Power consumption management method and device, electronic device, and storage medium

Abstract

The application provides a power consumption management method and device, electronic equipment and storage medium, and belongs to the field of intelligent adjustment. The method comprises the following steps: in response to receiving a voice signal of a target user, extracting acoustic features and semantic features of the voice signal, the target user being a user wearing smart glasses; weighting and fusing the acoustic features and the semantic features based on a conversation scene in which the target user is located to obtain target features; predicting a resource allocation strategy of the smart glasses based on the target features, and managing the power consumption of the smart glasses based on the resource allocation strategy, wherein the resource allocation strategy comprises an uplink resource allocation strategy and a downlink resource allocation strategy. The power consumption management method and device, electronic equipment and storage medium provided by the application can improve the endurance of the smart glasses.

Description

Technical Field

[0001] This application belongs to the field of intelligent regulation technology, and more specifically, relates to a power consumption management method and device, electronic device, and storage medium. Background Technology

[0002] Currently, with the continuous development of technology, the application scenarios and number of smart glasses are constantly increasing. Power management, as a key technology affecting the battery life of smart glasses, is undeniably important. At present, most smart glasses manage power consumption by setting sleep / wake strategies.

[0003] Existing technologies reserve fixed and relatively sufficient uplink and downlink resources during conversations. However, reserving fixed uplink and downlink resources can lead to resource waste and affect the battery life of smart glasses. Summary of the Invention

[0004] The purpose of this application is to provide a power management method, device, electronic device, and storage medium to improve the dynamic management capabilities of uplink and downlink resources of smart glasses and improve the battery life of smart glasses.

[0005] A first aspect of this application provides a power consumption management method, including:

[0006] In response to receiving the voice signal from the target user, the acoustic and semantic features of the voice signal are extracted. The target user is a user wearing smart glasses.

[0007] Based on the dialogue context of the target user, acoustic features and semantic features are weighted and fused to obtain target features;

[0008] Based on target features, a resource allocation strategy for smart glasses is predicted, and power consumption management of smart glasses is performed based on the resource allocation strategy. The resource allocation strategy includes an uplink resource allocation strategy and a downlink resource allocation strategy.

[0009] A second aspect of this application provides a power management device, including:

[0010] The feature extraction module is used to extract the acoustic and semantic features of the voice signal in response to the received voice signal from the target user, who is a user wearing smart glasses.

[0011] The feature fusion module is used to weightedly fuse acoustic features and semantic features based on the dialogue scenario of the target user to obtain the target features;

[0012] The power management module is used to predict the resource allocation strategy of the smart glasses based on target features, and to manage the power consumption of the smart glasses based on the resource allocation strategy. The resource allocation strategy includes uplink resource allocation strategy and downlink resource allocation strategy.

[0013] A third aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the power management method described above.

[0014] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the power management method described above.

[0015] The beneficial effects of the power management method, apparatus, electronic device, and storage medium provided in this application are as follows: By extracting acoustic and semantic features from the received voice signal of a target user, this application can effectively extract the user's dialogue state. Subsequently, target features are generated by fusing the acoustic and semantic features, and then the resource allocation strategy for the smart glasses is predicted based on these target features, dynamically adjusting the uplink and downlink resource allocation strategies of the smart glasses. Compared to fixed uplink and downlink resources, this improves resource utilization, reduces resource waste, and enhances the battery life of the smart glasses. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart illustrating a power management method provided in an embodiment of this application;

[0018] Figure 2 This is a structural block diagram of a power management device provided in an embodiment of this application;

[0019] Figure 3 This is a schematic block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0020] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following description will be provided in conjunction with the accompanying drawings and specific embodiments.

[0022] Please refer to Figure 1 , Figure 1 This is a flowchart illustrating a power management method provided in an embodiment of this application, which can be executed by a processor in smart glasses. The method may include steps S101 to S103.

[0023] S101, in response to receiving the voice signal from the target user, extracts the acoustic and semantic features of the voice signal, where the target user is a user wearing smart glasses.

[0024] In this embodiment, smart glasses are typically bound to a user through identity information, and this bound user is generally a target user. When the target user makes a voice call or performs voice control, the smart glasses can recognize or respond to the target user's voice signal.

[0025] After smart glasses detect a target user's voice signal through directional microphone pickup or voiceprint recognition, they can extract features from that voice signal in two ways. First, they can extract physical attribute features, obtaining acoustic features. Second, they can extract content information features, obtaining semantic features. Acoustic features can include frequency, energy, or spectral characteristics of the voice signal, representing its inherent characteristics and serving as a basis for evaluating its quality. Semantic features can include the intent or meaning of the voice signal, representing the target user's core needs.

