Audio playing control method and system for intelligent sound equipment, and storage medium

Abstract

The invention relates to the technical field of intelligent sound boxes, and discloses an audio playing control method and system for an intelligent sound box and a storage medium, and the audio playing control method comprises the steps: obtaining environment information and user information; performing environment state recognition based on the environment information; performing emotional state analysis based on the user information; according to the environment state and the emotional state, generating an audio playing control parameter; according to the audio playing control method and device, the audio playing parameters are adaptively adjusted through real-time environment sound analysis, the space sound field control technology is adopted, directional sound release is achieved, smooth transition of the audio parameters is achieved when the environment changes, the user experience is improved, and therefore the problem of environment noise interference is solved.

Description

Technical Field

[0001] This invention relates to the field of smart speakers, and more specifically, to an audio playback control method, system, and storage medium for smart speakers. Background Technology

[0002] With the widespread application of smart speakers in homes, offices, and medical settings, users' demands for intelligent and personalized audio playback are increasing. Existing smart speakers primarily employ rule-based audio playback control methods, which present the following technical challenges: When environmental noise, user emotional state, and application scenarios change dynamically, existing methods cannot achieve adaptive adjustment of audio playback parameters, specifically as follows: 1) Environmental noise interference: When environmental noise changes, fixed volume and timbre parameters lead to a decline in the user's listening experience. Furthermore, in spaces where multiple people are present, it is impossible to achieve directional control of the sound field, which can easily cause mutual interference. 2) Mismatch between emotional state and playback content: The user's emotional state is inconsistent with the playback content, which affects the user experience and lacks a real-time response mechanism for changes in the user's emotional state; 3) Insufficient scene adaptation; unable to optimize playback strategies specifically for different scenarios (such as learning, exercise, treatment, etc.); The root cause of the above problems is that existing technologies lack a systematic solution that organically combines environmental perception, emotion analysis, and scene recognition, resulting in smart speakers being unable to intelligently adjust audio playback according to the actual usage environment. Summary of the Invention

[0003] This invention provides an audio playback control method, system, and storage medium for smart speakers to solve the aforementioned technical problems.

[0004] The first aspect of this invention provides an audio playback control method for a smart speaker, comprising: Obtain environmental and user information; Environmental status identification is performed based on the aforementioned environmental information; Emotional state analysis is performed based on the user information; Based on the environmental and emotional states, audio playback control parameters are generated; Adjust audio playback according to the audio playback control parameters.

[0005] As a further optimization of the present invention, the environmental information includes: Ambient sound signals collected by a microphone array; Scene parameters collected by scene sensors; The scene parameters include at least one of temperature, humidity, and illuminance.

[0006] As a further optimization of the present invention, the user information includes: User voice data; User facial expression data; User behavior data; The user behavior data includes at least one of historical playback records and usage habit data.

[0007] As a further optimization of the present invention, the audio playback control parameters include: Volume parameters; Timbre parameters; Playback content parameters; Sound field control parameters.

[0008] As a further optimization of the present invention, it also includes: Obtain user historical data; A user behavior prediction model is constructed based on the historical data. Predict user intent; The audio playback parameters are pre-adjusted based on the user's intent.

[0009] As a further optimization of the present invention, it also includes: Obtain status information of smart devices; Generate device control commands based on the environmental and emotional states; The audio playback control parameters and device control commands are executed in a coordinated manner.

[0010] A second aspect of the present invention provides an audio playback control system for a smart speaker, comprising: The information collection module is used to acquire environmental and user information; The state analysis module is used to analyze environmental and emotional states. The parameter generation module is used to generate audio playback control parameters; The control execution module is used to perform audio playback control; The processor is used to execute the control logic of the above modules.

[0011] As a further optimization of the present invention, it also includes: The prediction module is used to predict user intent; The linkage control module is used to perform coordinated control of multiple devices; The storage module is used to store system operation data.

[0012] A third aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the audio playback control method for a smart speaker.

[0013] The beneficial effects of this invention are as follows: (1) By analyzing the ambient sound in real time, the audio playback parameters are adaptively adjusted, and spatial sound field control technology is used to achieve directional sound delivery and smooth transition of audio parameters when the environment changes, thereby improving the user experience and solving the problem of environmental noise interference.

