional state more accurately. Therefore, speech emotion recognition technology can be applied to the psychological counseling process to achieve all-round language monitoring of the target object during the psychological counseling process and identify the emotions expressed in the target object's speech signal.

[0004] In the process of psychological counseling, the psychological activities of the target object are characterized by frequent, subtle and complex changes. The existing speech emotion recognition model has low accuracy and poor interpretability of emotion recognition results due to the limited size and diversity of its database, which cannot effectively improve the efficiency and accuracy of psychological counseling diagnosis. Summary of the invention

[0005] In order to solve the above technical problems, the purpose of the present invention is to provide a speech emotion recognition method and system based on artificial intelligence.

[0006] According to the first aspect of the embodiment of the present application, a method for speech emotion recognition based on artificial intelligence is provided, and the technical solution adopted is as follows:

[0007] Collecting an analog voice signal of a target object and converting the analog voice signal into a digital voice signal;

[0008] Dividing the digital voice signal into a plurality of fluctuations, analyzing the change characteristics between the fluctuations, and obtaining characteristic values ​​of the fluctuations that may have emotional changes;

[0009] Performing RNN circular convolution on the eigenvalue to obtain a characteristic curve corresponding to the fluctuation;

[0010] Based on the corresponding characteristic curves under known emotional states, analyze the weight of the influence of the characteristic curves of each emotional type on each emotional state;

[0011] Analyze the numerical characteristics of the characteristic curve of the target object, combine the weight of the characteristic curve of each emotion type on each emotional state, and obtain the confidence of the target object in showing each emotional state;

[0012] Based on the confidence level, identifying the emotional state of the target object;

[0013] Analyzing the change characteristics between the fluctuations to obtain the characteristic values ​​of the fluctuations that may have emotional changes, including:

[0014] Analyze the temporal continuity between adjacent fluctuations, analyze the amplitude changes between adjacent fluctuations, and analyze the number of small amplitudes of the fluctuations, and mark the fluctuations that may have emotional changes;

[0015] The amplitude of the marked fluctuation and the number of small amplitudes are obtained as characteristic values ​​of the fluctuation that may have emotional changes.

[0016] In some embodiments of the present invention, collecting an analog voice signal of a target object and converting the analog voice signal into a digital voice signal includes:

[0017] Collecting analog voice signals of the target object;

[0018] Sampling and discretizing the analog voice signal to obtain a discrete voice signal;

[0019] Amplitude quantization is performed on the discrete speech signal to obtain a digital speech signal.

[0020] In some embodiments of the present invention, the digital voice signal is divided into a plurality of fluctuations, including:

[0021] Decoding the digital voice signal and dequantizing it to obtain a voice waveform;

[0022] By utilizing the gradient direction of the points on the edge of the waveform in the speech waveform diagram, the point where the gradient direction changes from right to left is set as the segmentation point, and the digital speech signal is divided into a plurality of fluctuations.

[0023] In some embodiments of the present invention, performing RNN circular convolution on the eigenvalue to obtain a characteristic curve corresponding to the fluctuation includes:

[0024] Organizing the eigenvalues ​​into a sequence form and performing standardization processing to obtain a standardized eigenvalue sequence;

[0025] The standardized feature value sequence is input into the trained RNN model, and the characteristic curve corresponding to the fluctuation is output.

[0026] In some embodiments of the present invention, based on the corresponding characteristic curve under the known emotional state, analyzing the weight of the influence of the characteristic curve of each emotional type on each emotional state includes:

[0027] Obtain characteristic curves of each emotion type corresponding to a plurality of simulated speech signals under different emotional states;

[0028] The numerical size characteristics of the characteristic curves of each emotion type corresponding to multiple simulated voice signals under various emotion states are analyzed to obtain the weight of the influence of each emotion type characteristic curve on each emotion state.