[0026] In the embodiments of this application, a lightweight dual-feature extraction model can be pre-constructed and embedded in the storage space of smart glasses, enabling rapid feature extraction of received speech signals and quick acquisition of acoustic and semantic features.

[0027] Specifically, the lightweight dual-feature extraction model mainly includes a shared feature layer, an acoustic feature extraction branch, and a semantic feature extraction branch.

[0028] The shared feature layer can extract the basic spectral features of speech signals quickly and in a lightweight manner through "depth-separable convolution + residual blocks". As a preprocessing step for acoustic and semantic feature extraction, it can reduce redundant computation.

[0029] The acoustic feature extraction branch can extract acoustic features from the basic spectral features using a lightweight time-series model. Acoustic features can include metrics such as the mean of Mel-Frequency Cepstral Coefficients (MCFF), Short-Time Energy (STE), and Zero Crossing Rate (ZCR).

[0030] The semantic feature extraction branch can extract semantic features from the basic spectral features using a two-layer Transformer Encoder. Semantic features can include intent information or question information, etc.

[0031] The loss function in the dual-feature extraction model of this application embodiment can be composed of two parts: the acoustic feature mean squared error loss (MSE) and the semantic feature cross-entropy loss. The weight ratio is 1:1, which can more evenly and accurately identify acoustic features and semantic features.

[0032] The dual-feature extraction model constructed in this application embodiment can simultaneously output reliable acoustic and semantic features on smart glasses, which is beneficial for subsequent resource allocation decisions.

[0033] S102, based on the dialogue scenario of the target user, the acoustic features and semantic features are weighted and fused to obtain the target features.

[0034] In this embodiment, the dialogue scenario refers to the environment and interaction type in which the target user engages in voice interaction using smart glasses. For example, a dialogue scenario may include a call scenario or a question-and-answer scenario. It may also include a subway call scenario, a subway question-and-answer scenario, a sports call scenario, or a sports question-and-answer scenario, etc.

[0035] Different dialogue scenarios have varying levels of background noise and requirements for speech clarity. Therefore, acoustic and semantic features can be weighted and fused based on the dialogue scenario to obtain the target features for the current dialogue scenario.

[0036] Specifically, by assigning different weights to acoustic and semantic features, and then concatenating them using a weighted sum, the semantic and acoustic features can be vector-concatenated to obtain the target features, which better reflect the current dialogue scenario. The sum of the weights of the acoustic and semantic features is 1.

[0037] Optionally, embodiments of this application can pre-set fixed weights for different dialogue scenarios, and select different weight allocations for different scenarios. For example, in a subway scenario (where strong noise requires priority noise reduction), the target feature is calculated as: acoustic feature × 0.7 + semantic feature × 0.3. In a bedroom scenario (where weak noise requires priority parsing), the target feature is calculated as: acoustic feature × 0.3 + semantic feature × 0.7.

[0038] Alternatively, the correlation between scene noise and acoustic and semantic features in a dialogue scenario can be calculated in real time, and weights can be allocated according to the correlation. For example, in an outdoor scene (scene noise 55dB), the correlation between acoustic features and scene noise is calculated to be 0.6, and the correlation between semantic features and scene noise is calculated to be 0.4. The weights can be allocated in a 3:2 ratio: Target feature → Acoustic feature × 0.6 + Semantic feature × 0.4. In a call scene (scene noise 45dB), the correlation between acoustic features and scene noise is calculated to be 0.5, and the correlation between semantic features and scene noise is calculated to be 0.5. The weights can be allocated in a 1:1 ratio: Target feature → Acoustic feature × 0.5 + Semantic feature × 0.5. The specific calculation method can be selected according to the actual situation.

[0039] This application embodiment obtains target features by weighted fusion of semantic features and acoustic features. These target features can cover environmental noise information and user demand information, providing strong support for the subsequent allocation of execution resources for smart glasses.

[0040] S103, predicts the resource allocation strategy of the smart glasses based on target features, and manages the power consumption of the smart glasses based on the resource allocation strategy, wherein the resource allocation strategy includes an uplink resource allocation strategy and a downlink resource allocation strategy.

[0041] In this embodiment, the uplink resource allocation strategy is mainly used to allocate uplink resources in the smart glasses. Uplink resources are the computing resources required by the smart glasses to process and understand the voice signals of the target user. Downlink resources are the computing resources required by the smart glasses to process and output information from other electronic devices connected to the smart glasses or from information that the smart glasses themselves need to output.