[0014] (2) By integrating voice and facial features, the user’s emotional state is accurately identified, and an emotion-audio mapping model is established to optimize content matching and support real-time tracking and dynamic response of emotional state, thereby realizing emotion-driven intelligent playback; (3) Provides specialized control strategies for different application scenarios, realizes seamless switching and parameter optimization between scenarios, supports personalized configuration of scenario modes, and thus provides scenario-based solutions. Attached Figure Description

[0015] Figure 1 This is a flowchart of an audio playback control method for a smart speaker according to the present invention; Figure 2 This is an environmental sound analysis data graph from Embodiment 1 of the present invention; Figure 3 This is a user sentiment analysis data graph from Embodiment 1 of the present invention; Figure 4 This is a diagram of audio playback control parameters according to Embodiment 1 of the present invention. Detailed Implementation

[0016] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0017] This invention provides an adaptive audio playback control method for intelligent speakers based on environmental perception and emotion analysis. The invention will be further described in detail below with reference to specific embodiments.

[0018] Example 1: Basic Audio Playback Control Scheme. This example provides a basic scheme for adaptive audio playback control of intelligent speakers based on environmental perception and emotion analysis, such as... Figure 1As shown, the solution includes the following steps: Step 1, environmental sound analysis, specifically includes: Step 101: Collect ambient sound signals through the built-in microphone array of the speaker; Step 102: Perform a Fast Fourier Transform (FFT) on the acquired audio signal to obtain the frequency domain information, as shown in the following formula: ; in, Let be the frequency domain signal after discrete Fourier transform, representing the _____. The complex values ​​of each frequency component It is a discrete-time signal. The number of sampling points. Indicates from arrive Discrete summation, The imaginary unit, These are the basis functions of the Fourier transform. Frequency number; Step 103: Calculate the energy distribution in different frequency bands, using the following formula: ; The corresponding frequency is: ; in, For the first Energy value at each frequency point For the first Complex spectral values ​​at each frequency point Representing complex numbers The model, For the first The actual frequency value (unit: Hz) corresponding to each frequency point. Sampling frequency (unit: Hz) The number of points for the Fast Fourier Transform; Step 104, calculate the... The energy distribution characteristics formed by the distribution at different frequency points are visualized to generate a spectrum, where the horizontal axis represents frequency and the vertical axis represents energy value. This spectrum is then input into a pre-trained CNN (Convolutional Neural Network) environmental sound classification model, which includes: ; The formula for convolution operation is as follows: ; In convolution operations: Indicates the input layer. Indicates a convolutional layer. Indicates a fully connected layer. Indicates the output layer. To output feature maps in The value of the position, Indicates to × Double summation is performed using convolution kernels of varying sizes. For convolution kernel weights, For the input feature map in The value of the position, The kernel size is [size]. For activation function, For max pooling; Step 105: Obtain the type of ambient sound (e.g., quiet, noisy, etc.).

