Wireless microphone intelligent control system and method

Abstract

The invention relates to the technical field of audio processing, and discloses an intelligent control system and method for a wireless microphone, and the method comprises the steps: collecting multi-point acoustic data through a distributed microphone network, and constructing a spatial acoustic topology model; based on a topological analysis result, constructing an adaptive notch filter bank to realize howling suppression; environment changes are monitored in real time through a distributed acoustic sensor network, and suppression parameters are dynamically optimized; integrating a lightweight convolutional neural network on microphone equipment to identify acoustic events; low-delay acoustic event processing and howling prevention are realized by adopting a layered edge computing architecture; according to the invention, the technical problems of unstable howling suppression, weak environmental adaptability and lack of spatial perception in a complex environment are effectively solved.

Description

Technical Field

[0001] The present invention relates to the technical field of audio processing, and more specifically, it relates to a wireless microphone intelligent control system and method. Background Art

[0002] Wireless microphone systems are widely used in scenarios such as large conference centers, multi-functional halls, lecture halls, and commercial venues, providing flexible and convenient sound pickup solutions for speakers and performers.

[0003] However, in complex acoustic environments, traditional wireless microphone systems face three major technical challenges: The complex and variable acoustic feedback path makes it difficult to control the howling problem. In large spaces, sound can return from the speaker to the microphone through multiple paths, forming an acoustic feedback loop. Traditional systems are difficult to accurately model the interaction between these complex acoustic feedback paths and multi-point howling, and only use single-frequency point detection and suppression methods, resulting in unstable howling suppression effects and limited gain margins.

[0004] The dynamically changing spatial configuration causes frequent changes in acoustic characteristics. Factors such as wall movement, furniture rearrangement, and audience density change in multi-functional venues will significantly affect the characteristics of the acoustic environment. Traditional systems need to be frequently recalibrated and are difficult to adapt to these changes in real time. When the environment changes, it is necessary to re-identify the acoustic feedback path, resulting in a decrease in system usability.

[0005] Traditional systems lack the ability of multi-device collaboration and spatial context understanding. They mostly process signals at the single-device level, cannot effectively integrate the information of multiple sound pickup points, lack the perception of the overall acoustic space, and are difficult to accurately identify and locate acoustic anomaly events in complex environments. Their functions are limited to basic sound pickup.

[0006] The above technical problems limit the application effect and function expansion of wireless microphone systems in complex acoustic environments. There is a need for an intelligent wireless microphone control system that can accurately model complex acoustic feedback paths, quickly adapt to dynamic changes in the acoustic environment, and have the ability of multi-device collaborative perception to improve the stability, adaptability, and function diversity of the system. Summary of the Invention

[0007] The present invention provides a wireless microphone intelligent control system and method to solve the technical problems of unstable howling suppression, weak environmental adaptability, and lack of spatial perception ability of wireless microphone systems in complex acoustic environments.

[0008] The present invention provides a wireless microphone intelligent control method, including the following steps: Collect the first type of multi-point acoustic data through multiple wireless microphones and fixed microphones distributed in space, construct a spatial acoustic transfer function matrix, analyze the characteristics of the acoustic feedback path using singular value decomposition, and form a spatial acoustic topology model; Based on the results of spatial acoustic topology analysis, construct and optimize an adaptive notch filter bank to achieve precise howling suppression; Collect the second type of acoustic data in real time through a distributed acoustic sensor network, construct a three-dimensional sound field distribution, detect changes in the spatial acoustic environment, and generate optimized suppression parameters for the changing area; Integrate a lightweight convolutional neural network on each microphone device, combine multi-microphone information fusion technology, and achieve the recognition and localization of acoustic anomaly events; Adopt a hierarchical edge computing architecture and an energy-saving communication strategy to process acoustic data in layers, and achieve low-latency acoustic event processing and howling prevention.

[0009] In a preferred embodiment, the steps of constructing the spatial acoustic transfer function matrix include: In the system initialization stage, output excitation signals through each speaker in sequence; All microphones simultaneously record the received signals; For each pair of speaker-microphone combinations, calculate the transfer function to form a spatial acoustic transfer function matrix.

[0010] In a preferred embodiment, the steps of constructing the adaptive notch filter bank include: According to the potential howling frequency points identified by acoustic topology analysis and their risk indices, construct a second-order notch filter unit for each potential howling frequency point; Optimize the attenuation factor of each notch filter according to the howling risk index; Construct a multi-stage linked notch filtering structure to handle the mutual influence between howling frequency points; Achieve predictive howling suppression and suppress howling before it is fully formed.

[0011] In a preferred embodiment, the steps of constructing the distributed acoustic sensor network include: Use the fixed microphone on the charging stand as the reference node and the wireless microphone as the dynamic node; Each wireless microphone is equipped with an inertial measurement unit to sense its own motion state; Estimate the relative position between the wireless microphone and the reference node based on time of flight or received signal strength indication; Realize data transmission and synchronization between nodes through a wireless communication network.

[0012] In a preferred embodiment, the step of detecting changes in the spatial acoustic environment includes: Periodically update the spatial acoustic transfer function matrix and calculate the change rate between the spatial acoustic transfer function matrix and the matrix at the previous moment; Use singular value decomposition to track the change direction of acoustic characteristics; Based on the components of the singular vectors, determine the spatial regions where the acoustic environment changes; When the change rate exceeds a preset threshold, trigger the adaptive update mechanism.

[0013] In a preferred embodiment, the structure of the lightweight convolutional neural network includes: An input layer for receiving feature vectors; Two convolutional layers and pooling layers for extracting acoustic features; Two fully connected layers for classifying acoustic events; Adopt weight pruning technology and fixed-point quantization representation to reduce the parameter scale, and control the overall model size within 500KB.

[0014] In a preferred embodiment, the multi-microphone information fusion technology includes: Based on the time difference and signal intensity ratio, calculate the spatial position of the sound source; Based on the spatio-temporal consistency principle, associate the same type of events identified by multiple devices; Use a confidence-weighted voting mechanism to fuse the event classification results of multiple devices; Construct an acoustic event spatial heat map and identify the high-activity regions as the focus of potential abnormal events.

