A long-distance self-organizing network full-duplex voice communication system for fire fighting

Abstract

The present invention discloses a long-distance self-organizing network full-duplex voice communication system for firefighting, which relates to the field of voice communication technology. The system realizes real-time voice collection of all personnel through the coordinated work of a microphone array module, a noise perception module, a noise filtering module, a communication state evaluation module and a signal gain adjustment module, and ensures two-way voice communication between firefighters and the command center. At the fire rescue site, the noise perception module can accurately identify various background noises in the fire environment, such as explosion sounds or equipment operation sounds, and timely classify and mark noise interference areas. The noise filtering module can dynamically adjust the noise suppression coefficient according to the noise type and intensity through an adaptive noise suppression algorithm, and effectively remove interference signals. Through this real-time and dynamic noise suppression, firefighters can clearly receive command information in a complex environment, reducing misunderstandings or communication failures caused by noise interference.

Description

Technical Field

[0001] The invention relates to the technical field of voice communication, in particular to a long-distance self-organizing network full-duplex voice communication system for fire fighting. Background Art

[0002] During the fire rescue process, the reliability and clarity of the communication system are crucial to ensuring rescue efficiency and personnel safety. Firefighters wear microphones and wireless communication equipment at the fire scene to achieve remote voice communication to ensure command, dispatch and collaborative work. However, the fire environment is usually accompanied by extremely complex and noisy background noises. These noises not only come from flames, explosions, alarms, building collapses, etc., but also include a variety of unpredictable noises such as the operation of firefighting equipment, water flow, and shouting of people. These noises will affect the quality of voice communication to a certain extent, causing the voice signal to be blurred and distorted. In severe cases, it may even completely block communication, greatly affecting the coordination and command of rescue.

[0003] As a new communication method, ad hoc networking technology enables each communication node to transmit information without central control through autonomous connection and collaboration between wireless network devices. This technology has great potential in fire rescue. However, in traditional wireless voice communication systems using ad hoc networking technology, it is common to use fixed signal processing schemes to reduce the impact of background noise on communication quality. Especially at the fire scene, drastic fluctuations in noise levels may cause the system to automatically switch to the wrong gain mode, further exacerbating communication distortion. In addition, traditional communication systems lack the ability to adaptively adjust to real-time noise changes, making it difficult to accurately identify and filter multiple types of noise, causing confusion between voice signals and noise signals, affecting the stability of communication quality. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a long-distance self-organizing network full-duplex voice communication system for fire fighting to solve the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a long-distance self-organizing network full-duplex voice communication system for fire fighting, including a microphone array module, a noise perception module, a noise filtering module, a communication status evaluation module and a signal gain adjustment module;

[0006] Microphone array module, which is used to deploy distributed MEMS microphone arrays at the fire scene and provide microphones for each firefighter. Each microphone node is connected through ad hoc networking technology to achieve full-duplex voice communication.

[0007] The noise perception module is used to monitor the background noise data of the fire scene in real time, identify and automatically classify it, and then construct the noise index of the jth microphone position. , if the noise index at the jth microphone position is > Noise threshold, indicating that the noise level interference in the environment of the current j-th microphone position is abnormal, generates a mark point for the noise level interference abnormal area, and issues a low noise suppression instruction;

[0008] The noise filtering module is used to generate the noise suppression coefficient of the jth microphone position using an adaptive filtering algorithm after receiving the low noise suppression instruction. , and after removing the noise component of the first speech signal at the j-th microphone position, generate a filtered second speech signal;

[0009] The communication status evaluation module is used to collect the real-time distance H between the j-th microphone position and the sending point based on the long-distance transmission requirements, and extract the second voice signal to obtain the voice signal strength at the j-th microphone position. , and associate the real-time distance H of the jth microphone position and the voice signal strength And the noise suppression coefficient , get the communication status evaluation value of the jth microphone position ;

[0010] Signal gain adjustment module, and the communication status evaluation value of the jth microphone position Evaluate to determine if the system should enter gain mode to adapt to changing conditions at the fire scene.

[0011] Preferably, the noise sensing module includes a distributed acquisition unit and a noise type identification unit;

[0012] The distributed acquisition unit is used to collect noise data at the fire scene in real time through a distributed MEMS micro-electromechanical microphone array in the fire environment, with a sampling frequency higher than 32kHz;

[0013] The noise type recognition unit is used to analyze noise signals in multiple frequency bands, and after performing fast Fourier transform, spectrum analysis, constructing feature space and comparison on the spectrum feature set, automatically classifies and obtains each type of noise source.

[0014] Preferably, the noise type identification unit includes an extraction submodule, a classification submodule and a noise index calculation submodule;

[0015] The extraction submodule is used to extract background noise data and process the noise signals of multiple frequency bands to obtain the noise feature vector F. The specific processing steps are as follows:

[0016] S11. Extract the time series represented by the collected signal from the noise data Where t is the sampling time, is the noise amplitude value of each frame signal;

[0017] S12, first perform Fourier transform on the signal to obtain its spectrum, and then calculate the noise amplitude value of each frame signal. , whose spectrum is expressed as , that is, the amplitude at frequency f, is converted to the frequency domain through fast Fourier transform FFT, and the expression is:

[0018] ; In the formula, It is the complex spectrum of the signal at frequency f, indicating the amplitude and phase information of the frequency component. is the signal sample value at discrete time n, from the time domain signal N represents the total number of samples, which controls the frequency resolution. It is composed of negative exponential form, which indicates the phase rotation of different frequency components. j indicates the imaginary unit, which is used to indicate the phase information in Fourier transform. f indicates the frequency variable, which determines the position of the frequency component. n indicates the index of the time sample, which indicates the position of each time sample in the signal. This spectrum is used to determine the frequency distribution characteristics of flame burning sound, alarm sound, and water flow noise.

[0019] S13, extracting multiple features from the spectrum graph, the multiple features including: power spectral density PSD, Mel frequency cepstrum decomposition and zero crossing rate ZCR;

[0020] S131, based on the complex spectrum of the signal at frequency f , analyze the energy distribution of each frequency component, and construct the power spectral density PSD through the following formula:

[0021] ;

[0022] Among them, PSD represents the power spectral density of each frequency, which describes the energy distribution of each frequency component;

[0023] S132, based on the human hearing characteristics, using the Mel filter bank to filter the complex spectrum of the signal at frequency f Convert to the Mel scale spectrum. The Mel scale is a nonlinear transformation based on the human ear's perception of different frequencies. The Mel frequency cepstral coefficient MFCC formula is:

[0024] ;

[0025] In the formula, refers to the center frequency of each filter in the Mel filter bank, Indicates the center frequency The spectral components on represents the frequency response of the kth Mel filter, T is the total number of Mel filters; log is a logarithmic operation, The filtered power spectrum is converted into a logarithmic value. DCT is discrete cosine transform. Mel frequency cepstral coefficients (MFCC) are obtained by using DCT compression features to represent the speech characteristics of the signal.

