Dual-mode interphone speech enhancement processing method and system based on AI recognition

Abstract

The invention discloses a dual-mode interphone speech enhancement processing method and system based on AI identification. The method comprises the following steps: acquiring working environment acoustic data and historical communication acoustic data of the dual-mode interphone in a target working area, and analyzing communication quality distribution; collecting communication behavior data in a preset time period, and performing clustering by using a clustering algorithm to generate a voice communication behavior density map; constructing a regionalized speech enhancement model in combination with the communication quality distribution and density map; acquiring voice communication incremental data of different positions in real time, and judging an emergency state according to the model; a corresponding voice processing strategy is made according to the emergency state, and the voice definition and reliability of the interphone in a complex or noise environment are improved. According to the method, the emergency response efficiency and the communication quality can be improved through AI intelligent identification and regionalized modeling, and the method is suitable for a dual-mode interphone communication system in various high-noise or emergency operation environments such as mines, electric power and fire fighting.

Description

Technical Field

[0001] The present invention relates to the field of voice processing technology, and in particular to a dual-mode intercom voice enhancement processing method and system based on AI recognition. Background Art

[0002] Dual-mode walkie-talkies, as integrated analog and digital communication devices, are widely used in on-site operations and emergency command and dispatch in complex environments such as mining, power generation, firefighting, and rail transit. With the diversification of operating environments and the increasing demand for communication reliability, traditional dual-mode walkie-talkies suffer from low voice clarity, limited noise suppression, and low command recognition accuracy in scenarios with high noise and strong interference, making them difficult to meet the needs of emergency response and multi-party collaborative communications.

[0003] Existing speech enhancement technologies primarily rely on fixed noise suppression algorithms, such as spectral subtraction and Wiener filtering. These algorithms lack comprehensive modeling of multi-source noise, dynamic acoustic characteristics, and communication behavior patterns within the operational area, making it difficult to adapt speech enhancement to different environments and communication loads. Furthermore, existing technologies typically rely on a single signal-to-noise ratio or energy threshold for noise suppression and speech enhancement, failing to flexibly adjust speech processing strategies based on differences in communication quality and communication activity density at different locations. This leads to frequent problems such as command output conflicts, speech distortion, and delays in emergency communications.

[0004] With the rapid development of artificial intelligence technology, especially deep learning, cluster analysis, and convolutional neural networks, environmental acoustic analysis, communication behavior modeling, and regionalized speech enhancement based on AI recognition have become important means to improve the communication quality of walkie-talkies and the efficiency of emergency response. However, there is currently a lack of a method and system that can integrate multi-dimensional acoustic data, historical communication behavior, and real-time emergency status to dynamically construct a regionalized speech enhancement model and generate targeted speech processing strategies. Therefore, it is urgent to propose a dual-mode walkie-talkie speech enhancement processing method and system based on AI recognition to address the shortcomings of existing technologies and achieve intelligent enhancement of walkie-talkie communication quality and reliability assurance of emergency command in high-noise environments. Summary of the Invention

[0005] In order to solve at least one of the above technical problems, the present invention proposes a dual-mode intercom voice enhancement processing method and system based on AI recognition.

[0006] The first aspect of the present invention provides a dual-mode intercom voice enhancement processing method based on AI recognition, comprising: Acquiring working environment acoustic data and historical communication acoustic data of the dual-mode walkie-talkie in the target operating area, and determining the communication quality distribution of the target operating area based on the working environment acoustic data and the historical communication acoustic data; Acquire communication behavior data of the dual-mode walkie-talkie within a preset time period, perform clustering operations on the communication behavior data based on a clustering algorithm, and construct a voice communication behavior density map; Determine and construct a regionalized speech enhancement model for the target operation area based on the communication quality distribution and the speech communication behavior density map; Acquire incremental data on voice communication behavior at different locations in the target operation area in real time, and determine the emergency status of different locations based on the incremental data and a regionalized voice enhancement model; A speech processing strategy for the target operation area is constructed according to the emergency state.

[0007] In this solution, the working environment acoustic data and historical communication acoustic data of the dual-mode walkie-talkie in the target operating area are obtained, and the communication quality distribution of the target operating area is determined based on the working environment acoustic data and historical communication acoustic data. Specifically, Obtain historical intercom record data of the dual-mode intercom in the target operation area, obtain working environment acoustic data and historical communication acoustic data of the dual-mode intercom in the target operation area based on the historical intercom record data, and obtain location information of the dual-mode intercom during communication; Converting the work environment acoustic data into a frequency domain signal based on a fast Fourier transform, extracting the Mel-frequency cepstral coefficients, spectral centroid, spectral entropy reduction, zero-crossing rate, and energy envelope of the work environment acoustic data according to the frequency domain signal, and constructing a feature vector of the work environment acoustic data; Building a work environment noise recognition model based on a convolutional neural network, importing the work environment acoustic data feature vector into the work environment noise recognition model to identify the work environment noise of the dual-mode intercom, and obtaining a work environment noise type recognition result; Determine the noise intensity of the working environment noise type according to the energy envelope, and determine the noise position according to the position information during the communication process of the dual-mode walkie-talkie to obtain noise data; Perform audio framing and windowing on historical communication acoustic data. Use a blind source separation algorithm to demix and extract the communication voice signal from the background noise. Calculate the energy values ​​of each frame of voice signal and the corresponding background noise, and generate a signal-to-noise ratio curve based on the ratio of voice energy to noise energy. determining speech clarity of historical communication acoustic data at different locations in the target operation area based on the signal-to-noise ratio curve and the location information; The noise data and speech clarity at different locations are spatially aligned. The communication quality at different locations is evaluated based on the aligned noise data and speech clarity. A communication quality distribution map is constructed to obtain the communication quality distribution of the target operation area.