[0042] Optionally, a resource allocation mapping table can be pre-established. This table includes multiple standard target features and corresponding resource allocation strategies for different standard target features. The resource allocation strategy for each standard target feature is the optimal resource allocation strategy under that standard target feature. Under this optimal resource allocation strategy, the uplink and downlink resource consumption of the smart glasses is low, the allocation is reasonable, and battery life is improved. In this embodiment, the resource allocation strategy can be determined by looking up the table after the target features are actually obtained. Alternatively, the target features can be input into a pre-determined resource allocation model to determine the resource allocation strategy for the smart glasses.

[0043] This application embodiment effectively extracts the user's dialogue state by extracting acoustic and semantic features from the received voice signal of the target user. Subsequently, target features are generated by fusing the acoustic and semantic features, and then the resource allocation strategy for the smart glasses is predicted based on these target features, dynamically adjusting the uplink and downlink resource allocation strategies of the smart glasses. Compared to fixed uplink and downlink resources, this improves resource utilization, reduces resource waste, and extends the battery life of the smart glasses.

[0044] In one embodiment of this application, the acoustic features include multiple first features with the same dimensions, and the semantic features include multiple second features with the same dimensions, wherein each first feature and each second feature has the same dimensions.

[0045] To ensure that acoustic features and semantic features have the same dimension and can be fused for computation, the embodiments of this application can perform same-dimensional processing during the extraction of acoustic features and semantic features of speech signals, thus processing the acoustic features and semantic features into features of the same dimension.

[0046] Generally, after extracting acoustic features from a speech signal, multiple first features are obtained, which represent different acoustic features. After extracting semantic features from a speech signal, multiple second features are obtained, which represent different semantic features. Each first feature and each second feature has the same dimension.

[0047] Based on the dialogue scenario of the target user, acoustic features and semantic features are weighted and fused to obtain target features, which may include S1021 to S1024.

[0048] S1021, Obtain the scene noise of the dialogue scene in which the target user is located.

[0049] Smart glasses can collect environmental signals from the conversation scene through a built-in microphone array, and process the environmental signals to obtain the scene noise of the conversation scene.

[0050] This application embodiment can calculate the equivalent continuous sound pressure level (LAeq, in dB) of the collected sound signal to characterize the noise intensity of the scene noise, and at the same time determine the noise type of the scene noise through spectrum analysis (short-time Fourier transform). For example, steady-state noise: spectrum fluctuation ≤10%; non-steady-state noise: spectrum fluctuation >10%.

[0051] This application provides an accurate noise benchmark for subsequent correlation calculations, avoiding correlation deviations caused by using fixed scene labels to replace real noise data. Simultaneously, it allows for dynamic updates to ensure real-time matching of noise data with the current scene, improving the timeliness of subsequent feature weights.

[0052] S1022, calculate the correlation degree between scene noise and each first feature to obtain a first correlation degree set, determine the weight of each first feature based on the first correlation degree set, and perform weighted fusion of each first feature based on the weight of each first feature to obtain acoustic fusion features.

[0053] In this embodiment of the application, for each first feature, the Pearson correlation coefficient between the first feature and the scene noise can be calculated as the first correlation degree; and all the first correlation degrees can be combined into a first correlation degree set.

[0054] Specifically, the first feature can be multiple continuous dimensions of the acoustic feature, each dimension being a continuous numerical sequence corresponding to feature values ​​at different time frames. In this case, the scene noise can be a noise intensity sequence, which can be aligned with the time frames of the first feature. For each first feature, the Pearson correlation coefficient between the first feature and the noise intensity sequence can be calculated. The absolute value of this Pearson correlation coefficient can be taken as the first correlation degree, with a range of (0, 1). A larger value indicates a stronger linear correlation between the first feature and the noise intensity.

[0055] After obtaining multiple first-level correlations, a set of first-level correlations can be obtained. Subsequently, this set can be normalized to obtain the weight of each first-level feature. The weight of each first-level feature is the proportion of the correlation corresponding to that feature weight in all correlations, where the sum of the weights of all first-level features is 1.

[0056] Finally, the acoustic fusion features can be obtained by weighted fusion of the first features based on their respective weights.

[0057] The embodiments of this application can effectively identify acoustic dimensions that change significantly with noise intensity, making the acoustic fusion features more reflective of the acoustic information most sensitive to noise, and providing targeted input for subsequent noise reduction processing and resource allocation.

[0058] S1023, calculate the correlation between scene noise and each second feature to obtain a second correlation set, determine the weight of each second feature based on the second correlation set, and perform weighted fusion of each second feature based on the weight of each second feature to obtain semantic fusion features.