[0026] S133, zero crossing rate ZCR is used to detect high-frequency components in the signal, indicating the number of times the sign of the signal changes, that is, the number of times the signal crosses from positive to negative or from negative to positive. The formula is expressed as: ;

[0027] In the formula, is the signal sample value at discrete time n, L is the length of the signal frame, is the indicator function, when When it returns 1, it means the sign changes once, otherwise it returns 0;

[0028] S14, based on the power spectrum density PSD, Mel frequency cepstrum decomposition and zero crossing rate ZCR, a noise feature vector F is formed, which is expressed as: ;

[0029] S15. Use vector machine (SVM), k nearest neighbor (KNN) or convolutional neural network (CNN) to establish a noise classification model and classify the noise feature vector Perform standardization so that the mean of each dimension is 0 and the variance is 1, and obtain the standardized noise feature vector It is expressed as: ;

[0030] in, is the mean of the noise feature vector, is the standard deviation of the noise feature vector.

[0031] Preferably, the classification submodule is used to transform the standardized noise feature vector Input to the trained noise classification model for binary classification, for each category , create a two-category SVM classifier, The samples are considered as positive and all other samples are considered as negative.

[0032] And mark the category The samples in the class are +1, and the samples of all other classes are -1. The output of each binary SVM classifier will indicate the input noise feature vector Whether it belongs to this category, count the votes of all classifiers, and take the category with the most votes as the prediction result of the noise type;

[0033] The prediction results include but are not limited to the following noise types: burning flames, explosions, building collapse, alarms, water flow and pump noises, demolition tool noises, rescue equipment noises, gas leaks, glass breaking, high-temperature metal deformation, and wind and smoke movement.

[0034] Preferably, the noise index calculation submodule is used to extract noise feature vectors at different positions in a distributed MEMS micro-electromechanical microphone array. , the noise feature vector extracted at the jth microphone position contains M noise types, recorded as: ; Where M is the total number of noise types;

[0035] And by weighted summing the noise feature vectors of each noise type, the noise index at the jth microphone position is obtained by the following formula: : ;

[0036] In the formula, The weight coefficient representing the Mth noise type is set according to the interference of each frequency band in the noise environment or determined through training; represents the noise feature vector of the g-th noise type at the j-th position,

[0037] And preset the noise threshold, if the noise index at the jth microphone position > Noise threshold, indicating that the noise level interference in the environment of the current j-th microphone position is abnormal, generates a mark point for the noise level interference abnormal area, and issues a low noise suppression instruction;

[0038] If the noise index at the jth microphone position ≤noise threshold, it means that the noise level interference in the environment of the current j-th microphone position is normal and no adjustment is required.

[0039] Preferably, the noise type identification unit further includes a priority submodule;

[0040] The priority submodule is used to identify the locations of the marker points in the abnormal area where the noise level interferes, and the noise index of the j-th microphone position is used. Generate the first priority processing queue from large to small.

[0041] Preferably, the noise filtering module includes a noise suppression calculation subunit and an adaptive filtering subunit;

[0042] The noise suppression calculation subunit is used to extract the noise index of the j-th microphone position after receiving the first priority processing queue. , mapping the noise suppression coefficient of the j-th microphone position : ;

[0043] In the formula, It is the adjustment factor of noise suppression. The range of the adjustment factor is between 0 and 1. The larger the value, the higher the suppression strength. It is used to control the overall suppression strength. is the sensitivity of the control suppression coefficient to the noise index, is the suppression threshold; exp represents the exponential function, and exp uses the Sigmoid function so that the noise suppression coefficient With the noise index The change of is gradually increased, avoiding excessively abrupt suppression effects, thereby reducing damage to the speech signal;

[0044] The adaptive filtering subunit is used to receive the first speech signal with noise collected by the microphone, divide it into short time frames, and use a 20-30 millisecond window for real-time collection;

[0045] Initialize the parameters of the adaptive filter, including the filter length and initial weight. The weight is initialized to 0 to avoid initial impact on the speech signal.

[0046] Use the noise suppression coefficient for the jth microphone position The filter weight is adjusted to remove the noise component from the first speech signal, and the filter generates a second speech signal after the noise is removed.

[0047] Preferably, the communication status evaluation module includes a real-time distance acquisition unit, a signal strength calculation unit and an associated unit;

[0048] The real-time distance acquisition unit is used to collect the real-time distance H between the j-th microphone position and the sending point in real time, because the real-time distance H will affect the propagation attenuation of the signal. As the real-time distance H increases, the signal strength will gradually decay;

[0049] The signal strength calculation unit is used to extract the voice signal strength in the second voice signal, and obtain the voice signal strength by calculating the following formula: : ;

[0050] in, represents the amplitude of the second speech signal at the jth microphone position at the i-th moment, which is acquired through the sound pressure sensor; R is the number of sampling points of the second speech signal;

[0051] The associated unit is used to extract the real-time distance H and speech signal strength of the jth microphone position. And the noise suppression coefficient , after dimensionless processing, the communication status evaluation value of the jth microphone position is generated by the following formula : ;

[0052] The meaning of the formula is: Communication status evaluation value Dynamically reflects the signal quality and transmission conditions in the current environment.

[0053] Preferably, the signal gain adjustment module includes a communication state evaluation unit and a strategy unit;

[0054] The communication state evaluation unit is used to preset a communication threshold X and set the communication state evaluation value of the jth microphone position The communication status is compared with the communication threshold X to obtain the communication status evaluation results, including:

[0055] If the communication status evaluation value of the jth microphone position ≥communication threshold X, indicating that the communication signal quality at the jth microphone position is qualified;

[0056] If the communication threshold X*80%≤the communication status evaluation value of the jth microphone position <communication threshold X, indicating that the communication signal quality at the jth microphone position is unqualified, and a first warning instruction is generated;

[0057] If the communication status evaluation value of the jth microphone position <communication threshold X*80%, indicating that the communication signal quality at the jth microphone position is unqualified, and a second early warning instruction is generated.