[0015] In a preferred embodiment, the hierarchical edge computing architecture includes: The device layer, composed of low-power microprocessors built into wireless microphones, is responsible for signal acquisition and preliminary feature extraction; The edge aggregation layer, composed of edge servers integrated in the charging dock, is responsible for multi-device data integration and model update; The cloud layer is used for long-term data storage and model optimization; A hierarchical message queue mechanism and an adaptive task migration algorithm are used for communication and resource optimization between layers.

[0016] In a preferred embodiment, the energy-saving communication strategy includes: Adopt an event-driven communication mechanism and trigger data upload only when potential events are detected; Dynamically adjust the data acquisition frequency according to the environmental complexity; Implement data compression transmission and only transmit effective features instead of raw audio data; Establish an optimized wireless communication protocol between the wireless microphone and the charging dock, supporting multi-device synchronization and low-power transmission.

[0017] In a preferred embodiment, a wireless microphone intelligent control system for implementing a wireless microphone intelligent control method includes: Multiple wireless microphone modules, each equipped with a microphone array, a signal processing unit, and a wireless communication module; The charging dock module, integrating a fixed microphone array and an edge computing server; The spatial acoustic topology analysis module for constructing and analyzing the spatial acoustic transfer function matrix; The adaptive howling suppression module for achieving precise howling control based on topology analysis; The distributed acoustic event detection module for achieving the recognition and localization of acoustic anomaly events with multi-device collaboration.

[0018] The beneficial effects of the present invention are as follows: By constructing a spatial acoustic topology model and an adaptive notch filter bank, the present invention effectively solves the howling problem in complex acoustic environments and achieves precise howling prediction and suppression; The present invention adopts the technology of adaptive acoustic characteristics in different zones, enhancing the environmental adaptability of the system; Through the distributed collaborative acoustic event detection technology, the present invention expands the function of the wireless microphone system from a simple human-computer interaction tool to an environmental safety monitoring platform. In a typical office or meeting environment, the system improves the classification accuracy of acoustic events and the positioning accuracy of events, can automatically record the audio evidence of abnormal acoustic events, and provide precise location information.

[0019] The present invention adopts a hierarchical edge computing architecture to achieve low-power and low-latency distributed intelligent processing. The system reasonably distributes tasks among different levels, improving the battery life of a single wireless microphone, while reducing the howling detection delay and acoustic event detection delay to meet the real-time interaction requirements.

[0020] The present invention also realizes device collaboration and system scalability. Multiple microphone devices form a collaborative working network. As the number of devices increases, the acoustic space perception ability and event positioning accuracy of the system continue to improve, and it supports the dynamic addition or removal of devices without a complex reconfiguration process. Brief Description of the Drawings

[0021] Figure 1 is a flowchart of a wireless microphone intelligent control method of the present invention. Detailed Embodiments

[0022] Reference will now be made to example embodiments to discuss the subject matter described herein. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein, and that changes can be made to the functions and arrangements of the elements discussed without departing from the scope of protection of the content of this specification. Each example may omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples may be combined in other examples.

[0023] In at least one embodiment of the present invention, a method for intelligent control of a wireless microphone is disclosed. As Figure 1 shown, it includes the following steps: Step 1, collecting first - type multi - point acoustic data through multiple wireless microphones and fixed microphones distributed in space, constructing a spatial acoustic transfer function matrix, and using singular value decomposition to analyze the characteristics of the acoustic feedback path to form a spatial acoustic topology model; The specific implementation includes the following sub - steps: Step 1.1, multi - point acoustic data collection; Using a wireless microphone array distributed in space and a condenser microphone on the charging base, acoustic signals are collected simultaneously at multiple key positions in space.

[0024] The acquisition system includes: a number of wireless microphones, each microphone equipped with a high - precision ADC (analog - to - digital converter) with a sampling rate of 48 kHz and a quantization accuracy of 24 bits, for capturing sound signals at different positions in space; a number of condenser microphones integrated on the charging base, with the same sampling rate as the wireless microphones, for providing acoustic data at a fixed reference point; The data recorded at each acquisition point includes the time - domain signal and the corresponding frequency - domain representation , where represents the index number of the acquisition point, and the value range is .

[0025] Step 1.2, construction of the acoustic transfer function matrix; Based on the multi - point acquired acoustic data, a spatial acoustic transfer function matrix is constructed.

[0026] This matrix describes the acoustic propagation characteristics between any two points in space, and its mathematical expression is: ; where, represents the acoustic transfer function from point to point , is a complex variable in the frequency domain for the complex variable representation, and are the indices of the signal source and the receiving point respectively, represents the total number of wireless microphones in the system, represents the total number of fixed microphones on the charging stand, and the dimension of the entire matrix is , which contains the acoustic transfer relationships between all microphones and speakers.

[0027] The transfer function is estimated by the following method: In the system initialization stage, an excitation signal (such as white noise or swept-frequency signal) is output through each speaker in turn; At the same time, the signals received by all microphones are recorded; For each pair of speaker-microphone combinations, calculate the transfer function: ; where, represents the acoustic transfer function from speaker to microphone , which describes the characteristics of the sound signal propagating from the speaker to the microphone in space; represents the representation of the excitation signal output by speaker in the frequency domain; represents the representation of the response signal received by microphone in the frequency domain; is a complex variable, representing a complex variable in the frequency domain, used to describe the frequency characteristics of the signal; is the index number of the speaker, representing a specific speaker position; is the index number of the microphone, representing a specific microphone position.

[0028] Step 1.3, singular value decomposition to analyze the characteristics of the acoustic feedback path; Perform singular value decomposition (Singular Value Decomposition, SVD) on the constructed acoustic transfer function matrix to analyze the main characteristics of the acoustic feedback path: ; where, is the acoustic transfer function matrix, which describes the acoustic propagation characteristics between points in space; is the left singular vector matrix, and its column vectors represent the main acoustic modes in the space received by the microphone; is the right singular vector matrix, and its column vectors represent the main spatial modes of the sound emitted by the speaker; is the singular value diagonal matrix, and the elements on the diagonal represent the energy magnitudes of the corresponding acoustic propagation paths. Denotes the conjugate transpose of the matrix , that is, for each complex element in , take the conjugate and then transpose the matrix; Singular value decomposition decomposes the complex acoustic transfer function matrix into orthogonal acoustic modes, facilitating the identification of the main acoustic feedback paths.