[0059] In this embodiment of the application, for each second feature, the mutual information degree between the second feature and the scene noise can be calculated as the second correlation degree; and all the second correlation degrees can be combined into a second correlation degree set.

[0060] The second feature can be multiple dimensions of the semantic feature, and can contain continuous or discrete values. In this case, scene noise can be labeled with a noise type. For each second feature, it can be discretized first. Then, the mutual information between the second feature and the noise type can be calculated to determine the strength of their nonlinear correlation. Finally, the obtained mutual information can be normalized to obtain multiple second correlation degrees. The value of each second correlation degree is in the range of (0, 1), and the larger the value, the more significant the influence of the noise type on the semantic feature.

[0061] After obtaining multiple second-degree associations, a set of second-degree associations can be obtained. Subsequently, based on the principle of prioritizing reliability, the normalized weights of each second feature can be calculated.

[0062] Finally, the second features can be weighted and fused based on their respective weights to obtain semantic fusion features.

[0063] The embodiments of this application can effectively identify semantic dimensions that are greatly affected by noise types. Semantic fusion features focus on semantic information that is still reliable under noise, avoiding low-quality semantic features from misleading subsequent decisions.

[0064] S1024, the acoustic fusion features and semantic fusion features are weighted and fused to obtain the target features.

[0065] Optionally, embodiments of this application may dynamically set the fusion weights of acoustic fusion features and semantic fusion features based on the noise intensity of the scene noise.

[0066] Specifically, if the noise intensity is less than or equal to the first intensity, the weight of the acoustic fusion feature is set as the first target fusion weight, and the weight of the semantic fusion feature is set as the second target fusion weight. The first target fusion weight is less than the second target fusion weight, prioritizing highly reliable semantic information.

[0067] If the noise intensity is greater than the first intensity and less than or equal to the second intensity, the weights of the acoustic fusion features are set to be the same as those of the semantic fusion features to balance acoustic and semantic information.

[0068] If the noise intensity is greater than the second intensity, the weight of the acoustic fusion feature is set as the second target fusion weight, and the weight of the semantic fusion feature is set as the first target fusion weight, so as to prioritize the noise reduction processing of noise-sensitive acoustic information.

[0069] For example, in a noise level ≤ 45dB (quiet scene): the acoustic fusion feature weight is 0.3, and the semantic fusion feature weight is 0.7.

[0070] 45dB < Noise intensity ≤ 60dB (medium noise): weight 0.5:0.5.

[0071] Noise intensity > 60dB (noisy scene): weight 0.7:0.3.

[0072] Alternatively, calculate the average of each correlation degree in the first correlation degree set to obtain the mean of the first correlation degree.

[0073] Calculate the average value of each correlation degree in the second correlation degree set to obtain the mean value of the second correlation degree.

[0074] The fusion weight is determined based on the relationship between the mean of the first correlation degree and the mean of the second correlation degree.

[0075] The acoustic fusion features and semantic fusion features are weighted and fused based on the fusion weight to obtain the target features.

[0076] The fusion weights include the first fusion weight corresponding to the acoustic fusion features and the second fusion weight corresponding to the semantic fusion features.

[0077] The fusion weights are determined based on the relationship between the mean of the first correlation degree and the mean of the second correlation degree, including:

[0078] If the mean of the first correlation degree is greater than the mean of the second correlation degree, then: First fusion weight = mean of the first correlation degree / (mean of the first correlation degree + mean of the second correlation degree), Second fusion weight = mean of the second correlation degree / (mean of the first correlation degree + mean of the second correlation degree).

[0079] If the mean of the first correlation degree is less than or equal to the mean of the second correlation degree, then the first fusion weight is equal to the second fusion weight.

[0080] In this embodiment, the first correlation mean is calculated, which reflects the average Pearson coefficient of acoustic features and noise, indicating the overall degree of acoustic influence from noise. The second correlation mean reflects the average mutual information of semantic features and noise, indicating the overall degree of semantic influence from noise. Then, the fusion weights are determined according to the magnitude of the mean: if the acoustic mean is large, the weights are allocated according to the proportion of the mean, with a higher acoustic weight; conversely, if the acoustic mean is small, both the acoustic and semantic weights are 0.5. Finally, the acoustic and semantic features are weighted and fused to obtain the target feature.