[0058] Preferably, the strategy unit is used to generate corresponding instructions when receiving the first warning instruction and the second warning instruction, including:

[0059] Generate a first gain adjustment instruction according to the first warning instruction, including: performing signal gain through a digital signal processor to increase the strength of the voice signal 10-30%;

[0060] Generating a second gain adjustment instruction according to the second warning instruction includes: performing signal gain through a digital signal processor to increase the strength of the voice signal If a signal interruption is detected, the system will automatically detect the signal interruption location, recalculate the best signal path, select an alternative path and restore the communication connection within 30 seconds.

[0061] The present invention provides a long-distance self-organizing network full-duplex voice communication system for fire fighting. It has the following beneficial effects:

[0062] (1) Distributed MEMS microphone arrays can collect firefighters' voice signals and noise data of the surrounding environment in real time; each firefighter wears a microphone to ensure that all participants can communicate in real time and clearly during the fire rescue process, ensuring the efficiency of command, dispatch and collaborative operations. Especially in the complex and dangerous environment of the fire scene, it can provide key remote voice commands and information feedback, effectively avoiding rescue errors caused by communication interruptions or delays. By connecting each microphone node through self-organizing network technology, the system can achieve full-duplex voice communication, which not only supports real-time voice transmission, but also supports two-way communication. At the fire scene, firefighters can receive instructions from the command center in real time, and at the same time feed back the voice of their own location to the command center, providing accurate command and collaborative information for fire rescue.

[0063] (2) The noise sensing module can monitor the background noise at the fire scene in real time and automatically classify various noises (such as flames, explosions, equipment operation sounds, etc.). When the noise index at a certain microphone position exceeds the set noise threshold, the system can promptly identify the abnormal noise interference area and generate noise interference markers to provide a basis for subsequent noise processing.

[0064] (3) The system uses an adaptive filtering algorithm based on the noise suppression instruction through the noise filtering module to generate a noise suppression coefficient, remove the noise component in the first voice signal, and thus generate a clearer second voice signal. This real-time, dynamic noise suppression capability effectively reduces the impact of complex background noise at the fire scene on voice communication, and improves voice intelligibility and transmission quality.

[0065] (4) The communication status evaluation module can collect the real-time distance between the microphone node and the transmission point, the voice signal strength and the noise suppression coefficient according to the long-distance transmission requirements of the fire scene, and comprehensively evaluate the communication status of each microphone position. This evaluation mechanism ensures that the system can evaluate the communication quality in real time under different environments and conditions, and provide accurate data support for subsequent gain adjustment. The system automatically adjusts the signal gain mode according to the real-time communication status evaluation results through the signal gain adjustment module. For areas with strong noise interference or poor communication quality, the system can automatically enter the gain mode to adapt to the changing noise and communication conditions at the fire scene, thereby improving the stability and reliability of the communication signal. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 The present invention is a schematic diagram of the flow chart of a long-distance self-organizing network full-duplex voice communication system for fire fighting. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] Example 1

[0069] See also Figure 1 , the present invention provides a long-distance self-organizing network full-duplex voice communication system for fire fighting, including a microphone array module, a noise perception module, a noise filtering module, a communication state evaluation module and a signal gain adjustment module;

[0070] Microphone array module, which is used to deploy distributed MEMS microphone arrays at the fire scene and provide microphones for each firefighter. Each microphone node is connected through ad hoc networking technology to achieve full-duplex voice communication.

[0071] The distributed MEMS microphone array can collect the voice signals of firefighters and the noise data of the surrounding environment in real time; each firefighter wears a microphone to ensure that all participants can communicate in real time and clearly during the fire rescue process, ensuring the efficiency of command, dispatch and collaborative operations. Especially in the complex and dangerous environment of the fire scene, it can provide key remote voice commands and information feedback, effectively avoiding rescue errors caused by communication interruptions or delays. By connecting each microphone node through self-organizing network technology, the system can achieve full-duplex voice communication, which not only supports real-time voice transmission, but also supports two-way communication. At the fire scene, firefighters can receive instructions from the command center in real time, and at the same time feed back the voice of their own location to the command center, providing accurate command and coordination information for fire rescue.

[0072] The noise perception module is used to monitor the background noise data of the fire scene in real time, identify and automatically classify it, and then construct the noise index of the jth microphone position. , if the noise index at the jth microphone position is > Noise threshold, indicating that the noise level interference in the environment of the current j-th microphone position is abnormal, generates a mark point for the noise level interference abnormal area, and issues a low noise suppression instruction;

[0073] The noise filtering module is used to generate the noise suppression coefficient of the jth microphone position using an adaptive filtering algorithm after receiving the low noise suppression instruction. , and after removing the noise component of the first speech signal at the j-th microphone position, generate a filtered second speech signal;

[0074] The communication status evaluation module is used to collect the real-time distance H between the j-th microphone position and the sending point based on the long-distance transmission requirements, and extract the second voice signal to obtain the voice signal strength at the j-th microphone position. , and associate the real-time distance H of the jth microphone position and the voice signal strength And the noise suppression coefficient , get the communication status evaluation value of the jth microphone position ;

[0075] Signal gain adjustment module, and the communication status evaluation value of the jth microphone position Evaluate to determine if the system should enter gain mode to adapt to changing conditions at the fire scene.

[0076] In this embodiment, the noise perception module can monitor the background noise at the fire scene in real time and automatically classify various noises (such as flame burning, explosion, equipment operation sound, etc.). When the noise index at a certain microphone position exceeds the set threshold, the system can promptly identify the abnormal noise interference area and generate a noise interference mark point to provide a basis for subsequent noise processing. This function ensures that the system can still maintain the accuracy and clarity of voice communication in an environment with large noise changes. Through the noise filtering module, the system uses an adaptive filtering algorithm according to the noise suppression instruction to generate a noise suppression coefficient, remove the noise component in the first voice signal, and thus generate a clearer second voice signal. This real-time and dynamic noise suppression capability effectively reduces the impact of complex background noise at the fire scene on voice communication, and improves the intelligibility and transmission quality of the voice. The communication status evaluation module can collect the real-time distance between the microphone node and the sending point, the voice signal strength and the noise suppression coefficient according to the long-distance transmission requirements of the fire scene, and comprehensively evaluate the communication status of each microphone position. This evaluation mechanism ensures that the system can evaluate the communication quality in real time under different environments and conditions, and provide accurate data support for subsequent gain adjustment. The system automatically adjusts the signal gain mode based on the real-time communication status evaluation results through the signal gain adjustment module. For areas with strong noise interference or poor communication quality, the system can automatically enter the gain mode to adapt to the changing noise and communication conditions at the fire scene, thereby improving the stability and reliability of the communication signal.