[0029] In some embodiments of the present invention, the numerical size characteristics of the characteristic curves of each emotion type corresponding to multiple simulated speech signals under various emotional states are analyzed to obtain the weight of the influence of each emotion type characteristic curve on each emotional state, including:

[0030] Set the threshold of each emotion type characteristic curve;

[0031] Analyze the relationship between the maximum value on each emotion type characteristic curve corresponding to the plurality of simulated voice signals under various emotional states and the threshold value to obtain a first value influencing parameter;

[0032] Analyze the magnitude relationship between all values ​​on each emotion type characteristic curve corresponding to multiple simulated voice signals under various emotional states and the threshold value to obtain a first quantity influence parameter;

[0033] According to the first numerical influence parameter and the first quantity influence parameter, obtaining numerical size characteristics of each emotion type characteristic curve corresponding to a plurality of simulated speech signals under various emotional states;

[0034] Traversing the numerical size features of all emotion type characteristic curves corresponding to multiple simulated speech signals under each emotion state, and summing up the numerical size features to obtain a total numerical size feature;

[0035] The proportional relationship between the numerical value feature and the total numerical value feature is calculated to obtain the weight of the influence of each emotion type characteristic curve on each emotion state.

[0036] In some embodiments of the present invention, the numerical characteristics of the characteristic curve of the target object are analyzed, and the weight of the influence of the characteristic curve of each emotion type on each emotional state is combined to obtain the confidence that the target object exhibits each emotional state, including:

[0037] Analyze the relationship between the maximum value on the characteristic curve of the target object and the threshold value to obtain a second value influencing parameter;

[0038] Analyze the magnitude relationship between all values ​​on the characteristic curve of the target object and the threshold value to obtain a second quantity influence parameter;

[0039] According to the second numerical influence parameter and the second quantity influence parameter, combined with the weight of the influence of the characteristic curve of each emotion type on each emotional state, the confidence level that the target object exhibits each emotional state is obtained.

[0040] According to a second aspect of an embodiment of the present application, a speech emotion recognition system based on artificial intelligence is provided, comprising: a memory and a processor, wherein:

[0041] The memory is used to store program codes;

[0042] The processor is used to read the program code stored in the memory and execute the method described in the first aspect of the embodiment of the present application.

[0043] In some embodiments of the present invention, the processor comprises:

[0044] A voice signal acquisition module, used to acquire an analog voice signal of a target object and convert the analog voice signal into a digital voice signal;

[0045] A feature value acquisition module, used to divide the digital voice signal into a plurality of fluctuations, analyze the change characteristics between the fluctuations, and obtain the feature values ​​of the fluctuations that may have emotional changes;

[0046] A characteristic curve acquisition module, used for performing RNN circular convolution on the characteristic value to obtain a characteristic curve corresponding to the fluctuation;

[0047] A weight analysis module, for analyzing the weight of the influence of the characteristic curve of each emotion type on each emotion state based on the corresponding characteristic curve under the known emotion state;

[0048] The confidence analysis module is used to analyze the numerical characteristics of the characteristic curve of the target object, and combine the weight of the characteristic curve of each emotion type on the influence of each emotional state to obtain the confidence of the target object showing each emotional state;

[0049] An emotional state recognition module, used to complete the recognition of the emotional state of the target object based on the confidence level;

[0050] Analyzing the change characteristics between the fluctuations to obtain the characteristic values ​​of the fluctuations that may have emotional changes, including:

[0051] Analyze the temporal continuity between adjacent fluctuations, analyze the amplitude changes between adjacent fluctuations, and analyze the number of small amplitudes of the fluctuations, and mark the fluctuations that may have emotional changes;

[0052] The amplitude of the marked fluctuation and the number of small amplitudes are obtained as characteristic values ​​of the fluctuation that may have emotional changes.

[0053] Compared with the prior art, the speech emotion recognition method and system based on artificial intelligence provided by the present invention have the following beneficial effects:

[0054] The present invention analyzes the change of the speech signal of the target object during the psychological consultation process, obtains the characteristic value of the fluctuation of the target object speech signal that may have emotional changes; then uses the characteristic value of the fluctuation of the target object speech signal to analyze, performs RNN circular convolution on the characteristic value, obtains the characteristic curve, and the characteristic curve exceeding the threshold is considered abnormal; by numerically analyzing the characteristic curve corresponding to the known emotional state, obtains the weight of the influence of each emotional type characteristic curve on each emotional state; finally, analyzes the numerical characteristics of the characteristic curve corresponding to the target object speech signal, combines the weight, obtains the confidence of the target object showing each emotional state, and completes the recognition of the target object emotional state. The method and system provided by the present invention can realize the comprehensive language monitoring of the target object during the psychological consultation process, identify the emotions shown in the target object speech signal, thereby helping the psychologist to efficiently capture the psychological sensitivity points and abnormal points shown by the target object, quickly understand the essential psychological problems of the target object, help the psychologist to complete the psychological diagnosis of the target object, and improve the efficiency of psychological consultation diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0056] Figure 1 A basic flow chart of a method for speech emotion recognition based on artificial intelligence provided by one embodiment of the present invention;