[0081] This application's embodiments allow for dynamic adjustment of weights based on the impact of noise on features. In noisy scenes, high acoustic weights enhance noise reduction information, while balanced weights in quiet scenes ensure accurate feature resolution, aligning with the actual usage scenarios of smart glasses. Simultaneously, lightweight computation adapts to the low computing power constraints of smart glasses. Finally, it balances environmental noise information with user demand information, avoiding resource allocation bias caused by feature bias and providing accurate and balanced target feature support for subsequent CPU resource scheduling.

[0082] For example, the implementation process of a power management method may include:

[0083] S1, voice signal reception and preprocessing.

[0084] Voice reception: The microphone array of the smart glasses detects the voice signal of the target user and triggers the response mechanism.

[0085] Preprocessing: The original speech signal is pre-emphasized, framed and windowed, and initially suppressed to output clean speech frames.

[0086] S2: Acoustic and semantic feature extraction

[0087] S2.1: Extract the first feature that is the same in multiple dimensions.

[0088] Four 8-dimensional first features are selected, all of which are continuous value vectors:

[0089] First Feature 1 (MFCC): The first 8 dimensions of the 13-dimensional MFCC are taken (including the first-order difference).

[0090] First Feature 2 (STE): An 8-dimensional statistical vector of short-time energy per frame (mean, standard deviation, maximum value, etc.);

[0091] First feature 3 (ZCR): an 8-dimensional statistical vector of short-time zero-crossing rate for each frame;

[0092] First Feature 4 (Spectral Centroid): An 8-dimensional temporal vector of the spectral centroid for each frame.

[0093] S2.2: Extract the second feature that is identical in multiple dimensions.

[0094] Four 8-dimensional second features were selected, including continuous and discrete values ​​(the discrete values ​​are already encoded):

[0095] Second feature 1 (intent probability): an 8-dimensional probability vector of "query class / control class / call class" (e.g., "query subway station" corresponds to a query class probability of 0.92);

[0096] Second feature 2 (entity confidence): Extract the 8-dimensional confidence vector (confidence of attributes such as location and distance) of the "subway station" entity;

[0097] Second feature 3 (dialogue rounds): 8-dimensional encoding vector of "single round / multiple rounds" (single round interaction is encoded as [1, 0, ... 0]);

[0098] Second feature 4 (semantic similarity): an 8-dimensional similarity vector with the historical "search subway station" command.

[0099] S3: Feature weighted fusion based on scene noise.

[0100] S3.1: Acquire scene noise

[0101] Noise quantization: A microphone array acquires ambient noise in real time, and calculates the equivalent continuous sound pressure level (LAeq) and noise type.

[0102] Subway scenario: Noise intensity 70dB (>60dB, steady-state noise);

[0103] Bedroom scenario: Noise intensity 35dB (≤45dB, steady-state noise);

[0104] Output: Scene noise data (intensity + type, used for subsequent correlation calculation).

[0105] S3.2: Calculate the first set of association degrees

[0106] Correlation Calculation: For each 8-dimensional first feature, calculate its Pearson correlation coefficient with the "scene noise intensity (70dB / 35dB) time series", and take the absolute value as the first correlation.

[0107] The first correlation set for the subway scene: [0.85 (MFCC), 0.78 (STE), 0.72 (ZCR), 0.65 (spectral centroid)];

[0108] First correlation set for the bedroom scene: [0.32 (MFCC), 0.28 (STE), 0.25 (ZCR), 0.22 (spectral centroid)].

[0109] S3.3: Calculate the second set of association degrees.

[0110] Correlation calculation: For each 8-dimensional second feature, the discrete value is first encoded into a continuous value, then its mutual information with the "scene noise type (steady-state / non-steady-state)" is calculated, and the normalized result is used as the second correlation.

[0111] The second set of relevance scores for the subway scenario: [0.25 (intent probability), 0.18 (entity confidence), 0.15 (dialogue turns), 0.12 (semantic similarity)];

[0112] Second set of relevance scores for the bedroom scene: [0.42 (intention probability), 0.38 (entity confidence), 0.35 (dialogue turn), 0.32 (semantic similarity)].

[0113] S3.4: Feature-weighted fusion.

[0114] Acoustic fusion feature generation: The first feature weights are obtained by normalizing the first correlation set, and the four 8-dimensional first features are weighted and summed.

[0115] The first feature weight for the subway scene is [0.29, 0.27, 0.24, 0.20] (normalized from the first correlation set).

[0116] Weighted formula: Acoustic fusion feature = Σ(first feature i × first feature weight i), outputting one 8-dimensional acoustic fusion feature.

[0117] Semantic fusion feature generation: The second relevance set is normalized by "1-relevance" to obtain the second feature weights (reliability priority), and the four 8-dimensional second features are weighted and summed.