[0077] The system of the present invention can effectively cope with the complex and noisy environment of the fire scene, ensure that the voice communication between firefighters is clear and stable, and the command center can obtain the on-site situation in real time and make timely command adjustments. The system's noise perception, noise filtering, communication status evaluation and gain adjustment functions provide more accurate and efficient communication guarantees for fire rescue, promoting the improvement of rescue efficiency and personnel safety.

[0078] Example 2

[0079] This embodiment is explained in Example 1. For details, please refer to Figure 1 ,The noise perception module includes a distributed acquisition unit and a noise type recognition unit;

[0080] The distributed acquisition unit is used to collect noise data at the fire scene in real time through a distributed MEMS microelectromechanical microphone array in the fire environment, with a sampling frequency higher than 32kHz. This high-frequency sampling can ensure accurate capture of complex noise signals at the scene, especially in high-noise environments, and can fully retain detailed noise information, providing high-quality data support for subsequent noise analysis and processing.

[0081] The noise type recognition unit is used to analyze noise signals in multiple frequency bands, and after performing fast Fourier transform, spectrum analysis, constructing feature space and comparison on the spectrum feature set, automatically classifies and obtains each type of noise source.

[0082] In this embodiment, by accurately identifying and classifying noise sources, the system can better manage and process the complex noise at the fire scene. In the noise filtering stage, the system can suppress specific noise sources in a targeted manner, thereby improving the clarity of the voice signal. This classification and recognition capability effectively improves the accuracy of the noise suppression algorithm and reduces unnecessary noise interference.

[0083] Example 3

[0084] This embodiment is explained in Example 2. For details, please refer to Figure 1 ,The noise type recognition unit includes an extraction submodule, a classification submodule and a noise index calculation submodule;

[0085] The extraction submodule is used to extract background noise data and process the noise signals of multiple frequency bands to obtain the noise feature vector F. The specific processing steps are as follows:

[0086] S11. Extract the time series represented by the collected signal from the noise data , where t is the sampling time, is the noise amplitude value of each frame signal; by collecting the noise amplitude value of each frame signal (i.e., time series), the changes of various noises at the fire scene can be accurately recorded. The density of sampling time and time series ensures the integrity of noise information, can capture the rapidly changing noise characteristics at the scene, and lays the foundation for subsequent spectrum analysis and noise classification. Using a higher sampling frequency can ensure that high-frequency noise components are captured, which is especially important when processing sounds with higher frequency components such as explosions or water pump noises.

[0087] S12, first perform Fourier transform on the signal to obtain its spectrum, and then calculate the noise amplitude value of each frame signal. , whose spectrum is expressed as , that is, the amplitude at frequency f, is converted to the frequency domain through fast Fourier transform FFT, and the expression is: ;

[0088] In the formula, It is the complex spectrum of the signal at frequency f, indicating the amplitude and phase information of the frequency component. is the signal sample value at discrete time n, from the time domain signal N represents the total number of samples, which controls the frequency resolution. It is composed of negative exponential form, which indicates the phase rotation of different frequency components. j indicates the imaginary unit, which is used to indicate the phase information in Fourier transform. f indicates the frequency variable, which determines the position of the frequency component. n indicates the index of the time sample, which indicates the position of each time sample in the signal. This spectrum is used to determine the frequency distribution characteristics of flame burning sound, alarm sound, and water flow noise. This step enables the distribution characteristics of the noise source at different frequencies to be clearly revealed, which is helpful for the subsequent noise source identification. For example, the flame burning sound may be in the low frequency band, while the explosion sound may contain higher frequency components.

[0089] S13. Extract multiple features from the spectrum graph, including power spectral density (PSD), Mel-frequency cepstrum decomposition, and zero-crossing rate (ZCR). At a fire scene, noise signals are usually complex and dynamically changing. Power spectral density can provide a clear energy distribution for each frequency, facilitating subsequent classification and noise source management.

[0090] S131, based on the complex spectrum of the signal at frequency f , analyze the energy distribution of each frequency component, and construct the power spectral density PSD through the following formula: ;

[0091] Among them, PSD represents the power spectral density of each frequency, which describes the energy distribution of each frequency component;

[0092] S132, based on the human hearing characteristics, using the Mel filter bank to filter the complex spectrum of the signal at frequency f Convert to the Mel scale spectrum. The Mel scale is a nonlinear transformation based on the human ear's perception of different frequencies. The Mel frequency cepstral coefficient MFCC formula is: ;

[0093] In the formula, refers to the center frequency of each filter in the Mel filter bank, Indicates the center frequency The spectral components on represents the frequency response of the kth Mel filter, T is the total number of Mel filters; log is a logarithmic operation, It converts the filtered power spectrum into a logarithmic value. DCT is discrete cosine transform. The Mel-frequency cepstrum coefficients MFCC are obtained by using DCT compression features to represent the speech features of the signal. DCT (discrete cosine transform) compresses the MFCC features and retains the most important spectral features, making the model easier to process and more efficient in calculation, while reducing redundant information and improving the efficiency and accuracy of the system.

[0094] S133, zero crossing rate ZCR is used to detect high-frequency components in the signal, indicating the number of times the sign of the signal changes, that is, the number of times the signal crosses from positive to negative or from negative to positive. The formula is expressed as: ;

[0095] In the formula, is the signal sample value at discrete time n, L is the length of the signal frame, is the indicator function, when When , it returns 1, indicating that the sign changes once, otherwise it returns 0; ZCR can quickly reflect the high-frequency changes of the signal by counting the number of times the sign changes in the signal, thereby enhancing the system's ability to respond to sudden noise.

[0096] S14, based on the power spectrum density PSD, Mel frequency cepstrum decomposition and zero crossing rate ZCR, a noise feature vector F is formed, which is expressed as: ;

[0097] The combination of power spectral density (PSD), Mel-frequency cepstrum decomposition, and zero crossing rate (ZCR) can comprehensively describe the frequency and time characteristics of the signal to form a comprehensive noise feature vector. This multi-dimensional feature fusion helps to identify different types of noise in a complex noise environment. Combining different spectral features to form a feature vector improves the differentiation of noise types. This feature vector can more accurately represent the unique properties of the noise source and help the classification model make more accurate predictions.