[0057] Figure 2 A schematic diagram of segmentation points and fluctuations of a speech waveform provided by an embodiment of the present invention;

[0058] Figure 3 A schematic diagram of the basic components of a speech emotion recognition system based on artificial intelligence provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0059] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of the speech emotion recognition method and system based on artificial intelligence proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0060] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. Terms such as "comprises", "comprising" or any other variants thereof are intended to cover non-exclusive inclusion, so that a circuit structure, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such article or device. In the absence of further restrictions, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the article or device including the element.

[0061] In the process of psychological consultation, a series of communication and exchanges are often conducted with the target object, i.e. the consulting object. The psychologist will observe the target object's emotions during the communication and exchanges with the target object to analyze and judge the target object's answers. However, since the human eye often has some deviations, the analysis effect is not good. Therefore, the present invention is mainly based on artificial intelligence to perform speech emotion recognition. By analyzing the characteristic expressions in the speech signal, the emotions that the target object may express in the speech are identified, thereby helping the psychologist to perceive the psychological problems of the target object and ensure the correctness of the psychologist's analysis of the problem.

[0062] The following is a detailed description of a specific solution of a speech emotion recognition method based on artificial intelligence provided by the present invention in conjunction with the accompanying drawings.

[0063] See also Figure 1 , which shows the basic process of a speech emotion recognition method based on artificial intelligence provided by an embodiment of the present invention.

[0064] like Figure 1 As shown, an artificial intelligence-based voice emotion recognition method provided by an embodiment of the present invention specifically includes:

[0065] S100: Collecting analog voice signals of a target object and converting the analog voice signals into digital voice signals.

[0066] The present invention is based on the analysis of the target object's voice signal, and the language signal emitted by the target object is a continuous analog voice signal in time series. In the process of voiceprint analysis, general processing equipment such as computers, general electronic equipment and other equipment are based on digital signals for processing, so it is necessary to convert the target object's analog voice signal into a digital signal to facilitate the subsequent analysis of the characteristic performance in the voice signal.

[0067] Therefore, this embodiment first collects the analog voice signal of the target object and converts the analog voice signal into a digital voice signal. Further, it includes collecting the analog voice signal of the target object; sampling and discretizing the analog voice signal to obtain a discrete voice signal; and performing amplitude quantization processing on the discrete voice signal to obtain a digital voice signal. The specific operation method is as follows:

[0068] First, install the dongle (built-in software) inside the voice processing device (such as a laptop), connect one end of the mind-body monitor to the voice processing device, and use the other end normally for the target object. Use the microphone of the all-in-one machine to collect the voice signal of the target object to obtain an analog voice signal. Then sample and discretize the analog voice signal, where the sampling frequency is 16kHz, to obtain the discrete voice signal of the target object. Then perform amplitude quantization on the discrete voice signal, map the continuous amplitude value of the analog voice signal to a discrete digital value, using a 16-bit quantization bit to obtain a 16-bit digital voice signal. Finally, the quantized 16-bit digital voice signal is converted into a binary representation of the value of each sampling point using pulse code modulation to obtain the binary form of the digital voice signal. Subsequent analysis is based on the binary form of the digital voice signal.

[0069] S200: Divide the digital voice signal into a plurality of fluctuations, analyze the change characteristics between the fluctuations, and obtain characteristic values ​​of the fluctuations that may have emotional changes.

[0070] In the process of speech emotion recognition of the target object, the collected digital speech signal is mainly analyzed. When the target object changes his emotion, his speech signal will change to a certain extent, such as slight tremor, sudden pause, etc. These changes are often the key to the subsequent recognition process.

[0071] The characteristics of the waveform in the digital voice signal can reflect some specific information to a certain extent. For example, when the digital voice signal of a certain sound is suddenly interrupted, it means that the target object may have a certain degree of psychological fluctuation. The words asked by the psychologist may hit some key points and need to be focused on and analyzed. When the target object starts to speak, the voice processing device will collect the voice signal, and the built-in software of the voice processing device will analyze the obtained digital voice signal every 10 seconds. Analyze the micro-tremor, spikes, mutations, inhibition, pauses and modulation in the digital voice signal to help the psychologist make judgments.