[0029] By analyzing: The magnitude of the singular value reflects the energy contribution of different acoustic feedback paths; The singular vector corresponding to the largest singular value represents the spatial distribution of the main acoustic feedback path; The distribution characteristics of the singular values reveal the complexity of the acoustic environment.

[0030] Step 1.4, extraction of key features of the feedback path; Based on the singular value decomposition results, extract the key features of the acoustic feedback path: Extract the acoustic feedback paths corresponding to the first largest singular values, where the value is adaptively determined according to the cumulative contribution rate, taking the smallest value when the cumulative contribution reaches 90%; For each main path, extract its frequency response characteristics, focusing on analyzing its amplitude-frequency characteristics and phase-frequency characteristics; Identify potential howling frequency points, that is, the frequency points where the transfer function gain is close to or exceeds the unit gain and the phase is close to an integer multiple of 2π; Calculate the howling risk index for each potential howling frequency point: ; Where represents the howling risk index of the th frequency point, and the larger the value, the higher the probability of howling at this frequency point; represents the gain value of the th frequency point, reflecting the amplification factor of the acoustic transfer function at this frequency point; represents the phase value of the th frequency point, describing the phase characteristics of the acoustic signal at this frequency point; represents normalizing the phase value to within the period; represents rounding the normalized phase value to find the closest integer value; represents the deviation degree between the phase and the closest integer multiple of ; represents the degree of closeness of the phase to an integer multiple of , and the closer this value is to 1, the easier it is to form howling.

[0031] The output spatial acoustic topology model contains the following core information: Complete acoustic transfer function matrix ; Characteristic description of the main acoustic feedback path; List of potential howling frequency points and their risk indices; Spatial distribution characteristics of the acoustic feedback path.

[0032] This model provides a basis for subsequent adaptive howling suppression and detection of changes in the spatial acoustic environment.

[0033] Step 2: Based on the results of spatial acoustic topology analysis, construct and optimize an adaptive notch filter bank to achieve precise howling suppression; The specific implementation includes the following sub-steps: Step 2.1: Construction of the adaptive notch filter bank; Based on the potential howling frequency points and their risk indices identified in Step 1.4, construct an adaptive notch filter bank. The transfer function of the filter bank is expressed as: ; Where, is the total transfer function of the adaptive notch filter bank, representing the processing response of the system to the input signal; represents the cascaded multiplication of all notch filter units, combining the effects of multiple filters; is the number of potential howling frequency points to be suppressed, determining the number of notch filter units included in the filter bank; is the normalized angular frequency of the th potential howling frequency point, , is the howling frequency, is the sampling frequency; is the unit delay operator, representing a signal delay of one sampling period; is the double unit delay operator, representing a signal delay of two sampling periods; determines the center frequency position of the notch filter, i.e., the target frequency of howling suppression; is the attenuation factor controlling the bandwidth of the notch filter, , The closer it is to 1, the narrower the bandwidth of the notch filter and the smaller the impact on the useful signal near the target frequency; represents a zero at the frequency, completely eliminating the signal at that frequency; represents a pole close to the frequency, controlling the bandwidth and depth of the notch; each potential howling frequency point corresponds to a second-order notch filter unit, and the second-order structure can precisely control the suppression effect at a specific frequency.

[0034] Step 2.2: Optimization of filter parameters based on risk indices; According to the howling risk index , optimize the parameters of each notch filter: The howling risk index is higher, and the corresponding attenuation factor is set closer to 1 (narrowband notch) to minimize the impact on the useful signal; For frequency points where the risk index is lower than the threshold , notch filtering should not be applied to avoid unnecessary signal distortion, where: represents the preset risk index threshold of the system. Frequency points below this threshold are considered to have a lower howling probability and do not require suppression processing; The attenuation factor and the risk index are related as follows: ; where, represents the attenuation factor of the th notch filter, which determines the bandwidth characteristics of the filter; represents the preset minimum attenuation factor, corresponding to a wider filter bandwidth, for frequency points with lower risk; represents the preset maximum attenuation factor, corresponding to a narrower filter bandwidth, for frequency points with higher risk; represents the current howling risk index of the th frequency point; represents the maximum risk index value among all frequency points detected by the system; represents normalizing the risk index to the [0, 1] interval for linearly mapping the attenuation factor.

[0035] Step 2.3, implementation of a multi-stage cascaded notch filter structure; Construct a multi-stage cascaded notch filter structure to handle the mutual influence between howling frequency points: Divide the filter bank into multiple cascaded sub-filter banks, with each group responsible for handling howling in a specific frequency band; Establish a linkage mechanism between notch filters: when the howling of one frequency point is suppressed, it may cause changes in the acoustic characteristics of other frequency points, and the system updates the risk index of each frequency point in real time; Implement a gradual adjustment of the filter parameters to avoid audio distortion caused by parameter mutations.

[0036] Step 2.4, predictive howling suppression algorithm; Implement predictive howling suppression to suppress howling before it is fully formed: Calculate the howling development trend index based on the time dynamic change of the spatial acoustic topology model: ; Among them, represents the howling development trend index, which is used to quantify the change rate of the howling risk; represents the derivative operation with respect to time, which is used to calculate the change rate of the risk index over time; represents the howling risk index of the th frequency point, which reflects the likelihood of howling at this frequency point; when is positive, it indicates that the howling risk is increasing, and the larger the value, the faster the howling formation speed; when is negative, it indicates that the howling risk is decreasing, and the system may be naturally stabilizing; this index is used for predictive howling control, enabling the system to take suppression measures before howling is fully formed; For frequency points greater than the preset threshold, activate the corresponding notch filter in advance; Adopt an increasing suppression intensity, that is, as increases, gradually increase the depth of the notch.