[0098] S15. Use vector machine (SVM), k nearest neighbor (KNN) or convolutional neural network (CNN) to establish a noise classification model and classify the noise feature vector Perform standardization so that the mean of each dimension is 0 and the variance is 1, and obtain the standardized noise feature vector , expressed as: ;

[0099] in, is the mean of the noise feature vector, is the standard deviation of the noise feature vector. The standardized noise feature vector is more suitable for training various classification models such as support vector machine (SVM), k-nearest neighbor (KNN) or convolutional neural network (CNN), which improves the generalization ability and adaptability of the model and can effectively handle diverse noise environments.

[0100] The classification submodule is used to transform the standardized noise feature vector Input to the trained noise classification model for binary classification, for each category , create a two-category SVM classifier, The samples are considered as positive and all other samples are considered as negative.

[0101] And mark the category The samples in the class are +1, and the samples of all other classes are -1. The output of each binary SVM classifier will indicate the input noise feature vector Whether it belongs to this category, the votes of all classifiers are counted, and the category with the most votes is used as the prediction result of the noise type; the system can accurately distinguish different noise types in a complex environment where multiple noise sources coexist, ensure that the communication quality is not excessively disturbed, and thus improve the command and coordination efficiency at the fire scene.

[0102] The prediction results include but are not limited to the following noise types: burning flames, explosions, building collapse, alarms, water flow and pump noises, demolition tool noises, rescue equipment noises, gas leaks, glass breaking, high-temperature metal deformation, and wind and smoke movement.

[0103] The following is an example of prediction results:

[0104] Flame burning sound: Frequency range: low frequency to medium frequency (about 20Hz-500Hz); Characteristics: low frequency, continuous sound, with irregular "buzzing" sound.

[0105] Popping sound (e.g. explosion caused by thermal expansion or uneven pressure); Frequency range: wide frequency range, including low-frequency to high-frequency components (20Hz–10kHz); Characteristics: instantaneous high-intensity sound, often containing a large amount of high-frequency energy.

[0106] Sound of building collapse: Frequency range: low frequency to medium frequency (about 20Hz-1kHz); Characteristics: low-frequency noise accompanied by vibration, wide frequency, high sound pressure and long duration.

[0107] Alarm sound: Frequency range: high frequency (about 2kHz–4kHz, the specific frequency will vary depending on the type of alarm);

[0108] Features: Continuous single-frequency or dual-frequency high-frequency sound, high intensity and stable frequency.

[0109] The sound of flowing water and the sound of water pumps: Frequency range: medium frequency to low frequency (about 100Hz-1kHz); Characteristics: The sound of flowing water is usually a continuous medium frequency, while the noise of water pumps is in the low frequency range and has regular vibration components.

[0110] Noise from demolition tools (such as cutters and saws): Frequency range: medium to high frequency (about 500Hz–8kHz); Characteristics: significant high-frequency components, with high-frequency sounds of motor operation and friction.

[0111] Rescue equipment noise (such as generators, smoke exhaust equipment): Frequency range: low frequency to medium frequency (about 50Hz-500Hz); Characteristics: strong low-frequency and medium-frequency components, regular sound, and vibration sound of mechanical operation.

[0112] Gas leakage sound: Frequency range: medium frequency to high frequency (about 1kHz-10kHz); Characteristics: high-frequency "hissing" sound, higher frequency, continuous sound.

[0113] Glass breaking sound: Frequency range: wide frequency range, significant high-frequency components (500Hz-12kHz); Characteristics: instantaneous crisp high frequency, with randomly distributed short breaking sounds.

[0114] Sound of high-temperature metal deformation (such as bending of steel beams due to high temperature): Frequency range: low frequency to medium frequency (about 50Hz-1kHz); Characteristics: short "click" sound, low-frequency vibration with instantaneous medium-frequency components.

[0115] Sound of wind and smoke movement: Frequency range: low to medium frequency (about 20Hz-500Hz); Characteristics: low-frequency whistling, continuous and irregular, accompanied by wind noise generated by air flow movement.

[0116] In this embodiment, the powerful ability of the extraction submodule and the classification submodule in the noise type recognition unit to accurately collect, process and classify noise signals in a high noise environment is achieved. Through a series of processes such as high-precision data collection, multi-dimensional feature extraction, and noise type classification, the system can achieve efficient and reliable voice communication in complex environments such as fires, ensuring clear and timely information transmission between firefighters, greatly improving the collaborative efficiency and safety during the rescue process.

[0117] Example 4

[0118] This embodiment is explained in Example 3. For details, please refer to Figure 1 The noise index calculation submodule is used to extract noise feature vectors at different locations in the distributed MEMS microphone array. , the noise feature vector extracted at the jth microphone position contains M noise types, recorded as: ; Where M is the total number of noise types;

[0119] And by weighted summing the noise feature vectors of each noise type, the noise index at the jth microphone position is obtained by the following formula: : ;

[0120] In the formula, The weight coefficient representing the Mth noise type is set according to the interference of each frequency band in the noise environment or determined through training; represents the noise feature vector of the g-th noise type at the j-th position,

[0121] And preset the noise threshold, if the noise index at the jth microphone position > Noise threshold, indicating that the noise level interference in the environment of the current j-th microphone position is abnormal, generates a mark point for the noise level interference abnormal area, and issues a low noise suppression instruction;

[0122] If the noise index at the jth microphone position ≤noise threshold, it means that the noise level interference in the environment of the current j-th microphone position is normal and no adjustment is required.

[0123] In this embodiment, in a distributed MEMS micro-electromechanical microphone array, noise feature vectors at different positions can provide comprehensive noise environment data. This method can obtain noise information at different positions in a fire or other complex environment, ensure that environmental noise is monitored from multiple angles, and avoid the situation where the noise information at a single position is insufficient or incomplete. Different microphone positions may be affected by different noise sources, such as explosions, alarms, or water flows at a fire scene. By extracting noise features from multiple microphone positions, the differences in noise characteristics at different positions can be accurately captured, which helps to comprehensively evaluate the overall noise level. The calculated noise index provides the system with a dynamic, quantitative noise level assessment that can monitor noise changes in different environments in real time. If the noise index at the jth microphone position is >Noise threshold, the system will trigger low noise suppression instructions, effectively reducing noise interference to the environment and improving signal transmission quality.

[0124] Example 5

[0125] This embodiment is explained in Example 1. For details, please refer to Figure 1 ,The noise type recognition unit also includes a priority submodule;

[0126] The priority submodule is used to identify the locations of the marker points in the abnormal area where the noise level interferes, and the noise index of the j-th microphone position is used. Generate the first priority processing queue from large to small.