[0072] Therefore, this embodiment divides the digital voice signal into a number of fluctuations, analyzes the change characteristics between the fluctuations, and obtains the characteristic values ​​of the fluctuations that may have emotional changes.

[0073] Further analysis shows that when the built-in software of the voice processing device analyzes the digital voice signal, it mainly analyzes the characteristic performance of the digital voice signal, including tremors, mutations, pauses and other phenomena, which can reflect the emotional performance of the target object to a certain extent. In the analysis process, the analysis of the overall waveform in 10s may result in the analysis object being too large, while the actual required results are all small changes, resulting in poor analysis results. Therefore, it is necessary to divide the digital voice signal into several fluctuations. Specifically, the collected digital voice signal is first decoded and dequantized to obtain a voice waveform graph, where the horizontal axis of the voice waveform graph is time and the vertical axis is amplitude. In the process of psychological counseling, normal counselors will not speak loudly, so there may be a problem of insufficient high-frequency data. In this embodiment, when the high-frequency data in the voice waveform graph is insufficient, pre-emphasis is used to process it; wherein, the waveform of the high-frequency part is usually sharper and changes faster, while the waveform of the low-frequency part is relatively flat and changes slower; by observing the frequency distribution on the vertical axis, the high-frequency part can be preliminarily judged. The vocal cords will produce a complete band from opening to closing, such as Figure 2 As shown, the gradient direction of the points on the edge of the waveform in the speech waveform is used, combined with the waveform performance, to set the amplitude close to 0, and at the same time the gradient direction changes from right to left (such as Figure 2 The point where the waveform changes from falling to rising is the split point, dividing the digital voice signal into several fluctuations.

[0074] Then, the characteristic performance is analyzed based on the specific information of each fluctuation. When there are tremors, spikes, mutations, inhibitions, pauses, and modulations in the fluctuation, it may correspond to a slight change in the target object's emotions, which requires more in-depth analysis. The details are as follows:

[0075] Under normal circumstances, the target object's speech signal will be relatively stable and continuous during the speech process. For adjacent fluctuations, the end time of the former and the start time of the latter will be connected; if the speech signal is interrupted, it means that the target object's emotions may fluctuate. Therefore, the time continuity between adjacent fluctuations is analyzed. Specifically, the time difference between adjacent fluctuations is calculated:

[0076]

[0077] In the formula, Indicates The fluctuation and The time difference between the waves; Indicates The start time of a wave; Indicates The end time of a wave.

[0078] right The corresponding The fluctuation and Mark the fluctuations.

[0079] Then analyze the amplitude of each fluctuation and the amplitude change between adjacent fluctuations. Since there are multiple small amplitudes on each fluctuation, a small amplitude threshold T can be set. The value of T can be 0.1, that is, if the amplitude difference between the lowest point and the highest point of a wave is less than T (the difference between the trough and the peak is less than T), then it is a small amplitude. The average value of these small amplitudes is used as the amplitude of this fluctuation:

[0080]

[0081] In the formula, Indicates The amplitude of a fluctuation; Indicates The fluctuation A tiny amplitude, which is the amplitude at the peak; Indicates The number of small amplitudes in a fluctuation.

[0082] Calculate the change in amplitude between adjacent fluctuations:

[0083]

[0084] In the formula, Indicates The amplitude of the fluctuation is The difference in the amplitude of the fluctuations; Indicates The amplitude of a fluctuation; Indicates The amplitude of a fluctuation.

[0085] Constructing a data series with varying amplitudes ,because It can be positive or negative, and the fluctuations corresponding to the data in the data sequence that have different signs from the data on both sides are marked. That is, when data of the type "positive-negative-positive\negative-positive-negative" appears in the data sequence, the fluctuations corresponding to the data in the middle are marked.

[0086] And the number of small amplitudes of fluctuations analyzed, that is, the number of small amplitudes of unlabeled fluctuations analyzed , set the quantity threshold , when the number of small amplitudes is greater than the quantity threshold, that is , it means that the target object produced more micro-tremors when speaking here, and we need to pay attention to it and mark this fluctuation.