[0008] In this solution, the communication behavior data of the dual-mode walkie-talkie within a preset time period is obtained, the communication behavior data is clustered based on a clustering algorithm, and a voice communication behavior density map is constructed, specifically: Acquire the communication behavior data of the dual-mode walkie-talkie within a preset time period, wherein the communication behavior data includes a communication timestamp, communication location information, and call frequency; Constructing a spatiotemporal coordinate sequence based on the communication timestamp and communication location information, generating a multidimensional feature vector set in combination with the call frequency, initializing density clustering algorithm parameters, setting the neighborhood radius eps and the minimum number of neighborhood points min_samples; Based on the density clustering algorithm, the Euclidean distance between all data points in the multidimensional feature vector set is calculated to determine the spatial topological relationship. Each data point in the multidimensional feature vector set is traversed. If the number of data points in the eps neighborhood of the current data point is not less than min_samples, it is marked as a core point, and each core point is marked as unvisited. Create a new cluster for each unvisited core point, obtain all data points in the neighborhood of each core point eps and store them in a queue, recursively visit each data point in the queue, and when the data point that is visited is not marked by any cluster, it is added to the current cluster. If the data point also meets the core point conditions, its neighboring data points are expanded and added to the queue until all data points are visited. The unclustered data points are marked as noise points, and the cluster assignment result is generated; Based on the cluster allocation results, the coverage area of ​​each cluster in the target operation area is calculated. The spatial density value is obtained according to the ratio of the number of data points in the cluster to the area of ​​the target operation area. The spatial density value is mapped to the corresponding two-dimensional plane area of ​​the target operation area, and a voice communication behavior density map of different locations in the target operation area within a preset time period is constructed.

[0009] In this solution, the regionalized speech enhancement model for the target operation area is determined based on the communication quality distribution and the voice communication behavior density map, specifically: Mapping the speech clarity data and noise data at different locations in the communication quality distribution into a speech communication behavior density map, and calculating the abnormal communication probability at each location in the target operation area based on the speech clarity data, noise data and communication behavior density at each location; If the probability of abnormal communication at a certain location is greater than a preset value, the area where the location is located is marked as a voice enhancement demand area, and noise signal data of each voice enhancement demand area is extracted based on the working environment acoustic data and historical communication acoustic data; Calculating the energy distribution of the noise signal data in a preset frequency band, identifying a noise-dominant frequency band whose energy exceeds a preset energy threshold, and calculating a noise suppression frequency band parameter based on the center frequency and bandwidth of the noise-dominant frequency band; Calculating statistical characteristics of the energy envelope of the noise signal data, including peak value, mean value, and variance; when the peak value of the energy envelope exceeds a preset peak value threshold, calculating an instantaneous gain suppression amplitude parameter based on the peak value; when the mean value of the energy envelope exceeds a preset mean value threshold, calculating a steady-state gain suppression amplitude parameter based on the mean value; and fusing the instantaneous gain suppression amplitude parameter and the steady-state gain suppression amplitude parameter to generate a noise suppression gain parameter; Determining a speech enhancement response sensitivity parameter according to a speech communication behavior density value of each speech enhancement demand area; Noise data of each area requiring speech enhancement is obtained, and a noise feature matrix is ​​constructed by combining the noise data with noise suppression frequency band parameters and noise suppression gain parameters. The noise feature matrix is ​​input into a BP neural network for parameter learning to construct a regionalized speech enhancement model. The response sensitivity of the speech enhancement processing parameters output by the regionalized speech enhancement model in different areas is determined according to the speech enhancement response sensitivity parameters.

[0010] In this solution, the real-time acquisition of incremental data on voice communication behavior at different locations in the target operation area and the determination of the emergency status of different locations based on the incremental data and the regionalized voice enhancement model are as follows: Acquire voice communication behavior increment data at different locations in the target operation area in real time, mark areas where voice communication behavior increments are greater than a preset increment value as emergency risk areas, and acquire real-time voice communication data for the emergency risk areas; extracting real-time noise data of the real-time voice communication data, inputting the real-time noise data into a regionalized voice enhancement model for matching, and determining voice enhancement processing parameters of the real-time voice communication data under the current real-time noise data; Performing voice enhancement processing on the real-time voice communication data according to the voice enhancement processing parameters, performing semantic analysis on the real-time call voice data after the voice enhancement processing, and converting it into voice text data; The emergency status of the emergency risk area is determined based on the voice and text data.

[0011] In this solution, the speech processing strategy for the target operation area constructed according to the emergency state is specifically as follows: Obtaining an emergency state calibration result of the emergency risk area within a preset time period according to the emergency state, calibrating the emergency risk area where the emergency state exists as an emergency area according to the emergency state calibration result, and determining the emergency state level of the emergency area; Generate estimated concurrent communication instruction quantity data based on the emergency state level and incremental voice communication behavior data of the emergency area, compare the estimated concurrent communication instruction quantity with a historical concurrency threshold, and when the estimated concurrent communication instruction quantity exceeds the historical concurrency threshold, determine that there is a command output conflict risk in the current emergency area, and extract real-time voice communication data of the emergency area under the command output conflict risk state; When the estimated number of concurrent communication instructions does not exceed the historical concurrency threshold, the cumulative probability of potential conflict is calculated based on the emergency state level and the incremental data of voice communication behavior. If the cumulative probability of potential conflict exceeds the preset conflict probability threshold, the area is determined to be a high-incidence area of ​​potential conflict, and the incremental data of voice communication behavior in the subsequent time window of the area is obtained in real time; Constructing a conflict risk assessment matrix based on the real-time voice communication data in the command output conflict risk state or the incremental voice communication behavior data in the potential conflict high-incidence area, and determining the command output conflict risk value of the emergency area based on the conflict risk assessment matrix; The command output priority areas within the target operation area are divided according to the command output conflict risk value and the emergency state level, and the resource allocation ratio of the regionalized speech enhancement model to the emergency area is adjusted to obtain the speech processing strategy of the target operation area.