[0127] In this embodiment, by sorting the noise index, the priority submodule helps the system to allocate resources reasonably. Resources (such as computing power or communication bandwidth) can be used first in areas with the most severe noise interference, while for areas with less noise interference, less processing resources can be used or processing can be delayed. This can avoid over-regulation or waste of processing resources and improve the resource utilization of the overall system.

[0128] Example 6

[0129] This embodiment is explained in Example 1. For details, please refer to Figure 1 , the noise filtering module includes a noise suppression calculation subunit and an adaptive filtering subunit;

[0130] The noise suppression calculation subunit is used to extract the noise index of the j-th microphone position after receiving the first priority processing queue. , mapping the noise suppression coefficient of the j-th microphone position : ;

[0131] In the formula, It is the adjustment factor of noise suppression. The range of the adjustment factor is between 0 and 1. The larger the value, the higher the suppression strength. It is used to control the overall suppression strength. is the sensitivity of the control suppression coefficient to the noise index, is the suppression threshold; exp represents the exponential function, and exp uses the Sigmoid function so that the noise suppression coefficient With the noise index The change of is gradually increased, avoiding excessively abrupt suppression effects, thereby reducing damage to the speech signal;

[0132] The adaptive filtering subunit is used to receive the first speech signal with noise collected by the microphone, divide it into short time frames, and use a 20-30 millisecond window for real-time collection;

[0133] Initialize the parameters of the adaptive filter, including the filter length and initial weight. The weight is initialized to 0 to avoid initial impact on the speech signal.

[0134] Use the noise suppression coefficient for the jth microphone position The filter weight is adjusted to remove the noise component from the first speech signal, and the filter generates a second speech signal after the noise is removed.

[0135] In this embodiment, the noise suppression calculation subunit maps the corresponding noise suppression coefficient according to the noise index of the j-th microphone position. The adjustment factor of this suppression coefficient ranges from 0 to 1, and the suppression strength can be dynamically adjusted according to the intensity of the noise interference. When the noise interference intensity is high, the suppression intensity increases, effectively reducing the impact of noise on the speech signal; conversely, if the noise interference is light, the suppression coefficient is small to avoid excessive suppression and maintain the original clarity of the speech signal.

[0136] Using the noise suppression coefficient, the weight of the adaptive filter will be adjusted according to the noise suppression coefficient to remove the noise component in real time. The filter is adjusted according to the dynamically changing noise intensity, making the noise removal process more targeted, especially at the fire scene, which is often accompanied by a variety of background noises (such as flames, explosions, alarms, etc.). Through this adaptive filtering method, unnecessary noise can be more effectively filtered out and the key information of the voice signal can be retained. The adaptive filter can process the first voice signal with noise into a denoised second voice signal, providing clearer and more accurate voice content, ensuring that the commander can receive clear instructions in real time. Especially in high-pressure environments such as fire rescue, command and dispatch and team collaboration rely on accurate voice communication, and the denoised voice signal plays a vital role in team collaboration.

[0137] Example 7

[0138] This embodiment is explained in Example 1. For details, please refer to Figure 1 ,The communication state evaluation module includes a real-time distance acquisition unit, a signal strength calculation unit and an associated unit;

[0139] The real-time distance acquisition unit is used to collect the real-time distance H between the j-th microphone position and the sending point in real time, because the real-time distance H will affect the propagation attenuation of the signal. As the real-time distance H increases, the signal strength will gradually decay;

[0140] The signal strength calculation unit is used to extract the voice signal strength in the second voice signal, and obtain the voice signal strength by calculating the following formula: : ;

[0141] in, represents the amplitude of the second speech signal at the jth microphone position at the i-th moment, which is acquired through the sound pressure sensor; R is the number of sampling points of the second speech signal;

[0142] The associated unit is used to extract the real-time distance H and speech signal strength of the jth microphone position. And the noise suppression coefficient , after dimensionless processing, the communication status evaluation value of the jth microphone position is generated by the following formula : ;

[0143] The meaning of the formula is: Communication status evaluation value Dynamically reflects the signal quality and transmission conditions in the current environment.

[0144] In this embodiment, the communication status evaluation module combines the real-time distance H of the j-th microphone position, the voice signal strength And the noise suppression coefficient Multiple factors to generate the communication status evaluation value of the jth microphone position , providing a comprehensive assessment of the communication status. Through the comprehensive analysis of these factors, the signal quality and transmission conditions can be dynamically reflected, and the actual status of the communication environment can be more accurately assessed.

[0145] Example 8

[0146] This embodiment is explained in Example 7. For details, please refer to Figure 1 ,The signal gain adjustment module includes a communication state evaluation unit and a strategy unit;

[0147] The communication state evaluation unit is used to preset a communication threshold X and set the communication state evaluation value of the jth microphone position The communication status is compared with the communication threshold X to obtain the communication status evaluation results, including:

[0148] If the communication status evaluation value of the jth microphone position ≥communication threshold X, indicating that the communication signal quality at the jth microphone position is qualified;

[0149] If the communication threshold X*80%≤the communication status evaluation value of the jth microphone position <communication threshold X, indicating that the communication signal quality at the jth microphone position is unqualified, and a first warning instruction is generated;

[0150] If the communication status evaluation value of the jth microphone position <communication threshold X*80%, indicating that the communication signal quality at the jth microphone position is unqualified, and a second early warning instruction is generated.

[0151] The strategy unit is used to generate corresponding instructions when receiving the first warning instruction and the second warning instruction, including:

[0152] Generate a first gain adjustment instruction according to the first warning instruction, including: performing signal gain through a digital signal processor to increase the strength of the voice signal 10-30%;

[0153] Generating a second gain adjustment instruction according to the second warning instruction includes: performing signal gain through a digital signal processor to increase the strength of the voice signal If a signal interruption is detected, the system will automatically detect the signal interruption location, recalculate the best signal path, select an alternative path and restore the communication connection within 30 seconds.