[0087] Through the above steps, the marking of fluctuations that may have emotional changes is completed, and it is necessary to judge and analyze according to their characteristic value performance. Therefore, the amplitude of the marked fluctuation and the number of small amplitudes are obtained as the characteristic values ​​of fluctuations that may have emotional changes. The characteristic value includes multiple amplitude data and a number of small amplitude data. Among them, the amplitude represents the degree of spikes or suppression in the target object's voice, and the number of small amplitudes represents the degree of tremor in the target object's voice.

[0088] At this point, the characteristic values ​​of the target object's possible emotional fluctuations are obtained.

[0089] S300: Perform RNN circular convolution on the eigenvalues ​​to obtain a characteristic curve corresponding to the fluctuation.

[0090] In the above process, the characteristic values ​​of the fluctuations of the target object's possible emotional changes are obtained. When the target object answers certain key questions during the communication process, there may be slight tremors, spikes or suppression in the voice signal, and these phenomena can be reflected in the fluctuating characteristic values. Therefore, it is necessary to analyze the fluctuating characteristic values.

[0091] First, perform RNN circular convolution on the eigenvalues ​​to obtain the characteristic curve corresponding to the fluctuation, and further include: organizing the eigenvalues ​​into a sequence form, each element in the sequence can be a two-tuple (amplitude, number of small amplitudes), and standardizing the eigenvalue sequence to obtain a standardized eigenvalue sequence; input the standardized eigenvalue sequence into the trained RNN model, and output the characteristic curve corresponding to the fluctuation. That is, the amplitude of the fluctuation and the number of small amplitudes are used as the input features of the RNN model. The amplitude of the fluctuation directly determines the height or strength of the characteristic curve at certain time points, and the number of small amplitudes on the characteristic curve is manifested as the smoothness or stability of the curve.

[0092] S400: Analyze the weight of the influence of the characteristic curve of each emotion type on each emotion state based on the corresponding characteristic curve under the known emotion state.

[0093] Since the target object behaves differently under different emotional states, it is necessary to determine the weight of the characteristic curve under a certain emotional state for the emotional state, wherein the emotional states include joy, anger, sorrow, happiness, fear, shock, thinking, etc. Therefore, this embodiment is based on the characteristic curve corresponding to the known emotional state, and the weight of the characteristic curve of each emotional type on each emotional state, wherein the emotional types include joy, anger, sorrow, happiness, fear, shock, thinking, etc. Further comprising: obtaining the characteristic curve of each emotional type corresponding to a plurality of analog voice signals under different emotional states, specifically, the characteristic curve of each emotional type corresponding to 5 analog voice signals under various emotional states can be obtained respectively; analyzing the numerical size characteristics of the characteristic curve of each emotional type corresponding to the plurality of analog voice signals under various emotional states, and obtaining the weight of the characteristic curve of each emotional type on each emotional state, specifically, for the characteristic curves of different emotional types, setting the threshold of the characteristic curve of each emotional type , when the characteristic curve exceeds the threshold , it indicates that the characteristic curve corresponding to the emotion type can express the emotion; analyze the relationship between the maximum value and the threshold value on each emotion type characteristic curve corresponding to multiple analog voice signals under various emotional states, and obtain the first numerical influence parameter; analyze the relationship between all the values ​​and the threshold value on each emotion type characteristic curve corresponding to multiple analog voice signals under various emotional states, and obtain the first quantity influence parameter; according to the first numerical influence parameter and the first quantity influence parameter, obtain the numerical size characteristics of each emotion type characteristic curve corresponding to multiple analog voice signals under various emotional states; traverse the numerical size characteristics of all emotion type characteristic curves corresponding to multiple analog voice signals under each emotional state and sum them up to obtain the total numerical size characteristics; calculate the proportional relationship between the numerical size characteristics and the total numerical size characteristics, and obtain the weight of each emotion type characteristic curve on each emotional state. It should be noted that, under the premise of a known emotional state, one emotional state corresponds to one emotional characteristic curve, that is, the characteristic curve obtained by one emotional state is defined as this kind of emotional characteristic curve. The weight of the influence of each emotion type characteristic curve on each emotional state is to measure the weight of the characteristic curve under one emotional state to represent the emotional state (different emotional states are reflected in language to different degrees).