[0012] A second aspect of the present invention further provides a dual-mode intercom voice enhancement processing system based on AI recognition, the system comprising: a memory, a processor, the memory comprising a dual-mode intercom voice enhancement processing method program based on AI recognition, and the dual-mode intercom voice enhancement processing method program based on AI recognition, when executed by the processor, implements the following steps: Acquiring working environment acoustic data and historical communication acoustic data of the dual-mode walkie-talkie in the target operating area, and determining the communication quality distribution of the target operating area based on the working environment acoustic data and the historical communication acoustic data; Acquire communication behavior data of the dual-mode walkie-talkie within a preset time period, perform clustering operations on the communication behavior data based on a clustering algorithm, and construct a voice communication behavior density map; Determine and construct a regionalized speech enhancement model for the target operation area based on the communication quality distribution and the speech communication behavior density map; Acquire incremental data on voice communication behavior at different locations in the target operation area in real time, and determine the emergency status of different locations based on the incremental data and a regionalized voice enhancement model; A speech processing strategy for the target operation area is constructed according to the emergency state.

[0013] The present invention discloses a dual-mode walkie-talkie voice enhancement processing method and system based on AI recognition. The method includes: obtaining the working environment acoustic data and historical communication acoustic data of the dual-mode walkie-talkie in the target operation area, and analyzing the communication quality distribution; collecting communication behavior data within a preset time period, clustering using a clustering algorithm, and generating a voice communication behavior density map; combining the communication quality distribution and the density map to construct a regionalized voice enhancement model; obtaining voice communication incremental data at different locations in real time, and judging the emergency status according to the model; formulating corresponding voice processing strategies according to the emergency status to achieve improved voice clarity and reliability of the walkie-talkie in complex or noisy environments. This method can improve emergency response efficiency and communication quality through AI intelligent recognition and regionalized modeling, and is suitable for dual-mode walkie-talkie communication systems in various high-noise or emergency operation environments such as mining, electricity, and fire fighting. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The flowchart of the present invention shows a dual-mode intercom voice enhancement processing method based on AI recognition; Figure 2 A flow chart showing the present invention for determining emergency status at different locations; Figure 3 A flow chart showing a speech processing strategy constructed in the present invention is shown; Figure 4 The block diagram of the dual-mode intercom voice enhancement processing system based on AI recognition of the present invention is shown. DETAILED DESCRIPTION

[0015] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0016] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0017] Figure 1 The flowchart of the present invention shows a dual-mode intercom voice enhancement processing method based on AI recognition.

[0018] like Figure 1 As shown, the first aspect of the present invention provides a dual-mode intercom voice enhancement processing method based on AI recognition, comprising: S102, obtaining working environment acoustic data and historical communication acoustic data of the dual-mode walkie-talkie in the target working area, and determining the communication quality distribution of the target working area based on the working environment acoustic data and the historical communication acoustic data; S104, acquiring communication behavior data of the dual-mode walkie-talkie within a preset time period, performing a clustering operation on the communication behavior data based on a clustering algorithm, and constructing a voice communication behavior density map; S106, determining and constructing a regionalized speech enhancement model for the target operation area based on the communication quality distribution and the speech communication behavior density map; S108, acquiring incremental data of voice communication behaviors at different locations in the target operation area in real time, and determining the emergency status of the different locations based on the incremental data and a regionalized voice enhancement model; S110: Construct a speech processing strategy for the target operation area according to the emergency state.

[0019] It should be noted that by analyzing the communication quality distribution of the working environment acoustic data and historical communication acoustic data of the dual-mode walkie-talkie in the target operation area, and based on the cluster analysis of the communication behavior data within the preset time period, it is possible to effectively identify high-frequency communication areas and call-intensive periods, and construct a voice communication behavior density map that reflects the laws of communication behavior; based on the communication quality distribution and the voice communication behavior density map, it is possible to simultaneously determine areas with severe noise interference or high communication demand in both spatial and behavioral dimensions, and to construct a targeted regional voice enhancement model to improve noise suppression accuracy and voice enhancement adaptability; by acquiring incremental voice communication behavior data at different locations in real time, it is possible to dynamically identify emergency risk areas with abnormal growth in communication behavior, and combine the regional voice enhancement model to judge the emergency state. When an emergency state exists, a voice processing strategy for the target operation area is generated, which can prioritize the allocation of voice enhancement resources in the emergency area, dynamically adjust the gain suppression parameters and command output priority, and achieve high reliability of voice communication and high response efficiency of emergency command. The AI ​​recognition includes a convolutional neural network for noise recognition, a density clustering algorithm for communication behavior analysis, a blind source separation algorithm for voice and noise demixing, a BP neural network for regional voice enhancement modeling, and a semantic analysis algorithm for emergency state recognition.