[0154] The system continuously monitors the network connection status by monitoring the signal quality and communication status between each microphone node and the receiving end in real time. When the communication threshold is less than X*50%, the system will default to the signal terminal and start the emergency handling mechanism; the system uses location tracking technology and network topology information to determine the location of the interrupted signal. For example, the system can mark the location of each microphone through the location information of the sensor (such as GPS positioning location) and infer the interruption location based on the signal strength or attenuation trend. After detecting the interruption location, the system will recalculate the optimal signal propagation path based on the current network status. For example, the Dijkstra algorithm or A* algorithm is used to calculate the shortest path from the interruption location to the destination (such as the command center), taking into account the real-time signal quality factor, and completing the path switching within 30 seconds. The system automatically switches to the new signal path and reconnects. This process includes:

[0155] Start a new signal transmission and reception channel. Switch to a microphone node or signal enhancement area with higher signal strength. The system sets a response time window (30 seconds), within which the system will optimize the network configuration and complete the path switching to ensure that communication is not interrupted for a long time. Within 30 seconds, the communication link is reestablished, and all devices involved in the communication (microphones, receiving terminals, etc.) will return to normal full-duplex voice communication status. After the signal is restored, the system will continue to monitor the communication status to ensure that the recovery process is stable. If the problem occurs again during the recovery process, the system will automatically repeat the above steps to repair it.

[0156] In this embodiment, by comparing the communication status evaluation value with the communication threshold, the system can accurately evaluate the quality of the signal. The signal quality is divided into three levels: qualified, slightly unqualified, and seriously unqualified, which provides a detailed signal quality judgment and helps to implement a more intelligent signal adjustment strategy; through the first warning instruction and the second warning instruction, the system can issue a warning when there is a problem with the signal quality, avoiding communication interruption or misunderstanding due to poor signal quality, and improving the system's responsiveness and early warning capabilities. The strategy unit can automatically adjust the gain amplitude according to different signal quality problems. For minor signal quality problems, moderate gain adjustment is performed; for serious signal quality problems, stronger gain adjustment is performed. This strategy ensures the accuracy of signal adjustment, avoids excessive gain or insufficient gain, and maximizes signal quality.

[0157] The threshold is set to facilitate comparison. The size of the threshold depends on the amount of sample data and the number of bases set by technicians in this field for each group of sample data; as long as it does not affect the proportional relationship between the parameter and the quantized value.

[0158] The above formulas are obtained by collecting a large amount of data for software simulation and selecting a formula that is close to the actual value. The coefficients in the formula are set by technical personnel in this field according to actual conditions. The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited to this. Any technical personnel familiar with the technical field within the technical scope disclosed by the present invention, according to the technical solution and the inventive concept of the present invention, make equivalent replacement or change, which should be covered within the protection scope of the present invention.

Claims

1. A long-distance self-organizing network full-duplex voice communication system for fire fighting, characterized in that: It includes a microphone array module, a noise perception module, a noise filtering module, a communication status evaluation module and a signal gain adjustment module; The microphone array module is used to deploy a distributed MEMS microphone array at the fire scene, and each firefighter wears a microphone, and connects each microphone node through self-organizing network technology to achieve full-duplex voice communication; The noise sensing module is used to monitor the background noise data of the fire scene in real time, identify and automatically classify it, and then construct the noise index of the jth microphone position. , if the noise index at the jth microphone position is > Noise threshold, indicating that the noise level interference in the environment of the current j-th microphone position is abnormal, generates a mark point for the noise level interference abnormal area, and issues a low noise suppression instruction; The noise filtering module is used to generate the noise suppression coefficient of the j-th microphone position by using an adaptive filtering algorithm after receiving the low noise suppression instruction. , and after removing the noise component of the first speech signal at the j-th microphone position, generate a filtered second speech signal; The communication status evaluation module is used to collect the real-time distance H between the j-th microphone position and the sending point based on the long-distance transmission requirements, and extract the second voice signal to obtain the voice signal strength at the j-th microphone position. , and associate the real-time distance H of the jth microphone position and the voice signal strength And the noise suppression coefficient , get the communication status evaluation value of the jth microphone position ; The communication status evaluation module includes a real-time distance acquisition unit, a signal strength calculation unit and an associated unit; The real-time distance acquisition unit is used to acquire the real-time distance H between the j-th microphone position and the sending point in real time; The signal strength calculation unit is used to extract the voice signal strength in the second voice signal, and the voice signal strength is calculated by the following formula : ； in, represents the amplitude of the second speech signal at the jth microphone position at the i-th moment, which is acquired through the sound pressure sensor; R is the number of sampling points of the second speech signal; The association unit is used to extract the real-time distance H of the j-th microphone position and the voice signal strength And the noise suppression coefficient , after dimensionless processing, the communication status evaluation value of the jth microphone position is generated by the following formula : ； The meaning of the formula is: Communication status evaluation value Dynamically reflect the signal quality and transmission conditions in the current environment; The signal gain adjustment module evaluates the communication status of the j-th microphone position. Evaluate to determine if the system should enter gain mode to adapt to changing conditions at the fire scene.

2. A fire-fighting long-distance self-organizing network full-duplex voice communication system according to claim 1, characterized in that: The noise perception module includes a distributed acquisition unit and a noise type identification unit; The distributed acquisition unit is used to collect noise data of the fire scene in real time through a distributed MEMS micro-electromechanical microphone array in the fire environment, and the sampling frequency is higher than 32kHz; The noise type identification unit is used to analyze noise signals of multiple frequency bands, and after performing fast Fourier transform, spectrum analysis, constructing feature space and comparison on the spectrum feature set, automatically classifies and obtains each type of noise source.