[0037] The filter can be implemented using an efficient IIR (Infinite Impulse Response) structure, and its difference equation is expressed as: ; Among them, represents the filter output signal at the current time ; represents the input signal at the current time ; represents the input signal at the previous time, that is, the input delayed by one sampling period; represents the input signal at the two previous times, that is, the input delayed by two sampling periods; represents the output signal at the previous time, that is, the output delayed by one sampling period; represents the output signal at the two previous times, that is, the output delayed by two sampling periods; , , respectively represent the feedforward coefficients of the filter at the current time, the previous time, and the two previous times, which are used to process the current and historical input signals; , respectively represent the feedback coefficients of the filter at the previous time and the two previous times, which are used to process the historical output signals; Each coefficient corresponds to the filter transfer function parameter: : The weight coefficient of the current input signal; : The weight coefficient of the input signal at the previous moment, where is the normalized angular frequency of the target howling frequency point; : The weight coefficients of the input signals at the previous two moments; : The weight coefficient of the output signal at the previous moment, where is the attenuation factor for controlling the bandwidth of the notch filter; : The weight coefficients of the output signals at the previous two moments, which are determined by the square of the attenuation factor .

[0038] Finally, output the signal processed by adaptive howling suppression filtering, which has the minimum howling risk while maintaining the sound quality and clarity of the original signal.

[0039] Step 3: Real-time collect the second type of acoustic data through the distributed acoustic sensor network, construct the three-dimensional sound field distribution, detect the changes in the spatial acoustic environment, and generate optimized suppression parameters for the changing area; The specific implementation includes the following sub-steps: Step 3.1: Construction of the distributed acoustic sensor network; Construct a distributed acoustic sensor network composed of wireless microphones and fixed microphones: Use M fixed microphones on the charging stand as reference nodes, and the positions of these nodes are fixed and known; Use N wireless microphones as dynamic nodes, and communicate with the reference nodes through a wireless communication network (such as Bluetooth Low Energy 5.0 or a proprietary wireless protocol); Each wireless microphone is equipped with an IMU (Inertial Measurement Unit), which includes a three-axis accelerometer and a three-axis gyroscope for sensing its own motion state; Estimate the relative position between the wireless microphone and the reference node based on TOF (Time of Flight) or RSSI (Received Signal Strength Indication).

[0040] The specific implementation of this distributed acoustic sensor network includes: Sensor node type and configuration: Fixed reference node: Each charging stand integrates 4 to 8 omnidirectional condenser microphones, with a sampling rate of 48 kHz and a signal-to-noise ratio > 70 dB, and a cost-optimized configuration; Dynamic mobile node: Each wireless microphone integrates 2 to 4 directional condenser microphones, with a sampling rate of 48 kHz and a signal-to-noise ratio > 65 dB, and a power consumption-optimized configuration; Each node is equipped with an independent signal processing unit, which includes a low-power DSP and a storage buffer, and can independently complete preliminary data processing.

[0041] Network topology and communication protocol: Adopt a star network topology, with the charging base as the central node and all wireless microphones as sub-nodes; Use the improved Bluetooth Low Energy 5.0 protocol, which supports broadcast, multicast, and point-to-point communication modes; Design a dynamic time slot allocation algorithm to dynamically adjust the communication bandwidth according to the transmission priority; Implement a communication error detection and automatic error correction mechanism to ensure communication reliability in a complex electromagnetic environment.

[0042] Node synchronization and positioning method: Periodically send synchronization signals, and each node calculates and corrects the clock deviation based on the received timestamp; Integrate TOF, RSSI, and signal angle estimation, and use the trilateration principle to determine the spatial position of the wireless microphone; Use IMU data for trajectory tracking and maintain the positioning accuracy through inertial navigation during communication interruption; Apply the Kalman filter algorithm to fuse multi-source positioning data to improve the positioning stability and accuracy.

[0043] This distributed acoustic sensor network exhibits the following characteristics in practical applications: In a typical 100-square-meter conference room environment, the positioning accuracy can reach ±0.3 meters; Supports up to 16 wireless microphones to access simultaneously and can be flexibly expanded; The position information update rate of each node is 10Hz, meeting the real-time tracking requirements; The network synchronization accuracy is better than 1 millisecond, ensuring the time consistency of multi-point acoustic data; The power consumption of a single wireless microphone node is less than 50mW (standby state) and 150mW (active state).

[0044] Step 3.2, Acoustic holography reconstructs the three-dimensional sound field distribution; Utilize the multi-point acoustic data collected by the distributed sensor network and apply acoustic holography technology to reconstruct the three-dimensional spatial sound field distribution: Adopt the spatial Fourier transform algorithm to convert the discrete acoustic sampling point data into a continuous sound field distribution; The mathematical model of sound field reconstruction is expressed as: ; Among them, represents the sound pressure distribution at point , is the position vector of any point in space; is from the source point to the point The Green's function represents the propagation of sound waves from the source point to the receiving point and its transfer characteristics; represents the normal vibration velocity at the source point, that is, the vibration velocity in the direction perpendicular to the surface of the sound source; is the surface of the sound source, representing the integration region; is an infinitesimal area element on the surface of the sound source; the entire integral represents the superimposed contribution of all points on the sound source surface to the sound pressure at any point in space; For discrete sampling points, the above equation can be approximated as: ; where represents the sound pressure distribution at point ; represents the number of wireless microphones; represents the number of fixed microphones; represents the microphone index, ranging from 1 to the total number of all microphones; is the acoustic holographic weight coefficient, used to adjust the contribution weights of different microphone sampling points; is the Green's function from the microphone position to any point in space, describing the sound wave propagation characteristics; is the measured sound pressure value at the sampling point , that is, the sound signal intensity collected by the microphone ;

[0045] Step 3.3, real-time detection of acoustic environment changes; Based on continuously collected acoustic data, real-time detection of changes in the spatial acoustic environment: Periodically update the spatial acoustic transfer function matrix and calculate its rate of change with the matrix at the previous moment: ; where represents the Frobenius norm of the matrix, used to calculate the square root of the sum of the squares of the matrix elements, measuring the overall difference between two matrices; represents the spatial acoustic transfer function matrix at the current moment ; represents the spatial acoustic transfer function matrix at the previous moment ; is the sampling time interval, that is, the time difference between two acoustic characteristic measurements; represents the quantization index of the acoustic environment change at the current moment, and the larger the value, the more significant the environmental change; Use singular value decomposition to track the main change directions of acoustic characteristics: ; Among them, represents the spatial acoustic transfer function matrix at the current moment ; represents the spatial acoustic transfer function matrix at the previous moment ; is an orthogonal matrix, and its column vectors represent the spatial distribution characteristics of the acoustic environment change; is a diagonal matrix, and the elements on the diagonal are singular values, representing the intensity of each main change direction; is 's conjugate transpose matrix, and its row vectors represent the frequency distribution characteristics of the acoustic environment change. By analyzing 's main column vectors and 's larger singular values, the spatial region and frequency range with the most significant acoustic environment change can be determined.