[0020] According to an embodiment of the present invention, the obtaining of working environment acoustic data and historical communication acoustic data of the dual-mode intercom in the target operating area, and determining the communication quality distribution of the target operating area based on the working environment acoustic data and the historical communication acoustic data, is specifically as follows: Obtain historical intercom record data of the dual-mode intercom in the target operation area, obtain working environment acoustic data and historical communication acoustic data of the dual-mode intercom in the target operation area based on the historical intercom record data, and obtain location information of the dual-mode intercom during communication; Converting the work environment acoustic data into a frequency domain signal based on a fast Fourier transform, extracting the Mel-frequency cepstral coefficients, spectral centroid, spectral entropy reduction, zero-crossing rate, and energy envelope of the work environment acoustic data according to the frequency domain signal, and constructing a feature vector of the work environment acoustic data; Building a work environment noise recognition model based on a convolutional neural network, importing the work environment acoustic data feature vector into the work environment noise recognition model to identify the work environment noise of the dual-mode intercom, and obtaining a work environment noise type recognition result; Determine the noise intensity of the working environment noise type according to the energy envelope, and determine the noise position according to the position information during the communication process of the dual-mode walkie-talkie to obtain noise data; Perform audio framing and windowing on historical communication acoustic data. Use a blind source separation algorithm to demix and extract the communication voice signal from the background noise. Calculate the energy values ​​of each frame of voice signal and the corresponding background noise, and generate a signal-to-noise ratio curve based on the ratio of voice energy to noise energy. determining speech clarity of historical communication acoustic data at different locations in the target operation area based on the signal-to-noise ratio curve and the location information; The noise data and speech clarity at different locations are spatially aligned. The communication quality at different locations is evaluated based on the aligned noise data and speech clarity. A communication quality distribution map is constructed to obtain the communication quality distribution of the target operation area.

[0021] It should be noted that due to the significant differences in noise type, noise intensity, and historical voice clarity at different locations, the accuracy and effectiveness of the dual-mode walkie-talkie voice enhancement strategy are directly affected. Fast Fourier transform is used to convert the raw ambient acoustic data into frequency domain signals. Multi-dimensional acoustic features such as Mel-frequency cepstral coefficients, spectral centroid, spectral entropy reduction, zero-crossing rate, and energy envelope are extracted to comprehensively characterize the working environment noise characteristics. Secondly, a convolutional neural network is used to intelligently classify and identify noise types, and the energy envelope is used to quantify noise intensity and spatial location to generate accurate noise data. Audio framing and blind source separation are then performed on the historical communication acoustic data to accurately separate the voice signal from the background noise. The ratio of voice to noise energy in each frame is calculated to generate a signal-to-noise ratio curve, which objectively reflects the voice clarity at each location. Subsequently, the noise data and voice clarity are spatially aligned to achieve a comprehensive quantitative assessment of communication quality at different locations. Finally, a communication quality distribution map is constructed. The working environment acoustic data and historical communication acoustic data are both audio signal data; the working environment noise recognition model includes a convolutional layer, a pooling layer and a fully connected layer, and a model with noise recognition capability is obtained by training the model by obtaining different types of noise features.

[0022] According to an embodiment of the present invention, the communication behavior data of the dual-mode walkie-talkie within a preset time period is obtained, and the communication behavior data is clustered based on a clustering algorithm to construct a voice communication behavior density map, specifically: Acquire the communication behavior data of the dual-mode walkie-talkie within a preset time period, wherein the communication behavior data includes a communication timestamp, communication location information, and call frequency; Constructing a spatiotemporal coordinate sequence based on the communication timestamp and communication location information, generating a multidimensional feature vector set in combination with the call frequency, initializing density clustering algorithm parameters, setting the neighborhood radius eps and the minimum number of neighborhood points min_samples; Based on the density clustering algorithm, the Euclidean distance between all data points in the multidimensional feature vector set is calculated to determine the spatial topological relationship. Each data point in the multidimensional feature vector set is traversed. If the number of data points in the eps neighborhood of the current data point is not less than min_samples, it is marked as a core point, and each core point is marked as unvisited. Create a new cluster for each unvisited core point, obtain all data points in the neighborhood of each core point eps and store them in a queue, recursively visit each data point in the queue, and when the data point that is visited is not marked by any cluster, it is added to the current cluster. If the data point also meets the core point conditions, its neighboring data points are expanded and added to the queue until all data points are visited. The unclustered data points are marked as noise points, and the cluster assignment result is generated; Based on the cluster allocation results, the coverage area of ​​each cluster in the target operation area is calculated. The spatial density value is obtained according to the ratio of the number of data points in the cluster to the area of ​​the target operation area. The spatial density value is mapped to the corresponding two-dimensional plane area of ​​the target operation area, and a voice communication behavior density map of different locations in the target operation area within a preset time period is constructed.

[0023] It should be noted that the timestamp and geographic location information of each communication behavior of the dual-mode walkie-talkie within a preset time period are converted into a spatiotemporal coordinate sequence, and call frequency is included as a key behavioral feature, together forming a multidimensional feature vector set that reflects the communication intensity and spatiotemporal distribution pattern. Next, the neighborhood radius and minimum neighborhood point threshold of the density clustering algorithm are initialized. The spatial topological relationship is established by calculating the Euclidean distance between all data points. The core point is recursively marked and the neighborhood is expanded with it as the center. Data points with similar spatial locations and continuous communication behavior patterns (such as high-frequency call areas or active points during specific time periods) are automatically merged into the same cluster. At the same time, noise points are filtered to eliminate sporadic communication interference. After obtaining the cluster allocation results, the spatial coverage area of ​​each cluster in the target operation area is further calculated. The ratio of the total number of communication behaviors within the cluster to the area of ​​the corresponding cluster area is converted into a spatial density value. Finally, these density values ​​are mapped to a two-dimensional spatial model of the target operation area, forming a voice communication behavior density map that intuitively displays the communication behavior cluster hotspots within the preset time period with different colors or intensities, thereby revealing the spatiotemporal distribution pattern of communication demand.