[0019] Step 2, user sentiment analysis steps, specifically include: Step 201, Voice Sentiment Analysis: (1) Collect user voice signals; (2) Extract MFCC features, calculation process: 1) Pre-intensification: ; in, The signal after pre-emphasis. The input signal at the current moment, The input signal from the previous moment; 2) Frame-by-frame windowing: using Hamming windows: ; in, This is the output value of the Hamming window function. This is the sequence number of the current sampling point. The length of the window function. It is a cosine function; 3) Take the logarithm after FFT transformation: ; in, For the first Logarithmic spectrum at frequency points For the first The absolute value of the FFT amplitude spectrum at each frequency point. Frequency number; 4) Mel filtering: ; in, For the first The output of the Mel filter This indicates that for all frequency points Summation, For the first The energy spectrum at each frequency point For the first A Mel filter at frequency point The response at the location, The number of FFT points; 5) The DCT transform yields the MFCC coefficients, where DCT stands for Discrete Cosine Transform and MFCC coefficients are Mel-frequency cepstral coefficients. (3) Input the MFCC features into the LSTM (Long Short-Term Memory) sentiment classification model, which includes: Forgotten Gate: ; in, The output value of the forget gate is in the range [0, 1]. Here is the forget gate weight matrix. For the sigmoid function, This is the hidden state from the previous moment. For the current input, Forget gate bias term; Input Gate: ; in, The input gate output value is in the range [0, 1]. The input gate weight matrix, This is the hidden state from the previous moment. For the current input, For the input gate bias term, For the sigmoid function: ; Output gate: ; in, The output value of the output gate is in the range [0, 1]. This is the output gate weight matrix. This is the hidden state from the previous moment. For the current input, For the output gate bias term, It is the sigmoid function; Cell status: ; in, This represents the current state of the cell. The output value of the forget gate controls how much of the previous cell state is retained. This refers to the cell state at the previous time step. The unit state weight matrix, The output value of the input gate is used to control the proportion of new information input. The hyperbolic tangent activation function maps values ​​to the interval [-1, 1]. This is the hidden state from the previous moment. For the current input, For cell state bias terms; Hidden state: ; in, Hide the current state. The output value of the output gate is used to control the proportion of the output information. The hyperbolic tangent activation function is used. This represents the current state of the cell. Step 202, Facial Expression and Emotion Analysis: (1) Acquire user facial images via camera ; (2) Perform face detection and feature point extraction to obtain the feature vector: ; (3) Input an SVM facial expression classification model, where SVM stands for Support Vector Machine and uses a Gaussian kernel function: ; Obtain the emotion category corresponding to the expression; Step 203, Emotional Integration: (1) The sentiment analysis results of speech and facial expressions were fused using a weighted average method; (2) Obtain the end user's emotional category (such as happy, sad, etc.).

[0020] Step 3, Audio playback adjustment steps: Step 301, establish a multidimensional mapping table M: ​​This mapping table associates the type of ambient sound with the user's emotional category. Through this association, a set of audio playback control parameters can be obtained, including volume parameters, timbre parameters, and genre parameters. Step 302: Based on the current ambient sound type and the user's emotional category, query the corresponding control parameters from the mapping table M; Step 303: Apply parameters to adjust audio playback.

[0021] Specific application examples of Example 1 The following example illustrates the practical application effect of this embodiment. The specific scenario is described as follows: A user is listening to music using a smart speaker in the family living room. During this time, other family members are talking, the ambient noise level changes, and the user's mood changes along with the music.

[0022] Environmental sound analysis data such as Figure 2 As shown; The frequency bands corresponding to the frequency domain energy distribution are: [100Hz, 500Hz, 1kHz, 2kHz, 4kHz]. User sentiment analysis data such as Figure 3 As shown; Among them, the frequency bands corresponding to MFCC features are: [0-1kHz, 1-2kHz, 2-3kHz, 3-4kHz]; the facial expression feature points are: [eyebrow spacing, eye angle, mouth angle, eyebrow height, forehead wrinkles]. Audio playback control parameters such as Figure 4 As shown; The EQ adjustment bands are: [100Hz, 500Hz, 1kHz, 2kHz, 4kHz].

[0023] Performance Analysis: The data above shows that the system is able to: 1) Accurately identify different environmental noise levels (confidence level above 0.87); 2) Accurately determine the user's emotional state (confidence level above 0.85); 3) Adaptively adjust audio playback parameters according to changes in environment and emotion: increase volume appropriately (+15dB) when the environment is noisy (60s); adjust the timbre to highlight the mid and low frequencies and select a cheerful style when the user is happy (30s); adjust the timbre to highlight the high frequencies and select a gentle style when the user is tired (60s).

[0024] Real-world testing shows that this solution can effectively improve the user's listening experience, increasing user satisfaction by 35%.