3. A fire-fighting long-distance self-organizing network full-duplex voice communication system according to claim 2, characterized in that: The noise type identification unit includes an extraction submodule, a classification submodule and a noise index calculation submodule; The extraction submodule is used to extract background noise data and process multi-band noise signals to obtain a noise feature vector F. The specific processing steps are as follows: S11. Extract the time series represented by the collected signal from the noise data , where t is the sampling time, is the noise amplitude value of each frame signal; S12, first perform Fourier transform on the signal to obtain its spectrum, and then calculate the noise amplitude value of each frame signal. , whose spectrum is expressed as , that is, the amplitude at frequency f, is converted to the frequency domain through the fast Fourier transform FFT, and the expression is: ; In the formula, is the complex spectrum of the signal at frequency f, is the signal sample value at discrete time n, from the time domain signal N represents the total number of samples, which controls the frequency resolution. It consists of a negative exponential form, which indicates the phase rotation of different frequency components. j represents the imaginary unit, which is used to represent the phase information in Fourier transform. f represents the frequency variable, which determines the position of the frequency component. n represents the index of the time sample, which indicates the position of each time sample in the signal. S13, extracting multiple features from the spectrum graph, the multiple features including: power spectral density PSD, Mel frequency cepstrum decomposition and zero crossing rate ZCR; S131, based on the complex spectrum of the signal at frequency f , analyze the energy distribution of each frequency component, and construct the power spectral density PSD through the following formula: ； Among them, PSD represents the power spectral density of each frequency, which describes the energy distribution of each frequency component; S132, based on the human hearing characteristics, using the Mel filter bank to filter the complex spectrum of the signal at frequency f Convert to the Mel scale spectrum. The Mel scale is a nonlinear transformation based on the human ear's perception of different frequencies. The Mel frequency cepstral coefficient MFCC formula is: ; In the formula, refers to the center frequency of each filter in the Mel filter bank, Indicates the center frequency The spectral components on represents the frequency response of the kth Mel filter, T is the total number of Mel filters; log is a logarithmic operation, It is to convert the filtered power spectrum into a logarithmic value; S133, zero crossing rate ZCR is used to detect high-frequency components in the signal, indicating the number of times the sign of the signal changes, that is, the number of times the signal crosses from positive to negative or from negative to positive. The formula is expressed as: ； In the formula, is the signal sample value at discrete time n, L is the length of the signal frame, is the indicator function, when When it returns 1, it means the sign changes once, otherwise it returns 0; S14, based on the power spectral density PSD, Mel frequency cepstrum decomposition and zero crossing rate ZCR, a noise feature vector F is formed, which is expressed as: ; S15. Use vector machine (SVM), k nearest neighbor (KNN) or convolutional neural network (CNN) to establish a noise classification model and classify the noise feature vector Perform standardization so that the mean of each dimension is 0 and the variance is 1, and obtain the standardized noise feature vector , expressed as: ; in, is the mean of the noise feature vector, is the standard deviation of the noise feature vector.

4. A fire-fighting long-distance self-organizing network full-duplex voice communication system according to claim 3, characterized in that: The classification submodule is used to transform the standardized noise feature vector Input to the trained noise classification model for binary classification, for each category , create a two-category SVM classifier, The samples are considered as positive and all other samples are considered as negative. And mark the category The samples in the class are +1, and the samples of all other classes are -1. The output of each binary SVM classifier will indicate the input noise feature vector Whether it belongs to this category, count the votes of all classifiers, and take the category with the most votes as the prediction result of the noise type; The prediction results include but are not limited to the following noise types: flame burning sound, explosion sound, building collapse sound, alarm sound, water flow sound and water pump sound, demolition tool noise, rescue equipment noise, gas leakage sound, glass breaking sound, high temperature metal deformation sound, and wind and smoke movement sound.

5. A fire-fighting long-distance self-organizing network full-duplex voice communication system according to claim 3, characterized in that: The noise index calculation submodule is used to extract noise feature vectors at different positions in a distributed MEMS micro-electromechanical microphone array. , the noise feature vector extracted at the jth microphone position contains M noise types, recorded as: ; Where M is the total number of noise types; And by weighted summing the noise feature vectors of each noise type, the noise index at the jth microphone position is obtained by the following formula: : ； In the formula, The weight coefficient representing the Mth noise type is set according to the interference of each frequency band in the noise environment or determined through training; represents the noise feature vector of the g-th noise type at the j-th position, And preset the noise threshold, if the noise index at the jth microphone position > Noise threshold, indicating that the noise level interference in the environment of the current j-th microphone position is abnormal, generates a mark point for the noise level interference abnormal area, and issues a low noise suppression instruction; If the noise index at the jth microphone position is ≤noise threshold, it means that the noise level interference in the environment of the current j-th microphone position is normal and no adjustment is required.

6. A fire-fighting long-distance self-organizing network full-duplex voice communication system according to claim 3, characterized in that: The noise type identification unit also includes a priority submodule; The priority submodule is used to determine the position of the marker point in the abnormal area where the noise level interferes, and the noise index of the j-th microphone position is used to determine the position of the marker point in the abnormal area. Generate the first priority processing queue from large to small.

7. A fire-fighting long-distance self-organizing network full-duplex voice communication system according to claim 6, characterized in that: The noise filtering module includes a noise suppression calculation subunit and an adaptive filtering subunit; The noise suppression calculation subunit is used to extract the noise index of the j-th microphone position after receiving the first priority processing queue. , mapping the noise suppression coefficient of the j-th microphone position : ； In the formula, It is the adjustment factor of noise suppression. The range of the adjustment factor is between 0 and 1. The larger the value, the higher the suppression strength. It is used to control the overall suppression strength. is the sensitivity of the control suppression coefficient to the noise index, is the suppression threshold; exp represents the exponential function, and exp uses the Sigmoid function; The adaptive filtering subunit is used to receive the first speech signal with noise collected by the microphone, divide it into short time frames, and use a 20-30 millisecond window for real-time collection; Initialize the parameters of the adaptive filter, including the filter length and initial weight, and the weight is initialized to 0; Use the noise suppression coefficient for the jth microphone position The filter weight is adjusted to remove the noise component from the first speech signal, and the filter generates a second speech signal after the noise is removed.

8. A fire-fighting long-distance self-organizing network full-duplex voice communication system according to claim 1, characterized in that: The signal gain adjustment module includes a communication state evaluation unit and a strategy unit; The communication state evaluation unit is used to preset a communication threshold X, and to calculate the communication state evaluation value of the j-th microphone position. The communication status is compared with the communication threshold X to obtain the communication status evaluation results, including: If the communication status evaluation value of the jth microphone position ≥communication threshold X, indicating that the communication signal quality at the jth microphone position is qualified; If the communication threshold X*80%≤the communication status evaluation value of the jth microphone position <communication threshold X, indicating that the communication signal quality at the jth microphone position is unqualified, and a first warning instruction is generated; If the communication status evaluation value of the jth microphone position <communication threshold X*80%, indicating that the communication signal quality at the jth microphone position is unqualified, and a second early warning instruction is generated.

9. A fire-fighting long-distance self-organizing network full-duplex voice communication system according to claim 8, characterized in that: The strategy unit is used to generate corresponding instructions when receiving the first warning instruction and the second warning instruction, including: Generate a first gain adjustment instruction according to the first warning instruction, including: performing signal gain through a digital signal processor to increase the strength of the voice signal 10-30%; Generating a second gain adjustment instruction according to the second warning instruction includes: performing signal gain through a digital signal processor to increase the strength of the voice signal If a signal interruption is detected, the system will automatically detect the signal interruption location, recalculate the best signal path, select an alternative path and restore the communication connection within 30 seconds.

Images