[0024] According to an embodiment of the present invention, determining and constructing a regionalized speech enhancement model for a target operation area based on the communication quality distribution and the voice communication behavior density map is specifically as follows: Mapping the speech clarity data and noise data at different locations in the communication quality distribution into a speech communication behavior density map, and calculating the abnormal communication probability at each location in the target operation area based on the speech clarity data, noise data and communication behavior density at each location; If the probability of abnormal communication at a certain location is greater than a preset value, the area where the location is located is marked as a voice enhancement demand area, and noise signal data of each voice enhancement demand area is extracted based on the working environment acoustic data and historical communication acoustic data; Calculating the energy distribution of the noise signal data in a preset frequency band, identifying a noise-dominant frequency band whose energy exceeds a preset energy threshold, and calculating a noise suppression frequency band parameter based on the center frequency and bandwidth of the noise-dominant frequency band; Calculating statistical characteristics of the energy envelope of the noise signal data, including peak value, mean value, and variance; when the peak value of the energy envelope exceeds a preset peak value threshold, calculating an instantaneous gain suppression amplitude parameter based on the peak value; when the mean value of the energy envelope exceeds a preset mean value threshold, calculating a steady-state gain suppression amplitude parameter based on the mean value; and fusing the instantaneous gain suppression amplitude parameter and the steady-state gain suppression amplitude parameter to generate a noise suppression gain parameter; Determining a speech enhancement response sensitivity parameter according to a speech communication behavior density value of each speech enhancement demand area; Noise data of each area requiring speech enhancement is obtained, and a noise feature matrix is ​​constructed by combining the noise data with noise suppression frequency band parameters and noise suppression gain parameters. The noise feature matrix is ​​input into a BP neural network for parameter learning to construct a regionalized speech enhancement model. The response sensitivity of the speech enhancement processing parameters output by the regionalized speech enhancement model in different areas is determined according to the speech enhancement response sensitivity parameters.

[0025] It should be noted that, since the noise characteristics affecting the dual-mode walkie-talkie voice at different locations in the target operating area are relatively fixed and predictable in most cases, such as the continuous narrowband noise generated by the operation of mechanical equipment, periodic impact noise or specific environmental background noise, the spatial distribution and spectral energy distribution vary little, so this claim first extracts noise signal data at different locations in advance through multiple collection and analysis of historical working environment acoustic data and historical communication acoustic data, and calculates the energy distribution of the noise signal in the preset frequency band at each location, identifies the noise dominant frequency band, and extracts the energy envelope statistical characteristics (including peak value, mean, variance), and then generates multi-dimensional voice enhancement processing parameters such as noise suppression frequency band parameters, instantaneous gain suppression amplitude parameters, and steady-state gain suppression amplitude parameters for the noise characteristics of each location; then, by comparing the noise data at these different locations with the corresponding generated voice The enhancement processing parameters together form a training sample input into a BP neural network for supervised learning. The BP neural network is used to deeply fit and update the weights of the nonlinear mapping relationship between noise characteristics and optimal speech enhancement processing parameters. After training, the neural network model is able to automatically infer the speech enhancement processing parameters for the corresponding area based on the real-time noise data input. During subsequent real-time communication, the system only needs to collect real-time noise data from the current communication location within the target operation area and directly input it into the trained BP neural network model, which outputs speech enhancement processing parameters that closely match the noise characteristics in real time. Without the need for complex spectrum analysis and parameter estimation, the optimal speech enhancement processing parameters can be quickly obtained directly from the current noise data during real-time communication, significantly reducing computational latency and speech output delay between dual-mode intercoms, greatly improving the success rate of emergency rescue in emergency rescue scenarios. The constructed regionalized speech enhancement model can be configured in the dual-mode intercom or in the speech processing center of the target operation area.

[0026] Figure 2 A flow chart of determining emergency status at different locations according to the present invention is shown.

[0027] According to an embodiment of the present invention, the real-time acquisition of incremental data on voice communication behaviors at different locations in the target operation area and the determination of the emergency status of the different locations based on the incremental data and the regionalized voice enhancement model are specifically as follows: S202, acquiring incremental voice communication behavior data at different locations in the target operation area in real time, marking areas where the voice communication behavior increment is greater than a preset increment value as emergency risk areas, and acquiring real-time voice communication data for the emergency risk areas; S204, extracting real-time noise data of the real-time voice communication data, inputting the real-time noise data into a regionalized voice enhancement model for matching, and determining voice enhancement processing parameters of the real-time voice communication data under the current real-time noise data; S206, performing voice enhancement processing on the real-time voice communication data according to the voice enhancement processing parameters, performing semantic analysis on the real-time call voice data after the voice enhancement processing, and converting it into voice text data; S208: Determine the emergency status of the emergency risk area according to the voice text data.