[0094] Based on the above analysis, we construct The characteristic curve of the emotion class The weight calculation formula of the influence of the emotional state is:

[0095]

[0096] In the formula, Indicates The characteristic curve of the emotion class The weight of the influence of emotional states; Indicates The first emotional state The first of the emotional characteristic curves The value size of the bar; Indicates The threshold of the characteristic curve of the emotion class; Indicates The first emotional state The first of the emotional characteristic curves The number of values ​​in the bar that exceed the threshold; Indicates The first emotional state The first of the emotional characteristic curves The number of all values ​​of the bar; Indicates the number of characteristic curve emotion types; Represents the maximum value function.

[0097] The difference between the maximum value in the characteristic curve and the threshold value indicates the first value influencing parameter. The larger the result, the larger the maximum value in the characteristic curve, indicating that the characteristic curve has a greater weight in representing the emotional state. The ratio of the number of values ​​in the characteristic curve that exceed the threshold to the total number of values ​​in the characteristic curve, that is, the first quantity influencing parameter, the larger the result, the more values ​​in the characteristic curve exceed the set threshold, and the greater the weight of the emotional state expressed by the characteristic curve. The above formula is mainly calculated based on the specific information of the characteristic curve obtained under a known emotional state. When the number of values ​​in the characteristic curve that exceed the set threshold is greater and the maximum value is larger, it can be said that the characteristic curve expresses a greater weight of the emotional state. However, since there are multiple characteristic curves that express the emotional state, the weight of each characteristic curve needs to be calculated under all characteristic curves.

[0098] S500: Analyze the numerical characteristics of the characteristic curve of the target object, and combine the weights of the influence of the characteristic curve of each emotion type on each emotional state to obtain the confidence that the target object exhibits each emotional state.

[0099] Analyze the numerical characteristics of the characteristic curve of the target object, combine the weights of the characteristic curves of each emotion type on each emotional state, and obtain the confidence that the target object exhibits each emotional state, including: analyzing the relationship between the maximum value and the threshold on the characteristic curve of the target object to obtain a second numerical influence parameter; analyzing the relationship between all values ​​and the threshold on the characteristic curve of the target object to obtain a second quantitative influence parameter; according to the second numerical influence parameter and the second quantitative influence parameter, combine the weights of the characteristic curves of each emotion type on each emotional state to obtain the confidence that the target object exhibits each emotional state.

[0100] Based on the above analysis, constructing the target object shows the The confidence calculation formula for the emotional state is:

[0101]

[0102] In the formula, Indicates that the target object shows Confidence of emotional states; Indicates The characteristic curve of the emotion class The weight of the impact of the emotional state; The target object The value on the characteristic curve; Indicates The number of values ​​of a characteristic curve exceeding the threshold; Indicates The number of all values ​​of a characteristic curve; Indicates The threshold of the characteristic curve; represents the number of target object characteristic curves obtained; represents the linear normalization function.

[0103] The difference between the maximum value in the characteristic curve and the threshold value indicates the second value influencing parameter. The larger the result is, the larger the maximum value in the characteristic curve is, indicating that the characteristic curve has a higher confidence in expressing the emotional state. The ratio of the number of values ​​in the characteristic curve exceeding the threshold to the total number of values ​​in the characteristic curve, that is, the second quantity influence parameter, the larger the result is, the more values ​​in the characteristic curve exceed the set threshold, and the higher the confidence of the characteristic curve in expressing the emotional state is. The above formula mainly uses the influence weight of the characteristic curve under an emotional state on the emotional state, and calculates it in combination with the specific information of the characteristic curve obtained based on the target object's voice signal.

[0104] At this point, the confidence level of the target object's various emotional states at this time is obtained.

[0105] S600: Based on the confidence level, the emotional state of the target object is identified.

[0106] In the above process, the confidence of the target object showing various emotional states at this time is obtained. The confidence of different emotional states is constructed from high to low data sequence The confidence values ​​corresponding to different emotional states are displayed on the dedicated software. The larger the confidence value, the greater the possibility that the target object is expressing this emotion at this time. The emotional state corresponding to the maximum confidence value is taken as the current emotional state of the target object.

[0107] Based on the same inventive concept as the above method, this embodiment also provides a speech emotion recognition system based on artificial intelligence.

[0108] See also Figure 3 , which shows the basic composition of a speech emotion recognition system based on artificial intelligence provided by an embodiment of the present invention.