[0046] Based on and 's main components, determine the spatial region of the acoustic environment change; When the change rate exceeds the preset threshold , trigger the adaptive update mechanism.

[0047] Step 3.4, regionalized adaptive parameter optimization; For the detected acoustic change region, dynamically generate optimized suppression parameters: Divide the space into acoustic regions, and the acoustic characteristics within each region are approximately the same; For each region , calculate its acoustic characteristic change index: ; Among them, represents the region index number, and the value range is from 1 to ; represents a quantitative index of the acoustic characteristic change degree of this region, and the larger the value, the more significant the change; represents the summation of the transfer functions of all microphone-speaker pairs within region ; represents a pair of combinations composed of microphone and speaker ; represents the set of all microphone-speaker pairs included in region ; represents the microphone and speaker at the current moment the acoustic transfer function between; representing the previous moment of the same transfer function; representing the absolute difference between the transfer functions at two moments, used to quantify the degree of change in the acoustic characteristics of this specific microphone-speaker pair; According to the change index , different adaptive parameters are assigned to each region: High-change region: Update the acoustic topology model more frequently and use a more aggressive howling prediction threshold; Low-change region: Reduce the update frequency, maintain the existing suppression parameters, and reduce the computational burden; Introduce a smooth transition mechanism to prevent parameter mutations at the region boundaries.

[0048] This partitioned adaptive system can automatically adjust the system parameters without recalibration when the spatial acoustic conditions change (such as personnel movement, door and window opening and closing, space reconstruction, etc.), maintain efficient howling suppression performance, and at the same time significantly reduce the consumption of computing resources. Through this regional processing, the system can focus on the parts of the acoustic environment that actually change, improving the response speed and algorithm efficiency.

[0049] Step 4, Integrate a lightweight convolutional neural network on each microphone device, combine multi-microphone information fusion technology, and achieve the recognition and localization of acoustic anomaly events; The specific implementation includes the following sub-steps: Step 4.1, Single-device lightweight acoustic event detection; Integrate a lightweight acoustic event detector in each wireless microphone device: Preprocess the audio signal, including frame segmentation, windowing, feature extraction, etc. The length of each frame is 20 ms, and the frame shift is 10 ms; Extract acoustic features such as Mel-Frequency Cepstral Coefficients (MFCC), spectral centroid, spectral flux, zero-crossing rate, etc. to form a feature vector ; Use a lightweight convolutional neural network to analyze the feature vector. The network structure is: Input layer: Receive the feature vector , with a dimension of 128× , where is the number of features; Convolutional layer 1: 32 convolution kernels, ReLU activation function, stride of 1, padding of 1, and the output feature map dimension of 128× ×32; Pooling layer 1: Max pooling with a stride of 2, reducing the output feature map dimension by half to 64×( / 2)×32; Convolutional layer 2: 64 convolution kernels, ReLU activation function, stride of 1, padding of 1, and the output feature map dimension is 64×( / 2)×64; Pooling layer 2: Max pooling with a stride of 2, reducing the output feature map dimension by half to 32×( / 4)×64; Fully connected layer 1: 128 neurons, ReLU activation; Fully connected layer 2: Output layer, neurons ( is the number of acoustic event categories to be detected), Softmax activation; The training method of this lightweight convolutional neural network: Use a supervised learning method with an annotated acoustic event dataset; Adopt a cross-entropy loss function to evaluate the difference between the model prediction and the true label; Use the Adam optimizer to adjust the network parameters, set the initial learning rate to 0.001, and reduce the learning rate to 0.1 of the original every 50 epochs; Use the Dropout technique (probability of 0.5) to prevent overfitting; Use batch normalization technology to improve the training stability and convergence speed; Model compression and optimization method: Use the weight pruning technique to remove connections with little contribution to the output, achieving a compression rate of 40%; Adopt 8-bit fixed-point quantization representation to convert floating-point weights to integers and reduce the storage requirements; Use the knowledge distillation technique to guide the lightweight model learning with a large pre-trained model; In specific applications, selectively activate specific classifiers according to the scene characteristics to reduce the computational load.

[0050] The network output can be expressed as: ; Among them, represents the input acoustic feature vector; represents the processing result of the convolutional neural network on the input features; represents the weight matrix of the first fully connected layer; represents the bias vector of the first fully connected layer; represents the rectified linear unit activation function, which is used to introduce non-linearity; Represents the weight matrix of the second fully connected layer (output layer); Represents the bias vector of the second fully connected layer; Represents the softmax function, which converts the output into a probability distribution; Represents the final classification result, that is, the probability distribution of each acoustic event category.

[0051] The 15 acoustic event categories that the system can identify include: human speech; coughing; applause; laughter; knocking; keyboard tapping; door opening / closing; chair movement; paper flipping; glass breaking; mobile phone ringing; alarm sound; electrical appliance humming; abnormal screaming; background music. These categories cover both normal acoustic events commonly seen in scenarios such as meeting rooms, classrooms, and offices, as well as abnormal acoustic events that require special attention.

[0052] The network uses quantization technology to reduce the parameter scale, and the overall model size is controlled within 500KB, making it suitable for deployment on resource-constrained microphone devices.

[0053] When this lightweight convolutional neural network is applied in a large meeting room environment, it can real-time detect 10 common acoustic events including speech, applause, knocking, door opening / closing, glass breaking, etc., with an average detection accuracy of 92% and a single detection latency of less than 50 milliseconds. At the same time, through the incremental learning method, the model can gradually adapt to and learn new acoustic event types in specific scenarios, showing strong adaptability.