[0118] The weights of the second feature in the subway scenario are: [0.28, 0.30, 0.32, 0.34] (normalized from "1-second correlation set").

[0119] Weighted formula: Semantic fusion feature = Σ(second feature j × second feature weight j), outputting one 8-dimensional semantic fusion feature.

[0120] S3.5: Cross-feature weighted fusion

[0121] S3.5.1: Calculate the mean correlation degree

[0122] Subway scenario: Mean of first correlation

[0123] R1=(0.85+0.78+0.72+0.65) / 4=0.75;

[0124] Second correlation mean

[0125] R2=(0.25+0.18+0.15+0.12) / 4=0.18;

[0126] Bedroom Scene: Mean of First Relevance

[0127] R1=(0.32+0.28+0.25+0.22) / 4=0.27;

[0128] Second correlation mean

[0129] R2=(0.42+0.38+0.35+0.32) / 4=0.37.

[0130] S3.5.2: Determine the fusion weights

[0131] Subway scene:

[0132] R1>R2→First fusion weight w1=0.75 / (0.75+0.18)=0.81;

[0133] The second fusion weight w2 = 0.19.

[0134] Bedroom scene:

[0135] R1≤R2→First fusion weight w1=Second fusion weight w2=0.5.

[0136] S3.5.3: Generate target features

[0137] Target feature = acoustic fusion feature × w1 + semantic fusion feature × w2, outputting one 8-dimensional target feature (subway scene focuses on acoustics, bedroom scene is balanced).

[0138] S4: Resource Allocation Strategy Prediction

[0139] If the target characteristics belong to a pre-determined resource allocation model, output a resource allocation strategy.

[0140] Subway scenario (command interaction): Target feature input model → Uplink CPU utilization 48% (processing voice noise reduction + command parsing), downlink CPU utilization 19% (processing subway station information broadcasting).

[0141] Bedroom scenario (call interaction): Target feature input model → Uplink CPU utilization 27% (processing call voice encoding), downlink CPU utilization 35% (processing the other party's voice decoding).

[0142] S5: Power Management Based on Resource Allocation Strategy

[0143] Dynamically adjust CPU frequency based on resource allocation strategy (core power consumption control method):

[0144] Subway scenario: Uplink CPU utilization 48% (high load) → CPU frequency increased to 1.5GHz (to ensure noise reduction / resolution speed); Downlink CPU utilization 19% (low load) → Time-sharing multiplexing of cores, no additional frequency consumption; Power consumption per interaction cycle (5 seconds) ≈ 0.6mAh;

[0145] Bedroom scenario: Uplink / downlink CPU utilization is both <40% (medium load) → CPU frequency drops to 1.2GHz (balancing performance and power consumption); power consumption per call cycle (30 seconds) ≈3.2mAh;

[0146] During idle periods (without voice interaction): CPU frequency drops to 800MHz, and power consumption drops to 0.1mAh / second (compared to a fixed high frequency, battery life is extended by 1.8 times).

[0147] Compared to fixed uplink and downlink resources, the embodiments of this application can improve resource utilization, reduce resource waste, and improve the battery life of smart glasses.

[0148] Corresponding to the power management method in the above embodiments, Figure 2 This is a structural block diagram of a power management device provided according to an embodiment of this application. For ease of explanation, only the parts relevant to the embodiment of this application are shown. References Figure 2 The power management device 20 includes a feature extraction module 201, a feature fusion module 202, and a power management module 203.

[0149] The feature extraction module 201 is used to extract the acoustic and semantic features of the voice signal in response to receiving the voice signal of the target user, where the target user is a user wearing smart glasses.

[0150] The feature fusion module 202 is used to perform weighted fusion of acoustic features and semantic features based on the dialogue scenario of the target user to obtain the target features;

[0151] The power management module 203 is used to predict the resource allocation strategy of the smart glasses based on the target features, and to manage the power consumption of the smart glasses based on the resource allocation strategy. The resource allocation strategy includes an uplink resource allocation strategy and a downlink resource allocation strategy.

[0152] In one embodiment of this application, the acoustic features include multiple first features with the same dimensions, and the semantic features include multiple second features with the same dimensions, wherein each first feature and each second feature has the same dimensions.

[0153] The feature fusion module 202 is specifically used to acquire scene noise in the dialogue scenario of the target user; calculate the correlation between scene noise and each first feature to obtain a first correlation set; determine the weight of each first feature based on the first correlation set; and perform weighted fusion of each first feature based on the weight of each first feature to obtain acoustic fusion features; calculate the correlation between scene noise and each second feature to obtain a second correlation set; determine the weight of each second feature based on the second correlation set; and perform weighted fusion of each second feature based on the weight of each second feature to obtain semantic fusion features; and perform weighted fusion of acoustic fusion features and semantic fusion features to obtain target features.