[0109] like Figure 3 As shown, a speech emotion recognition system based on artificial intelligence includes: a memory 10 and a processor 20, wherein:

[0110] A memory 10, used for storing program codes;

[0111] The processor 20 is used to read the program code stored in the memory 10, and execute the acquisition of the analog voice signal of the target object, and convert the analog voice signal into a digital voice signal; divide the digital voice signal into a number of fluctuations, analyze the change characteristics between the fluctuations, and obtain the characteristic values ​​of the fluctuations that may have emotional changes; perform RNN circular convolution on the characteristic values ​​to obtain the characteristic curves corresponding to the fluctuations; based on the corresponding characteristic curves under known emotional states, analyze the weights of the characteristic curves of each emotional type on each emotional state; analyze the numerical characteristics of the characteristic curves of the target object, and combine the weights of the characteristic curves of each emotional type on each emotional state to obtain the confidence of the target object in expressing each emotional state; based on the confidence, complete the recognition of the target object's emotional state.

[0112] The processor 20 includes a speech signal acquisition module 21, a feature value acquisition module 22, a feature curve acquisition module 23, a weight analysis module 24, a confidence analysis module 25 and an emotional state recognition module 26.

[0113] The voice signal acquisition module 21 is used to acquire the analog voice signal of the target object and convert the analog voice signal into a digital voice signal;

[0114] The feature value acquisition module 22 is used to divide the digital voice signal into a plurality of fluctuations, analyze the change characteristics between the fluctuations, and obtain the feature values ​​of the fluctuations that may have emotional changes;

[0115] The characteristic curve acquisition module 23 is used to perform RNN circular convolution on the characteristic value to obtain the characteristic curve corresponding to the fluctuation;

[0116] A weight analysis module 24, for analyzing the weight of the influence of the characteristic curve of each emotion type on each emotion state based on the corresponding characteristic curve under the known emotion state;

[0117] The confidence analysis module 25 is used to analyze the numerical characteristics of the characteristic curve of the target object, and combine the weight of the characteristic curve of each emotion type on each emotional state to obtain the confidence of the target object showing each emotional state;

[0118] The emotional state recognition module 26 is used to recognize the emotional state of the target object based on the confidence level.

[0119] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. 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.

[0120] Various embodiments in this specification are described in a progressive manner, and the same or similar parts between various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A speech emotion recognition method based on artificial intelligence, characterized in that: The method comprises: Collecting an analog voice signal of a target object and converting the analog voice signal into a digital voice signal; Dividing the digital voice signal into a plurality of fluctuations, analyzing the change characteristics between the fluctuations, and obtaining characteristic values ​​of the fluctuations that may have emotional changes; Performing RNN circular convolution on the eigenvalue to obtain a characteristic curve corresponding to the fluctuation; Based on the corresponding characteristic curves under known emotional states, analyze the weight of the influence of the characteristic curves of each emotional type on each emotional state; Analyze the numerical characteristics of the characteristic curve of the target object, combine the weight of the characteristic curve of each emotion type on each emotional state, and obtain the confidence of the target object in showing each emotional state; Based on the confidence level, identifying the emotional state of the target object; Analyzing the change characteristics between the fluctuations to obtain the characteristic values ​​of the fluctuations that may have emotional changes, including: Analyze the temporal continuity between adjacent fluctuations, analyze the amplitude changes between adjacent fluctuations, and analyze the number of small amplitudes of the fluctuations, and mark the fluctuations that may have emotional changes; The amplitude of the marked fluctuation and the number of small amplitudes are obtained as characteristic values ​​of the fluctuation that may have emotional changes.

2. The method for speech emotion recognition based on artificial intelligence according to claim 1, characterized in that: Collecting an analog voice signal of a target object and converting the analog voice signal into a digital voice signal includes: Collecting analog voice signals of the target object; Sampling and discretizing the analog voice signal to obtain a discrete voice signal; Amplitude quantization is performed on the discrete speech signal to obtain a digital speech signal.

3. The method for speech emotion recognition based on artificial intelligence according to claim 2, characterized in that: The digital voice signal is divided into a plurality of fluctuations, including: Decoding the digital voice signal and dequantizing it to obtain a voice waveform; By utilizing the gradient direction of the points on the edge of the waveform in the speech waveform diagram, the point where the gradient direction changes from right to left is set as the segmentation point, and the digital speech signal is divided into a plurality of fluctuations.

4. The method for speech emotion recognition based on artificial intelligence according to claim 1, characterized in that: Performing RNN circular convolution on the eigenvalue to obtain a characteristic curve corresponding to the fluctuation, including: Organizing the eigenvalues ​​into a sequence form and performing standardization processing to obtain a standardized eigenvalue sequence; The standardized feature value sequence is input into the trained RNN model, and the characteristic curve corresponding to the fluctuation is output.