[0028] It should be noted that by acquiring incremental data on voice communication behavior at different locations in real time, areas with significantly abnormal communication behavior are accurately identified and marked as emergency risk areas. Real-time noise data for the corresponding areas is further extracted and input into a regionalized voice enhancement model for matching. The voice enhancement processing parameters most suitable for the current noise environment are quickly generated, and the real-time voice communication data is enhanced to improve voice clarity. Then, through semantic analysis of the enhanced voice, it is converted into text data to achieve accurate recognition and judgment of the content of emergency calls, thereby enabling timely and accurate determination of the emergency status of each emergency risk area, effectively improving the reliability and response efficiency of emergency command and dispatch. If the voice enhancement processing parameters cannot be matched, it means that the noise is not sufficient to affect the voice. At this time, the voice enhancement processing operation is not required, saving system resources.

[0029] Figure 3 A flow chart of constructing a speech processing strategy according to the present invention is shown.

[0030] According to an embodiment of the present invention, the speech processing strategy for constructing the target operation area according to the emergency state is specifically as follows: S302: Obtain emergency status calibration results for emergency risk areas within a preset time period based on the emergency status, calibrate the emergency risk areas where the emergency status exists as emergency areas based on the emergency status calibration results, and determine the emergency status level of the emergency areas; S304: generating an estimated number of concurrent communication instructions based on the emergency state level and incremental voice communication behavior data in the emergency area, comparing the estimated number of concurrent communication instructions with a historical concurrency threshold, and determining that a command output conflict risk exists in the current emergency area when the estimated number of concurrent communication instructions exceeds the historical concurrency threshold, and extracting real-time voice communication data of the emergency area in the command output conflict risk state; S306: When the estimated number of concurrent communication instructions does not exceed the historical concurrency threshold, the cumulative probability of potential conflicts is calculated based on the emergency state level and the incremental data of voice communication behaviors. If the cumulative probability of potential conflicts exceeds the preset conflict probability threshold, the area is determined to be a high-incidence area of ​​potential conflicts, and incremental data of voice communication behaviors in the area in a subsequent time window is obtained in real time. S308, constructing a conflict risk assessment matrix based on the real-time voice communication data in the command output conflict risk state or the incremental voice communication behavior data in the potential conflict high-incidence area, and determining the command output conflict risk value of the emergency area based on the conflict risk assessment matrix; S310 , dividing the command output priority areas within the target operation area according to the command output conflict risk value and the emergency state level, and adjusting the resource allocation ratio of the regionalized speech enhancement model to the emergency area to obtain a speech processing strategy for the target operation area.

[0031] It should be noted that, in an emergency state, multiple concurrent communication instructions in the target operation area may lead to problems such as command output conflicts, information confusion, and imbalanced resource allocation. By analyzing the number of concurrent communication instructions in the emergency area where the emergency state exists, when the estimated number of concurrent communication instructions exceeds the historical concurrent threshold, it means that the communication load in the current emergency area is too high, and there is a risk of command output conflicts, which may lead to command confusion or delays, affecting the efficiency of emergency response. In response to this situation, extracting real-time voice communication data of the emergency area in a command output conflict risk state can timely monitor and analyze the communication status; when the estimated number of concurrent communication instructions does not exceed the historical concurrent threshold, it means that the communication load in the current emergency area is still within a controllable range, and there is no obvious conflict risk in the command output for the time being; at this time, by calculating the cumulative probability of potential conflicts, it is possible to identify high-incidence areas of conflicts in advance, monitor the increment of voice communication behavior in real time, and issue timely warnings of potential risks to prevent the occurrence of communication conflicts. By constructing a conflict risk assessment matrix, the conflict risk of command output within the emergency zone is quantified, accurately assessing the communication pressure and potential conflicts in different areas. Command output priority zones are divided based on conflict risk values ​​and emergency status levels, effectively optimizing resource allocation and improving the responsiveness of the voice enhancement model in key areas. This ensures the clarity and stability of communication in high-priority areas, thereby enabling dynamic regulation and intelligent management of voice communications within the target operating area and ensuring the timely and accurate delivery of emergency commands. The cumulative probability of potential conflicts is calculated by statistically analyzing incremental historical and real-time voice communication behavior data within the emergency zone, combining the temporal overlap rate and spatial distribution characteristics of communication commands, and using a probabilistic model to comprehensively calculate the frequency and cumulative trend of command conflict events. The conflict risk assessment matrix collects real-time voice communication data and incremental voice communication behavior data within the emergency zone, combines the temporal, spatial, and content relevance of communication commands, and uses multidimensional feature vectors for quantitative analysis and normalization to construct a matrix representation reflecting the probability and impact of conflicts between different commands. The command output priority zones are weighted by the command output conflict risk value and emergency status level to ultimately determine the voice communication command output priority for the dual-mode radios in each emergency zone.

[0032] According to an embodiment of the present invention, the further embodiment includes: Obtaining physical space structure vector data of the target operation area, extracting obstacle boundary coordinate sets and connected path topology sequences based on the physical space structure vector data, and constructing a spatial attribute adjacency matrix; Obtain a communication behavior dataset of a dual-mode walkie-talkie, extract the communication location point coordinate sequence and timestamp interval data, and calculate the rate of change of the location point movement direction based on the timestamp interval data; Mapping the communication location point coordinate sequence to the spatial attribute adjacency matrix to generate a spatial node embedding vector, and fusing the location point moving direction change rate to construct a dynamic behavior feature tensor; Building a spatial topology encoder based on a graph convolutional network, inputting the dynamic behavior feature tensor into the spatial topology encoder for multi-layer message passing operations to generate a spatial-behavior fusion feature graph; Calculating a path constraint weight matrix according to the connected path topology sequence, performing attention weighted aggregation on the space-behavior fusion feature map using the path constraint weight matrix to generate a path normalized density distribution; Based on the obstacle boundary coordinate set, boundary reflection compensation calculation is performed on the path normalized density distribution to construct a communication behavior density correction map of the target operation area.