[0054] Step 4.2, multi-microphone sound source localization; Based on the time delay and signal strength difference, calculate the spatial position of the sound source: Calculate the time difference of the sound arriving at different microphones: ; Among them, represents the time difference between the sound arriving at microphone and microphone ; represents the time point when the sound arrives at microphone ; represents the time point when the sound arrives at microphone ; When is positive, it means the sound arrives at microphone first and then arrives at microphone ; When is negative, it means the sound arrives at microphone first and then arrives at microphone ; This time difference is a key input parameter for sound source localization calculation; Calculate the sound source position using the hyperbolic positioning method based on the known microphone position information and time difference: ; Wherein, is the sound source position vector, representing the coordinates of the to-be-determined sound source in three-dimensional space; and are the position vectors of microphones and respectively, representing the known coordinates of the two microphones in three-dimensional space; is the speed of sound, with a value of approximately 340 m / s (in air at standard atmospheric pressure and a temperature of 20 °C); is the time difference for the sound to reach microphones and ; represents the Euclidean distance from the sound source to microphone , that is, the length of the sound wave propagation path; this equation describes the mathematical relationship that the sound source is located on a hyperbola (a hyperboloid in three-dimensional space) with two microphones as foci; Combine the time difference and intensity ratio information to construct a weighted positioning model to improve the positioning accuracy: ; Wherein, is the estimated sound source position vector, representing the coordinates of the finally calculated sound source in three-dimensional space; represents finding the value that minimizes the following expression; represents summing over all microphone pairs ; is the weight coefficient of the positioning model, related to the signal quality. The better the signal quality, the greater the weight; is the sound source position vector, which is a variable in the optimization process; and are the position vectors of microphones and respectively; represents the Euclidean distance from the sound source to microphone ; is the speed of sound, usually approximately 340 m / s; is the time difference for the sound to reach microphones and ; The whole formula represents determining the optimal sound source position by minimizing the sum of the squares of the time difference estimation errors for all microphone pairs.

[0055] Step 4.3, multi-device information fusion and collaborative detection; Integrate the detection results of multiple microphone devices to improve the accuracy of acoustic event recognition: Based on the principle of spatio-temporal consistency, associate the same type of events identified by multiple devices. If the difference in event timestamps is less than the threshold and the difference in spatial positions is less than the threshold , then they are regarded as the same event; Use a confidence-weighted voting mechanism to fuse the event classification results of multiple devices: ; Among them, is the final event category, representing the acoustic event type determined after the system synthesizes the detection results of multiple microphone devices; represents finding the category that maximizes the following expression ; represents for all the contributions of microphone devices are accumulated; represents the number of the microphone device, ranging from 1 to ; is the device for the category the predicted probability, indicating the confidence that the th microphone device judges that the current acoustic event belongs to the category , with a value range of [0, 1]; is the device weight, reflecting the reliability and importance of the microphone device in the overall system, usually dynamically adjusted based on factors such as the signal quality and historical accuracy of the device; this formula describes a weighted voting mechanism that selects the category with the largest weighted probability sum as the final judgment result by comprehensively considering the detection results of multiple microphone devices; For low-confidence regions, increase the weights of neighboring devices to improve the overall judgment accuracy.

[0056] Step 4.4, construction of the acoustic event spatial heat map; Based on the collaborative detection results of multiple devices, construct an acoustic event spatial heat map: Divide the space into three-dimensional grids, and each grid point represents the acoustic event heat at a certain position in space at time ; The heat calculation formula is: ; Among them, represents the acoustic event heat value at the coordinate point in space at time ; represents for all the contributions of microphone devices are accumulated; represents the number of the microphone, ranging from 1 to ; represents the weight coefficient of the microphone device which reflects the reliability and importance of the device in the overall system; represents a spatial point to the microphone Euclidean distance; represents the signal strength detected at the microphone location; represents the distance attenuation function, which describes how the sound intensity attenuates with the increase of distance while considering the influence of the original signal strength ; this function is usually a non-linear attenuation model, reflecting the physical characteristics of sound wave propagation in space; Adopt a time decay factor to gradually weaken the influence of historical events over time: ; where represents the acoustic event heat value at the spatial coordinate at the time point ; is the time decay coefficient, which controls the rate of weakening of the influence of historical events. The larger the value, the faster the influence of historical events decays; is the current system time; represents the time difference from the event occurrence time to the current time ; is an exponential decay function, which rapidly decreases in value as the time difference increases, making the influence of earlier events on the current heat map gradually weaken; the entire formula describes how to dynamically update the acoustic event heat map according to time movement, ensuring that the heat map can reflect the latest acoustic activity status; Use heat map threshold segmentation to identify high-activity areas as the focus of potential abnormal events.

[0057] When this distributed collaborative detection system detects high-confidence abnormal acoustic events (such as glass breaking sounds, abnormal screams, etc.), it can automatically record the audio evidence of the event and provide accurate location information, expanding the function of the wireless microphone system from a simple human-computer interaction tool to an environmental safety monitoring platform.

[0058] Step 5: Adopt a hierarchical edge computing architecture and an energy-saving communication strategy to perform hierarchical processing on acoustic data, achieving low-latency acoustic event processing and howling prevention; The specific implementation includes the following sub-steps: Step 5.1: Construction of the hierarchical computing architecture; Establish a three-layer edge computing architecture and reasonably allocate computing tasks: The first layer (device layer): The low-power microprocessor built into the wireless microphone is mainly responsible for signal acquisition, preliminary feature extraction, and event candidate detection; The second layer (edge aggregation layer): The edge server integrated in the charging dock, with medium computing power, is responsible for multi-device data integration, acoustic topology model update, and collaborative event verification; The third layer (cloud layer): An optional cloud server for long-term data storage, model optimization, and complex algorithm training.

[0059] The specific implementation of this hierarchical edge computing architecture includes: Hardware platform configuration: Device layer: Each wireless microphone integrates an ARM Cortex-M4F or equivalent performance microcontroller, with a main frequency of 80 MHz, 256 KB of on-chip RAM, and 1 MB of flash memory; Edge aggregation layer: The charging dock integrates a high-performance SoC processor, such as an ARM Cortex-A53 quad-core, with a main frequency of 1.5 GHz, 2 GB of RAM, and 16 GB of eMMC storage; Cloud layer: Standard cloud server instances, with computing resources configured on demand.

[0060] Software architecture implementation: The device layer uses a real-time operating system (Real-Time Operating System, RTOS), such as FreeRTOS, to ensure the time determinacy of critical tasks - the edge aggregation layer runs a lightweight Linux system, providing flexible task scheduling and network connection capabilities; Inter-layer communication uses a hierarchical message queue mechanism, and high-priority messages can interrupt low-priority transmissions; The data stream adopts a pipeline processing mode, and each processing unit works in parallel to reduce the end-to-end delay.