[0025] Example 2: An audio playback control scheme with music prediction function. This example adds music prediction and pre-buffering functions to the scheme based on Example 1. For example, based on the steps in Embodiment 1, it further includes: Step 4, Song prediction model construction steps: Step 401: Obtain the user's historical music playback record sequence. ; Step 402, construct a prediction model based on GRU (Gated Recurrent Unit): Update gate calculation: ; in, for Update the gate vector at each time step. To update the gate weight matrix, for The hidden state at all times for The input vector at time t, It is the sigmoid activation function; Reset door calculation: ; in, for Reset gate vector at time step To reset the gate weight matrix, for The hidden state at all times for The input vector at time t, It is the sigmoid activation function; Candidate hidden state calculation: ; in, for The candidate hidden state at time step. The hyperbolic tangent activation function is used. Let be the candidate hidden state weight matrix. To reset the gate vector, for The hidden state at all times for The input vector at time t, Represents element-wise multiplication of vectors; Hidden status update: ; in, for The hidden state at all times To update the gate vector, This indicates the proportion of historical information retained. for The hidden state at all times In the candidate hidden state, This represents element-wise multiplication of vectors.

[0026] Step 5, Application of Prediction Results: Step 501, Expand the mapping table N: This mapping table associates the ambient sound type, the user's emotional category, and the predicted song list. Through this association, a set of audio playback control parameters can be obtained, including volume parameters, timbre parameters, genre parameters, and buffering instructions. Step 502: Obtain parameters from mapping table N based on the current state; Step 503: Perform audio playback adjustments and song pre-caching.

[0027] Example 3: Audio playback control scheme for smart home appliance linkage. This example adds linkage control function with other smart home appliances to the existing example 1.

[0028] For example, based on the steps in Embodiment 1, it further includes: Step 6, Smart Home Appliance Status Acquisition Steps: Step 601: Obtain the status information of smart lighting devices via smart home protocols (such as ZigBee, Wi-Fi): Brightness value: L∈[0,100]; Color temperature value: T∈[2700K, 6500K]; Step 602, Obtain air conditioning equipment status information: Temperature values: Temp ∈ [16℃, 30℃]; Operating mode: Mode∈{cooling, heating, dehumidification, ventilation}.

[0029] Step 7, Home Appliance Linkage Decision-Making Steps: Step 701, construct the decision tree model DecisionTree, which includes: Input node: Emotion category; Intermediate nodes: Select the optimal splitting feature based on information gain; Leaf nodes: Equipment adjustment parameter combinations; Step 702, Decision Process: ifEmotion==“Happy”: return{Brightness: +20, Color Temperature: +500K, Temperature: -1℃}elifEmotion==“Sadness”: return{Brightness: -10, Color Temperature: -300K, Temperature: +1℃}else: return{Brightness: 0, Color Temperature: 0, Temperature: 0}.

[0030] Step 8, Execution steps of linkage control: Step 801, Expand the mapping table P: This mapping table associates the ambient sound type, user emotion category and air conditioning device status. Through this association, a set of audio playback control parameters can be obtained, including audio parameters, lighting parameters and air conditioning parameters. Step 802: Obtain parameters from mapping table P based on the current state; Step 803: Send control commands to each device via the smart home protocol.

[0031] Example 4: Audio playback control scheme for language learning scenarios. This example is designed for language learning scenarios and adds an intelligent adjustment function for learning content based on Example 1.

[0032] For example, based on the steps in Embodiment 1, it further includes: Step 9, Learning Status Assessment Steps: Step 901: Obtain the user's learning history ,in Include: (1) Difficulty of learning content: ; (2) Test results: ; (3) Study duration: ; Step 902, construct the Bayesian network model: (1) Define the learning level hypothesis variables ; (2) Calculate the posterior probability: ; in, Assuming given evidence E The posterior probability, Given the hypothesis Evidence observed at the time The likelihood probability, Assumption The prior probability, As evidence The marginal probability of is calculated using the following formula: ; The summation is applied to all possible assumptions. conduct.

[0033] Step 10, Learning Content Selection Steps: Step 1001, based on the user's emotion category (Emotion) and learning level Construct the content selection function Select: If Emotion == "happy" and H == "high level": return to the advanced oral practice content; If Emotion == "frustrated" and H == "low level": return to basic vocabulary learning content; Other situations: Return to the listening content of medium difficulty.

[0034] Step 11, Learning Content Playback Steps: Step 1101, Expand the mapping table Q: This mapping table associates the type of ambient sound, the user's emotional category, and the type of learning content. Through this association, a set of audio playback control parameters can be obtained, including volume parameters, speech rate parameters, and repetition count. Step 1102: Obtain parameters from mapping table Q based on the current state; Step 1103: Execute playback control for the learning content.