[0154] In one embodiment of this application, the feature fusion module 202 is specifically used to calculate the Pearson correlation coefficient between each first feature and the scene noise as a first correlation degree; and to combine all the first correlation degrees into a first correlation degree set.

[0155] In one embodiment of this application, the feature fusion module 202 is specifically used to calculate the mutual information degree between the second feature and the scene noise for each second feature, as a second correlation degree; and combine all the second correlation degrees into a second correlation degree set.

[0156] In one embodiment of this application, the feature fusion module 202 is specifically used to calculate the average value of each correlation degree in the first correlation degree set to obtain the first correlation degree mean; calculate the average value of each correlation degree in the second correlation degree set to obtain the second correlation degree mean; determine the fusion weight based on the relationship between the first correlation degree mean and the second correlation degree mean; and perform weighted fusion of acoustic fusion features and semantic fusion features based on the fusion weight to obtain the target feature.

[0157] In one embodiment of this application, the fusion weights include a first fusion weight corresponding to the acoustic fusion features and a second fusion weight corresponding to the semantic fusion features.

[0158] The feature fusion module 202 is specifically used to: if the mean of the first correlation degree is greater than the mean of the second correlation degree, then: first fusion weight = mean of the first correlation degree / (mean of the first correlation degree + mean of the second correlation degree), second fusion weight = mean of the second correlation degree / (mean of the first correlation degree + mean of the second correlation degree); if the mean of the first correlation degree is less than or equal to the mean of the second correlation degree, then the first fusion weight is equal to the second fusion weight.

[0159] In one embodiment of this application, the power management module 203 is specifically used to input the target features into a predetermined resource allocation model to determine the resource allocation strategy of the smart glasses.

[0160] See Figure 3 , Figure 3 This is a schematic block diagram of an electronic device provided according to an embodiment of this application. Figure 3 The electronic device 300 in this embodiment may include one or more processors 301, one or more input devices 302, one or more output devices 303, and one or more memories 304. The processors 301, input devices 302, output devices 303, and memories 304 communicate with each other via a communication bus 305. The memories 304 store computer programs, including program instructions. The processors 301 execute the program instructions stored in the memories 304. Specifically, the processors 301 are configured to invoke the program instructions to perform the functions of the modules in the aforementioned device embodiments, for example... Figure 2 The functions of the feature extraction module 201, feature fusion module 202, and power management module 203 are shown.

[0161] It should be understood that, in the embodiments of this application, the processor 301 may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0162] Input device 302 may include a touchpad, a fingerprint sensor (for collecting the user's fingerprint information and fingerprint orientation information), a microphone, etc., and output device 303 may include a display (LCD, etc.), a speaker, etc.

[0163] The memory 304 may include read-only memory and random access memory, and provides instructions and data to the processor 301. A portion of the memory 304 may also include non-volatile random access memory.

[0164] In specific implementations, the processor 301, input device 302, and output device 303 described in the embodiments of this application can execute the implementation method described in the power management method provided in the embodiments of this application, or they can execute the implementation method of the electronic device described in the embodiments of this application, which will not be repeated here.

[0165] In another embodiment of this application, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which includes program instructions. When executed by a processor, the program instructions implement all or part of the processes in the methods described above. Alternatively, the computer program can instruct related hardware to complete the process. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include any entity or device capable of carrying computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0166] The computer-readable storage medium can be an internal storage unit of the electronic device in any of the foregoing embodiments, such as a hard disk or memory of the electronic device. The computer-readable storage medium can also be an external storage device of the electronic device, such as a plug-in hard disk, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the electronic device. Furthermore, the computer-readable storage medium can include both internal and external storage units of the electronic device. The computer-readable storage medium is used to store computer programs and other programs and data required by the electronic device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0167] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0168] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the electronic devices and units described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0169] In the several embodiments provided in this application, it should be understood that the disclosed electronic devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and other division methods may exist in actual implementation. Furthermore, the mutual couplings or direct couplings or communication connections shown or discussed can be indirect couplings or communication connections through some interfaces or units, or they can be electrical, mechanical, or other forms of connection.