[0061] Resource allocation strategy: Based on the task type and system load status, dynamically adjust the computing resource allocation of each layer; Reserve dedicated computing resources for critical tasks (such as howling prediction) to ensure performance in the worst case; Implement an adaptive task migration algorithm to intelligently allocate computing tasks between layers and optimize the overall performance; Set an upper limit on resource usage to prevent a single task from occupying too many resources and causing an increase in system latency.

[0062] In actual application scenarios, this hierarchical edge computing architecture exhibits the following characteristics: The end-to-end delay of howling prediction and suppression is less than 10 milliseconds, meeting the requirements of real-time audio processing; The end-to-end delay control for acoustic event detection is within 100 milliseconds, enabling users to perceive real-time response. The system can independently complete over 90% of its functions without cloud connection, and only advanced analysis and model updates require cloud connection. The edge aggregation layer can simultaneously process data streams from up to 16 wireless microphones, with the average CPU load remaining below 30%. The battery life of the wireless microphone is improved by 30% compared to traditional solutions, thanks to the reasonable allocation of computing tasks.

[0063] Step 5.2, Computing task allocation and scheduling; Optimize computing task allocation according to task characteristics and real-time requirements: Tasks executed at the device layer include: Audio signal preprocessing (sampling rate conversion, noise reduction, etc.); Basic feature extraction (MFCC, spectral features, etc.); Lightweight event classification (using the convolutional neural network in 4.1); Initial howling detection (based on simplified frequency domain analysis); Tasks executed at the edge aggregation layer include: Multi-device information fusion and collaborative detection; Acoustic transfer function matrix update and analysis; Execution of complex howling prediction algorithms; Adaptive filter parameter optimization; Tasks executed at the cloud layer include: Model training and optimization; Historical data analysis; System performance evaluation; Step 5.3, Energy-saving communication strategy; Implement an efficient energy-saving communication method to reduce system power consumption: Adopt an event-driven communication method, and trigger data upload only when potential events or significant changes in the acoustic environment are detected; Use adaptive sampling rate technology to dynamically adjust the data acquisition frequency according to environmental complexity: Silent environment: Reduce the sampling rate and processing frequency; Complex or changing environment: Increase the sampling rate and processing frequency; Implement data compression and transmission, and only transmit effective features instead of raw audio data. The compression algorithm is selected based on the balance between computational complexity and compression efficiency; Establish an optimized wireless communication protocol between the wireless microphone and the charging dock to support multi-device synchronization and low-power transmission.

[0064] Step 5.4, System Collaborative Optimization Algorithm; Achieve collaborative optimization among subsystems: Build a status sharing module to enable each device to understand the overall system status, such as the working modes of charging stations and other microphones, battery levels, etc.; Introduce a dynamic task migration algorithm to migrate computing tasks from high-load devices to low-load devices when the load is unbalanced; Build a priority processing queue to ensure that critical tasks (such as howling prevention) obtain priority processing resources; Implement an adaptive sleep strategy to strategically sleep some devices or functional modules according to the usage scenario and battery status.

[0065] This hierarchical architecture method makes full use of the advantages of edge computing, reasonably distributes computing tasks among different levels, not only ensures the system's real-time response ability to critical events (such as potential howling, abnormal acoustic events), but also effectively reduces the overall system power consumption and communication bandwidth requirements, enabling the wireless microphone system to work stably for a long time and providing users with reliable sound pickup and environmental monitoring services.

[0066] Application example of this embodiment: To verify the effectiveness of this embodiment, we deployed this wireless microphone intelligent control system in a multi-functional conference hall and conducted on-site tests and data collection. The following is the specific implementation process and effect verification of the system in actual application.

[0067] The test site is a 300-square-meter multi-functional conference hall, which has the following characteristics: Movable partition walls that can adjust the space size and shape according to requirements; The ceiling is made of reflective material, the floor is carpeted, and part of the surrounding walls are glass and part are sound-absorbing materials; Under standard configuration, there are 4 main speakers and 2 auxiliary speakers; Daily activities in the space include meetings, speeches, small music performances and other uses.

[0068] The following devices were deployed in this scenario: 8 wireless microphones, each equipped with 3 microphone units and a built-in signal processing unit; 2 charging stations, each integrating 6 fixed capacitor microphones and an edge computing server; The main console connected to the conference hall audio system.

[0069] Example of building a spatial acoustic topology model: During the system initialization phase, spatial acoustic characteristic data was obtained by sending test signals to the environment. Table 1 shows the main eigenvalue of the acoustic transfer function between some measurement points and the characteristics of the determined main acoustic feedback paths.

[0070] Table 1: Main eigenvalue of the spatial acoustic transfer function and characteristics of the feedback path;

[0071] Based on these data, the system constructed a complete spatial acoustic topology model for subsequent howling prediction and suppression. After performing singular value decomposition on the obtained complete transfer function matrix, 4 main acoustic feedback paths were determined, and these paths together contributed 92.7% of the total feedback energy.

[0072] Adaptive howling suppression example: According to the spatial acoustic topology model, the system automatically identified potential howling frequency points and configured the corresponding parameters of the adaptive notch filter. Table 2 shows these parameters and their adaptive adjustment after spatial variation.

[0073] Table 2: Parameter configuration and adjustment of the adaptive notch filter;

[0074] Acoustic event detection and localization example: During the operation of the system in the conference hall, multiple acoustic events were detected and located. Table 3 shows some typical acoustic events identified by the system and their characteristics.

[0075] Table 3: Detected typical acoustic events and their characteristics;

[0076] The acoustic event space heat map constructed by the system helps the staff to monitor the acoustic condition of the entire conference hall in real time, improving the response speed to abnormal situations. In a test, the system successfully detected the sound of broken glass in the back row of the auditorium and provided accurate location information within less than 100 milliseconds.

[0077] Edge computing performance example: Under different load conditions, the computing resource occupancy and latency performance of each layer of the system are shown in Table 4.