[0035] Example 5: Audio playback control scheme for gym scene. This example is for gym scene and adds motion state perception and music rhythm matching functions on the basis of Example 1.

[0036] For example, based on the steps in Embodiment 1, it further includes: Step 12, Motion State Perception Step: Step 1201, Obtain user motion data: (1) Heart rate data: HR∈[40,200] beats / minute; (2) Type of exercise: Type∈{running, weightlifting, yoga, ...}; Step 1202, Exercise intensity calculation: (1) Maximum heart rate: HRmax = 220 - age; (2) Exercise intensity: Intensity = HR / HRmax × 100%; (3) Strength grading: Low intensity: Intensity < 50%; Medium intensity: 50% ≤ Intensity < 70%; High strength: Intensity ≥ 70%.

[0037] Step 13, Music Rhythm Matching Steps: Step 1301, Construct a motion-music mapping model: Input feature vector: [Intensity, Type, Emotion]; Output parameter vector: [Tempo, Volume, Style]; Step 1302, rhythm matching rules: ifType==“Running”: Tempo=HR×1.2 / / BPM synchronized with heart rate elifType==“Weightlifting”: Tempo=120+Intensity×40 / / The higher the intensity, the faster the rhythm elifType==“Yoga”: Tempo=60+Intensity×20 / / Maintain a slower rhythm.

[0038] Step 14, Audio Playback Control Steps: Step 1401, Expand the mapping table R: This mapping table associates the type of ambient sound, the user's emotional category, and the motion state. Through this association, a set of audio playback control parameters can be obtained, including rhythm parameters, volume parameters, and genre parameters. Step 1402: Obtain parameters from mapping table R based on the current state; Step 1403: Execute audio playback control and adjust in real time to match the rhythm of the movement.

[0039] Example 6: Audio playback control scheme for hospital ward scene. This example is for hospital ward scene and adds disease perception and spatial sound control functions on the basis of Example 1.

[0040] For example, based on the steps in Embodiment 1, it further includes: Step 15, Steps for obtaining the patient's condition status: Step 1501: Obtain patient information through the hospital information system. Recovery stages: Stage ∈ {early postoperative period, mid-recovery period, late recovery period}; Step 1502: Obtain sleep status through sleep monitoring equipment: Electroencephalogram (EEG) characteristics: EEG(t); Sleep stages: Sleep ∈ {awake, light sleep, deep sleep}.

[0041] Step 16, Spatial Sound Control Steps: Step 1601, Construct the sound field control model: Input: Room dimensions Sound speaker location Location of hospital bed ; Calculate the sound pressure level distribution: ; in, For any point in space The sound pressure level at that location is measured in dB. The sound pressure level of the sound source. This is the geometric divergence attenuation term of the sound wave. For transmission distance, The air absorption coefficient; Step 1602, optimize playback parameters: Objective function: ; in, This represents minimizing the objective function. This indicates the maximum sound pressure level at other hospital bed locations. The sound pressure level at the target bed location, This is the minimum effective sound pressure level.

[0042] Step 17, Treatment Audio Control Steps: Step 1701, Expand the mapping table S: This mapping table associates the type of ambient sound, the user's emotional category, and the state of illness. Through this association, a set of audio playback control parameters can be obtained, including volume parameters, timbre parameters, and spatial control parameters. Step 1702: Obtain parameters from mapping table S based on the current state; Step 1703: Execute audio playback control to ensure that the sound only affects the target area.

[0043] Example 7: Audio playback control scheme for driving scenarios. This example is designed for car driving scenarios and adds driving status perception and intelligent adjustment of navigation voice functions on the basis of Example 1.