5. The method for speech emotion recognition based on artificial intelligence according to claim 1, characterized in that: Based on the corresponding characteristic curves under known emotional states, the weight of the influence of the characteristic curves of each emotional type on each emotional state is analyzed, including: Obtain characteristic curves of each emotion type corresponding to a plurality of simulated speech signals under different emotional states; The numerical size characteristics of the characteristic curves of each emotion type corresponding to multiple simulated voice signals under various emotion states are analyzed to obtain the weight of the influence of each emotion type characteristic curve on each emotion state.

6. The method for speech emotion recognition based on artificial intelligence according to claim 5, characterized in that: Analyze the numerical size characteristics of the characteristic curves of each emotion type corresponding to multiple simulated voice signals under various emotional states, and obtain the weight of the influence of each emotion type characteristic curve on each emotional state, including: Set the threshold of each emotion type characteristic curve; Analyze the relationship between the maximum value on each emotion type characteristic curve corresponding to the plurality of simulated voice signals under various emotional states and the threshold value to obtain a first value influencing parameter; Analyze the magnitude relationship between all values ​​on each emotion type characteristic curve corresponding to multiple simulated voice signals under various emotional states and the threshold value to obtain a first quantity influence parameter; According to the first numerical influencing parameter and the first quantity influencing parameter, obtaining numerical size characteristics of each emotion type characteristic curve corresponding to a plurality of simulated speech signals under various emotional states; Traversing the numerical size features of all emotion type characteristic curves corresponding to multiple simulated speech signals under each emotion state, and summing up the numerical size features to obtain a total numerical size feature; The proportional relationship between the numerical value feature and the total numerical value feature is calculated to obtain the weight of the influence of each emotion type characteristic curve on each emotion state.

7. The method for speech emotion recognition based on artificial intelligence according to claim 6, characterized in that: Analyze the numerical characteristics of the characteristic curve of the target object, combine the weight of the characteristic curve of each emotion type on each emotional state, and obtain the confidence of the target object in showing each emotional state, including: Analyze the relationship between the maximum value on the characteristic curve of the target object and the threshold value to obtain a second value influencing parameter; Analyze the magnitude relationship between all values ​​on the characteristic curve of the target object and the threshold value to obtain a second quantity influence parameter; According to the second numerical influence parameter and the second quantity influence parameter, combined with the weight of the influence of the characteristic curve of each emotion type on each emotional state, the confidence level that the target object exhibits each emotional state is obtained.

8. A speech emotion recognition system based on artificial intelligence, characterized in that: The system comprises: a memory and a processor, wherein: The memory is used to store program codes; The processor is configured to read the program code stored in the memory and execute the method according to any one of claims 1 to 7.

9. The artificial intelligence-based speech emotion recognition system according to claim 8, characterized in that: The processor comprises: A voice signal acquisition module, used to acquire an analog voice signal of a target object and convert the analog voice signal into a digital voice signal; A feature value acquisition module, used to divide the digital voice signal into a plurality of fluctuations, analyze the change characteristics between the fluctuations, and obtain the feature values ​​of the fluctuations that may have emotional changes; A characteristic curve acquisition module, used for performing RNN circular convolution on the characteristic value to obtain a characteristic curve corresponding to the fluctuation; A weight analysis module, for analyzing the weight of the influence of the characteristic curve of each emotion type on each emotion state based on the corresponding characteristic curve under the known emotion state; The confidence analysis module is used to analyze the numerical characteristics of the characteristic curve of the target object, and combine the weight of the characteristic curve of each emotion type on the influence of each emotional state to obtain the confidence of the target object showing each emotional state; An emotional state recognition module, used to complete the recognition of the emotional state of the target object based on the confidence level; Analyzing the change characteristics between the fluctuations to obtain the characteristic values ​​of the fluctuations that may have emotional changes, including: Analyze the temporal continuity between adjacent fluctuations, analyze the amplitude changes between adjacent fluctuations, and analyze the number of small amplitudes of the fluctuations, and mark the fluctuations that may have emotional changes; The amplitude of the marked fluctuation and the number of small amplitudes are obtained as characteristic values ​​of the fluctuation that may have emotional changes.

Images