[0033] According to an embodiment of the present invention, the attention-weighted aggregation of the space-behavior fusion feature map is performed through the path constraint weight matrix to generate a path normalized density distribution, specifically: parsing the branch node coordinates in the connected path topology sequence, and dividing the target operation area into a set of linear path segments according to the branch node coordinates; Calculate the direction vector and length ratio parameters of each linear path segment, and generate the path segment permeability coefficient by combining the spatial attribute adjacency matrix; Extracting the initial density estimation value of each spatial node from the spatial-behavior fusion feature map, and constructing the density diffusion transfer equation according to the permeability coefficient of the path segment; Constructing a loss function including an obstacle penetration penalty term and a path connectivity gain term, and optimizing the parameter weights of the density diffusion transfer equation through back propagation; Acquire incremental data of communication position points in real time, generate a position offset matrix based on the incremental data, and perform a tensor multiplication operation on the position offset matrix and the optimized density diffusion transfer equation; A direction-aligned integral operation is performed on the calculation results along the connected path topology sequence to generate a path normalized density distribution that eliminates spatial distortion.

[0034] It should be noted that in emergency rescue scenarios with restricted physical topology, such as mine tunnels and underground pipeline corridors, the narrow linear channel structure causes serious distortion in traditional communication behavior density calculation methods. First, the effective coordinate distribution range of tunnel intersections is compressed due to the actual physical space constraints, causing high-frequency calls to be weakened as low-value artifacts in the density map. Second, the linear movement trajectories of rescue teams are fragmented into discrete segments due to spatial continuity constraints, causing the density distribution map to exhibit discontinuous patchy distortion. Therefore, we first construct a spatial attribute adjacency matrix based on the connected path topology sequence to accurately quantify the curvature and branching structure characteristics of the tunnels, making the density calculation model physically aware of the space. Second, through the multi-layer message passing mechanism of the graph convolutional network, the rate of change of the movement direction of the location point is dynamically encoded into the space-behavior fusion feature map, ensuring the spatiotemporal continuity of the movement trajectory. Finally, combining the attention weighting of the path constraint weight matrix with boundary reflection compensation, we achieve density compensation in the intersection area and suppress density fluctuations in the linear path segments. This generates a corrected communication behavior density map that eliminates spatial distortion and provides a spatially topologically realistic density benchmark for subsequent calculation of abnormal communication probability. The physical space structure vector data includes a set of three-dimensional coordinate points of the lane, a sequence of vertex points of obstacle boundary polygons, a topological connection relationship of the channel centerline, a set of marking points of ventilation shafts and intersections, and a polygon for dividing a safety area.

[0035] Figure 4 The block diagram of the dual-mode intercom voice enhancement processing system based on AI recognition of the present invention is shown.

[0036] The second aspect of the present invention further provides a dual-mode intercom voice enhancement processing system 4 based on AI recognition, which includes: a memory 41 and a processor 42. The memory includes a dual-mode intercom voice enhancement processing method program based on AI recognition. When the dual-mode intercom voice enhancement processing method program based on AI recognition is executed by the processor, the following steps are implemented: Acquiring working environment acoustic data and historical communication acoustic data of the dual-mode walkie-talkie in the target operating area, and determining the communication quality distribution of the target operating area based on the working environment acoustic data and the historical communication acoustic data; Acquire communication behavior data of the dual-mode walkie-talkie within a preset time period, perform clustering operations on the communication behavior data based on a clustering algorithm, and construct a voice communication behavior density map; Determine and construct a regionalized speech enhancement model for the target operation area based on the communication quality distribution and the speech communication behavior density map; Acquire incremental data on voice communication behavior at different locations in the target operation area in real time, and determine the emergency status of different locations based on the incremental data and a regionalized voice enhancement model; A speech processing strategy for the target operation area is constructed according to the emergency state.

[0037] The present invention discloses a dual-mode walkie-talkie voice enhancement processing method and system based on AI recognition. The method includes: obtaining the working environment acoustic data and historical communication acoustic data of the dual-mode walkie-talkie in the target operation area, and analyzing the communication quality distribution; collecting communication behavior data within a preset time period, clustering using a clustering algorithm, and generating a voice communication behavior density map; combining the communication quality distribution and the density map to construct a regionalized voice enhancement model; obtaining voice communication incremental data at different locations in real time, and judging the emergency status according to the model; formulating corresponding voice processing strategies according to the emergency status to achieve improved voice clarity and reliability of the walkie-talkie in complex or noisy environments. This method can improve emergency response efficiency and communication quality through AI intelligent recognition and regionalized modeling, and is suitable for dual-mode walkie-talkie communication systems in various high-noise or emergency operation environments such as mining, electricity, and fire fighting.

[0038] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0039] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0040] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0041] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0042] Alternatively, if the integrated units described above are implemented as software modules and sold or used as standalone products, they can also be stored on a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present invention, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product, stored on a storage medium, includes instructions for enabling a computer device (such as a personal computer, server, or network device) to execute all or part of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as removable storage devices, ROM, RAM, magnetic disks, or optical disks.