[0044] For example, based on the steps in Embodiment 1, it further includes: Step 18, Driving Environment Perception Steps: Step 1801: Obtain vehicle driving status: Speed: ∈[0, 200] km / h; Road conditions: ∈{smooth traffic, congestion, accident}; Step 1802, Obtain driving environment information: In-vehicle noise: Measurements were taken using an in-vehicle microphone array. Weather outside the car: ∈{sunny, rainy, snowy, foggy}; Step 1803, Calculate the environment complexity index: Environment complexity calculation formula: ; in, This is an environmental complexity index, representing the overall complexity of the current driving environment. Current vehicle speed (unit: km / h) This represents the current in-vehicle noise level. The maximum speed limit is usually set at 200 km / h. This is the maximum noise limit (usually taken as 90dB). For the road condition normalization function: , Weather influencing factors , Let be the weighting coefficient, satisfying .

[0045] Step 19, Navigation voice adjustment steps: Step 1901, Construct an adaptive speech parameter model: Input: Environment complexity Emotional Category ; Output: Speech parameter vector ; Step 1902, Parameter Calculation Rules: Volume adjustment: ; Speech speed adjustment: ; Tone adjustment: ; in, Base volume (usually 65dB). The basic speaking speed is (usually 180 words per minute). Set the basic timbre parameter to 0. For sentiment evaluation function: , The adjustment coefficient is determined experimentally to obtain the optimal value.

[0046] Step 20, Integrated Control Steps: Step 2001, Extend the mapping table T: This mapping table associates the ambient sound type, user emotion category and driving state. Through this association, a set of audio playback control parameters can be obtained, including navigation parameters, music parameters and prompt tone parameters. Step 2002: Obtain parameters from mapping table T based on the current state; Step 2003, execute hierarchical control: Emergency situation: Only navigation and warning prompts will be played; Normal driving: Balance navigation and music playback; Parking wait: Focus on music playback.

[0047] Through the above seven embodiments, this invention comprehensively demonstrates the application schemes of the intelligent speaker adaptive audio playback control system based on environmental perception and emotion analysis in different scenarios, reflecting the technical innovation and practical value of this invention. By organically combining environmental perception and emotion analysis, it realizes intelligent control of intelligent speakers in various scenarios, significantly improving the user experience.

[0048] The embodiments of the present invention have been described above, but the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the embodiments described above, all of which are within the protection scope of the embodiments described above.

Claims

1. An audio playback control method for a smart speaker, characterized in that, include: Obtain environmental and user information; Environmental status identification based on environmental information; Sentiment analysis based on user information; Generate audio playback control parameters based on environmental and emotional states; Adjust audio playback according to the audio playback control parameters.

2. The audio playback control method for a smart speaker according to claim 1, characterized in that, Environmental status identification based on environmental information includes the following steps: Step 101: Collect ambient sound signals through the built-in microphone array of the speaker; Step 102: Perform a Fast Fourier Transform on the acquired sound signal to obtain the frequency domain information, as shown in the following formula: ； in, Let be the frequency domain signal after discrete Fourier transform, representing the _____. The complex values ​​of each frequency component It is a discrete-time signal. The number of sampling points. Indicates from arrive Discrete summation, The imaginary unit, These are the basis functions of the Fourier transform. Frequency number; Step 103: Calculate the energy distribution in different frequency bands, using the following formula: ； The corresponding frequency is: ； in, For the first Energy value at each frequency point For the first Complex spectral values ​​at each frequency point Representing complex numbers The model, For the first The actual frequency value corresponding to each frequency point Sampling frequency, The number of points for the Fast Fourier Transform; Step 104, calculate the... The energy distribution characteristics formed by the distribution at different frequency points are visualized to generate a spectrogram, where the horizontal axis represents frequency and the vertical axis represents energy value. This spectrogram is then input into a pre-trained CNN environmental sound classification model, which includes: ； The formula for convolution operation is as follows: ； In convolution operations: Indicates the input layer. Indicates a convolutional layer. Indicates a fully connected layer. Indicates the output layer. To output feature maps in The value of the position, Indicates to × Double summation is performed using convolution kernels of varying sizes. For convolution kernel weights, For the input feature map in The value of the position, The kernel size is [size]. For activation function, For max pooling; Step 105: Obtain the ambient sound type.