[0170] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0171] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A power consumption management method, characterized in that, include: In response to receiving a voice signal from a target user, the acoustic and semantic features of the voice signal are extracted, wherein the target user is a user wearing smart glasses; The acoustic features and semantic features are weighted and fused based on the dialogue scenario of the target user to obtain the target features; Based on the target features, a resource allocation strategy for the smart glasses is predicted, and power consumption management of the smart glasses is performed based on the resource allocation strategy. The resource allocation strategy includes an uplink resource allocation strategy and a downlink resource allocation strategy. The acoustic features include multiple first features with the same dimensions, and the semantic features include multiple second features with the same dimensions, wherein each first feature and each second feature has the same dimensions. The target features are obtained by weighted fusion of the acoustic features and the semantic features based on the dialogue scenario in which the target user is located, including: Obtain the scene noise of the dialogue scenario in which the target user is located; The correlation degree between the scene noise and each first feature is calculated to obtain a first correlation degree set. The weight of each first feature is determined according to the first correlation degree set, and the first features are weighted and fused based on the weight of each first feature to obtain the acoustic fusion feature. The correlation degree between the scene noise and each second feature is calculated to obtain a second correlation degree set. The weight of each second feature is determined according to the second correlation degree set, and the second features are weighted and fused based on the weight of each second feature to obtain semantic fusion features. The acoustic fusion features and the semantic fusion features are weighted and fused to obtain the target features.

2. The power management method as described in claim 1, characterized in that, The calculation of the correlation degree between the scene noise and each first feature to obtain a first correlation degree set includes: For each first feature, calculate the Pearson correlation coefficient between the first feature and the scene noise, as the first correlation degree; Combine all the first degree of association into a first degree of association set.

3. The power management method as described in claim 1, characterized in that, The calculation of the correlation degree between the scene noise and each second feature yields a second correlation degree set, including: For each second feature, the mutual information between the second feature and the scene noise is calculated as the second correlation degree; Combine all the second degree of association into a second degree of association set.

4. The power management method as described in claim 1, characterized in that, The step of weightedly fusing the acoustic fusion features and the semantic fusion features to obtain the target features includes: Calculate the average value of each correlation degree in the first correlation degree set to obtain the first correlation degree mean; Calculate the average value of each correlation degree in the second correlation degree set to obtain the average value of the second correlation degree; The fusion weight is determined based on the relationship between the mean of the first correlation degree and the mean of the second correlation degree; The acoustic fusion feature and the semantic fusion feature are weighted and fused based on the fusion weight to obtain the target feature.

5. The power management method as described in claim 4, characterized in that, The fusion weights include a first fusion weight corresponding to the acoustic fusion features and a second fusion weight corresponding to the semantic fusion features; The step of determining the fusion weight based on the relationship between the mean of the first correlation degree and the mean of the second correlation degree includes: If the mean of the first correlation degree is greater than the mean of the second correlation degree, then: First fusion weight = mean of the first correlation degree / (mean of the first correlation degree + mean of the second correlation degree), second fusion weight = mean of the second correlation degree / (mean of the first correlation degree + mean of the second correlation degree). If the mean of the first correlation degree is less than or equal to the mean of the second correlation degree, then the first fusion weight is equal to the second fusion weight.

6. The power consumption management method according to any one of claims 1 to 5, characterized in that, The method of predicting the resource allocation strategy for the smart glasses based on the target features includes: The target features are input into a pre-determined resource allocation model to determine the resource allocation strategy for the smart glasses.

7. A power consumption management device, characterized in that, include: The feature extraction module is used to extract acoustic features and semantic features of the received voice signal from a target user, wherein the target user is a user wearing smart glasses; the acoustic features include multiple first features with the same dimensions, and the semantic features include multiple second features with the same dimensions, wherein each first feature and each second feature has the same dimension; The feature fusion module is used to perform weighted fusion of the acoustic features and the semantic features based on the dialogue scenario of the target user to obtain the target features; The feature fusion module is specifically used for: acquiring scene noise in the dialogue scenario of the target user; calculating the correlation between the scene noise and each first feature to obtain a first correlation set; determining the weight of each first feature based on the first correlation set; and performing weighted fusion of each first feature based on the weight of each first feature to obtain acoustic fusion features. The correlation degree between the scene noise and each second feature is calculated to obtain a second correlation degree set. The weight of each second feature is determined according to the second correlation degree set, and the second features are weighted and fused based on the weight of each second feature to obtain semantic fusion features. The acoustic fusion features and the semantic fusion features are weighted and fused to obtain the target features; A power management module is used to predict the resource allocation strategy of the smart glasses based on the target features, and to manage the power consumption of the smart glasses based on the resource allocation strategy, wherein the resource allocation strategy includes an uplink resource allocation strategy and a downlink resource allocation strategy.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1 to 6.

Images