[0078] Table 4: Performance indicators of the edge computing architecture;

[0079] Verification of technical effects: To verify the effectiveness of this embodiment, we compared the performance of traditional howling suppression techniques and this solution in various scenarios. Tables 5 and 6 show two key technical effects of the system: howling prediction and suppression effects, and the ability to adapt to spatial changes.

[0080] Table 5: Comparison of howling prediction and suppression effects;

[0081] Table 6: Comparison of the ability to adapt to changes in spatial acoustic conditions;

[0082] Table 7: Comprehensive evaluation of the overall system performance;

[0083] Through the above real application examples, the significant advantages of the wireless microphone intelligent control method of this embodiment in howling prediction and suppression, and environmental adaptability are verified. The system can not only detect potential howling 2 - 3 dB in advance, increasing the system gain margin by 4.5 dB, but also maintain a suppression efficiency of over 95% when the spatial acoustic conditions change, while endowing the wireless microphone system with a new function of environmental safety monitoring.

[0084] The above describes the embodiments of the present invention, but these embodiments are not limited to the above specific embodiments. The above specific embodiments are merely illustrative and not restrictive. Under the inspiration of this embodiment, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of this embodiment.

Claims

1. A method for intelligent control of a wireless microphone, characterized in that, It includes the following steps: Collect the first type of multi-point acoustic data through multiple wireless microphones and fixed microphones distributed in space, construct a spatial acoustic transfer function matrix, analyze the characteristics of the acoustic feedback path using singular value decomposition, and form a spatial acoustic topology model; Based on the results of spatial acoustic topology analysis, construct and optimize an adaptive notch filter bank to achieve precise howling suppression; Collect the second type of acoustic data in real time through a distributed acoustic sensor network, construct a three-dimensional sound field distribution, detect changes in the spatial acoustic environment, and generate optimized suppression parameters for the changing area; Integrate a lightweight convolutional neural network on each microphone device, combine multi-microphone information fusion technology, and achieve the recognition and localization of acoustic anomaly events; Adopt a hierarchical edge computing architecture and an energy-saving communication strategy to perform hierarchical processing on acoustic data, and achieve low-latency acoustic event processing and howling prevention.

2. The wireless microphone intelligent control method according to claim 1, wherein, The steps of constructing the spatial acoustic transfer function matrix include: In the system initialization stage, sequentially output excitation signals through each speaker; All microphones simultaneously record the received signals; For each pair of speaker-microphone combinations, calculate the transfer function to form a spatial acoustic transfer function matrix.

3. A method for intelligent control of a wireless microphone according to claim 1, characterized in that, The steps of constructing the adaptive notch filter bank include: According to the potential howling frequency points identified by acoustic topology analysis and their risk indices, construct a second-order notch filter unit for each potential howling frequency point; Optimize the attenuation factor of each notch filter according to the howling risk index; Construct a multi-stage linked notch filtering structure to handle the mutual influence between howling frequency points; Achieve predictive howling suppression and suppress it before the howling is fully formed.

4. A method for intelligent control of a wireless microphone according to claim 1, characterized in that, The steps of constructing the distributed acoustic sensor network include: Use the fixed microphone on the charging stand as the reference node and the wireless microphone as the dynamic node; Each wireless microphone is equipped with an inertial measurement unit to sense its own motion state; Estimate the relative position between the wireless microphone and the reference node based on time of flight or received signal strength indication; Realize data transmission and synchronization between nodes through a wireless communication network.

5. A method for intelligent control of a wireless microphone according to claim 1, characterized in that, The steps of detecting changes in the spatial acoustic environment include: Periodically update the spatial acoustic transfer function matrix and calculate the change rate between the spatial acoustic transfer function matrix and the matrix at the previous moment; Use singular value decomposition to track the change direction of acoustic characteristics; Based on the components of the singular vector, determine the spatial area of acoustic environment change; When the change rate exceeds the preset threshold, trigger the adaptive update mechanism.

6. The intelligent control method of a wireless microphone according to claim 1, wherein, The structure of the lightweight convolutional neural network includes: An input layer for receiving feature vectors; Two convolutional layers and pooling layers for extracting acoustic features; Two fully connected layers for classifying acoustic events; Adopt weight pruning technology and fixed-point quantization representation to reduce the parameter scale, and control the overall model size within 500KB.

7. A method for intelligent control of a wireless microphone according to claim 1, characterized in that, The multi-microphone information fusion technology includes: Calculate the spatial position of the sound source based on the time difference and signal strength ratio; Associate the same type of events identified by multiple devices based on the spatio-temporal consistency principle; Use a confidence-weighted voting mechanism to fuse the event classification results of multiple devices; Construct an acoustic event spatial heat map and identify the high-activity area as the focus of potential anomaly events.

8. A method for intelligent control of a wireless microphone according to claim 1, characterized in that, The described hierarchical edge computing architecture includes: The device layer, which consists of low-power microprocessors built into wireless microphones and is responsible for signal acquisition and preliminary feature extraction; The edge aggregation layer, which consists of edge servers integrated into the charging dock and is responsible for multi-device data integration and model update; The cloud layer, which is used for long-term data storage and model optimization; A hierarchical message queue mechanism and an adaptive task migration algorithm are used for communication and resource optimization between each layer.

9. A method for intelligent control of a wireless microphone according to claim 1, wherein, The described energy-saving communication strategy includes: Adopting an event-driven communication mechanism and triggering data upload only when potential events are detected; Dynamically adjusting the data acquisition frequency according to the environmental complexity; Implementing compressed data transmission and only transmitting effective features instead of raw audio data; Establishing an optimized wireless communication protocol between the wireless microphone and the charging dock to support multi-device synchronization and low-power transmission.

10. A wireless microphone intelligent control system for implementing the wireless microphone intelligent control method according to any one of claims 1-9, characterized in that, Including: Multiple wireless microphone modules, each equipped with a microphone array, a signal processing unit, and a wireless communication module; The charging dock module, which integrates a fixed microphone array and an edge computing server; The spatial acoustic topology analysis module, which is used to construct and analyze the spatial acoustic transfer function matrix; The adaptive feedback suppression module, which is used to achieve precise feedback control based on topology analysis; The distributed acoustic event detection module, which is used to achieve the identification and localization of acoustic anomaly events with multi-device collaboration.

Images