3. The audio playback control method for a smart speaker according to claim 1, characterized in that, The steps involved in sentiment analysis based on user information include: Step 201, Voice Emotion Analysis: Collect user voice signals; Extracting MFCC features, calculation process: Pre-intensity: ； in, The signal after pre-emphasis. The input signal at the current moment, The input signal from the previous moment; Frame-by-frame windowing: using Hamming windows: ； in, This is the output value of the Hamming window function. This is the sequence number of the current sampling point. The length of the window function. It is a cosine function; Take the logarithm after FFT transformation: ； in, For the first Logarithmic spectrum at a frequency point For the first The absolute value of the FFT amplitude spectrum at each frequency point. Frequency number; Mel filtering: ； in, For the first The output of the Mel filter This indicates that for all frequency points Summation, For the first The energy spectrum at each frequency point For the first A Mel filter at frequency point The response at the location, The number of points for the Fast Fourier Transform; Discrete cosine transform yields the MFCC coefficients, where the MFCC coefficients are Mel-frequency cepstral coefficients; The MFCC features are input into an LSTM sentiment classification model, which includes: Forgotten Gate: ； in, The output value of the forget gate is in the range [0, 1]. Here is the forget gate weight matrix. For the sigmoid function, This is the hidden state from the previous moment. For the current input, Forget gate bias term; Input Gate: ； in, The input gate output value is in the range [0, 1]. The input gate weight matrix, This is the hidden state from the previous moment. For the current input, For the input gate bias term, It is the sigmoid function; Output gate: ； in, The output value of the output gate is in the range [0, 1]. This is the output gate weight matrix. This is the hidden state from the previous moment. For the current input, For the output gate bias term, The sigmoid function is used; cell states: ； in, This represents the current state of the cell. The output value of the forget gate controls how much of the previous cell state is retained. This refers to the cell state at the previous time step. The unit state weight matrix, The output value of the input gate is used to control the proportion of new information input. The hyperbolic tangent activation function maps values ​​to the interval [-1, 1]. This is the hidden state from the previous moment. For the current input, For cell state bias terms; Hidden state: ； in, Hide the current state. The output value of the output gate is used to control the proportion of the output information. The hyperbolic tangent activation function is used. This represents the current state of the cell. Step 202, Facial Expression and Emotion Analysis: Capture user's facial image via camera ; Face detection and feature point extraction are performed to obtain feature vectors: ； Will Input a support vector machine facial expression classification model, using a Gaussian kernel function: ； Obtain the emotion category corresponding to the expression; Step 203, Emotional Integration: The sentiment analysis results of speech and facial expressions are combined using a weighted average method; Obtain the end user's emotional category.

4. The audio playback control method for a smart speaker according to claim 1, characterized in that, The steps for generating audio playback control parameters based on environmental and emotional states include: Step 301, establish a multidimensional mapping table M: ​​the mapping table M associates the type of ambient sound with the user's emotional category, and obtains the corresponding control parameters for audio playback; Step 302: Based on the current ambient sound type and the user's emotional category, query the corresponding audio playback control parameters from the mapping table M. The audio playback control parameters include: Volume parameters; Timbre parameters; Playback content parameters; Sound field control parameters.

5. The audio playback control method for a smart speaker according to claim 1, characterized in that, Also includes: Obtain user historical data; Build a user behavior prediction model based on historical data; Predict user intent; Pre-adjust audio playback parameters based on user intent.

6. The audio playback control method for a smart speaker according to claim 1, characterized in that, Also includes: Obtain status information of smart devices; Generate device control commands based on environmental and emotional states; Coordinated control of audio playback control parameters and device control commands.

7. An audio playback control system for a smart speaker, used to execute the audio playback control method for a smart speaker according to any one of claims 1-6, characterized in that, include: The information collection module is used to acquire environmental and user information; The state analysis module is used to analyze environmental and emotional states. The parameter generation module is used to generate audio playback control parameters; The control execution module is used to perform audio playback control; The processor is used to execute the control logic of the information acquisition module, status analysis module, parameter generation module, and control execution module.

8. An audio playback control system for a smart speaker according to claim 7, characterized in that, Also includes: The prediction module is used to predict user intent; The linkage control module is used to perform coordinated control of multiple devices; The storage module is used to store system operation data.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, is capable of performing an audio playback control method for a smart speaker as described in any one of claims 1-6.

Images