[0043] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A dual-mode intercom voice enhancement processing method based on AI recognition, characterized in that: The following steps are involved: Acquiring working environment acoustic data and historical communication acoustic data of the dual-mode walkie-talkie in the target operating area, and determining the communication quality distribution of the target operating area based on the working environment acoustic data and the historical communication acoustic data; Acquire communication behavior data of the dual-mode walkie-talkie within a preset time period, perform clustering operations on the communication behavior data based on a clustering algorithm, and construct a voice communication behavior density map; Determine and construct a regionalized speech enhancement model for the target operation area based on the communication quality distribution and the speech communication behavior density map; Acquire incremental data on voice communication behavior at different locations in the target operation area in real time, and determine the emergency status of different locations based on the incremental data and a regionalized voice enhancement model; A speech processing strategy for the target operation area is constructed according to the emergency state.

2. The method for voice enhancement processing of a dual-mode intercom based on AI recognition according to claim 1 is characterized in that: The step of obtaining working environment acoustic data and historical communication acoustic data of the dual-mode intercom in the target working area, and determining the communication quality distribution of the target working area based on the working environment acoustic data and the historical communication acoustic data, is specifically as follows: Obtain historical intercom record data of the dual-mode intercom in the target operation area, obtain working environment acoustic data and historical communication acoustic data of the dual-mode intercom in the target operation area based on the historical intercom record data, and obtain location information of the dual-mode intercom during communication; Converting the work environment acoustic data into a frequency domain signal based on a fast Fourier transform, extracting the Mel-frequency cepstral coefficients, spectral centroid, spectral entropy reduction, zero-crossing rate, and energy envelope of the work environment acoustic data according to the frequency domain signal, and constructing a feature vector of the work environment acoustic data; Building a work environment noise recognition model based on a convolutional neural network, importing the work environment acoustic data feature vector into the work environment noise recognition model to identify the work environment noise of the dual-mode intercom, and obtaining a work environment noise type recognition result; Determine the noise intensity of the working environment noise type according to the energy envelope, and determine the noise position according to the position information during the communication process of the dual-mode walkie-talkie to obtain noise data; Perform audio framing and windowing on historical communication acoustic data. Use a blind source separation algorithm to demix and extract the communication voice signal from the background noise. Calculate the energy values ​​of each frame of voice signal and the corresponding background noise, and generate a signal-to-noise ratio curve based on the ratio of voice energy to noise energy. determining speech clarity of historical communication acoustic data at different locations in the target operation area based on the signal-to-noise ratio curve and the location information; The noise data and speech clarity at different locations are spatially aligned. The communication quality at different locations is evaluated based on the aligned noise data and speech clarity. A communication quality distribution map is constructed to obtain the communication quality distribution of the target operation area.

3. The method for voice enhancement processing of a dual-mode intercom based on AI recognition according to claim 1 is characterized in that: The communication behavior data of the dual-mode intercom within a preset time period is obtained, and the communication behavior data is clustered based on a clustering algorithm to construct a voice communication behavior density map, specifically: Acquire the communication behavior data of the dual-mode walkie-talkie within a preset time period, wherein the communication behavior data includes a communication timestamp, communication location information, and call frequency; Constructing a spatiotemporal coordinate sequence based on the communication timestamp and communication location information, generating a multidimensional feature vector set in combination with the call frequency, initializing density clustering algorithm parameters, setting the neighborhood radius eps and the minimum number of neighborhood points min_samples; Based on the density clustering algorithm, the Euclidean distance between all data points in the multidimensional feature vector set is calculated to determine the spatial topological relationship. Each data point in the multidimensional feature vector set is traversed. If the number of data points in the eps neighborhood of the current data point is not less than min_samples, it is marked as a core point, and each core point is marked as unvisited. Create a new cluster for each unvisited core point, obtain all data points in the neighborhood of each core point eps and store them in a queue, recursively visit each data point in the queue, and when the data point that is visited is not marked by any cluster, it is added to the current cluster. If the data point also meets the core point conditions, its neighboring data points are expanded and added to the queue until all data points are visited. The unclustered data points are marked as noise points, and the cluster assignment result is generated; Based on the cluster allocation results, the coverage area of ​​each cluster in the target operation area is calculated. The spatial density value is obtained according to the ratio of the number of data points in the cluster to the area of ​​the target operation area. The spatial density value is mapped to the corresponding two-dimensional plane area of ​​the target operation area, and a voice communication behavior density map of different locations in the target operation area within a preset time period is constructed.

4. The method for voice enhancement processing of a dual-mode intercom based on AI recognition according to claim 1, characterized in that: The determining and constructing of a regionalized speech enhancement model for a target operation area according to the communication quality distribution and the voice communication behavior density map is specifically as follows: Mapping the speech clarity data and noise data at different locations in the communication quality distribution into a speech communication behavior density map, and calculating the abnormal communication probability at each location in the target operation area based on the speech clarity data, noise data and communication behavior density at each location; If the probability of abnormal communication at a certain location is greater than a preset value, the area where the location is located is marked as a voice enhancement demand area, and noise signal data of each voice enhancement demand area is extracted based on the working environment acoustic data and historical communication acoustic data; Calculating the energy distribution of the noise signal data in a preset frequency band, identifying a noise-dominant frequency band whose energy exceeds a preset energy threshold, and calculating a noise suppression frequency band parameter based on the center frequency and bandwidth of the noise-dominant frequency band; Calculating statistical characteristics of the energy envelope of the noise signal data, including peak value, mean value, and variance; when the peak value of the energy envelope exceeds a preset peak value threshold, calculating an instantaneous gain suppression amplitude parameter based on the peak value; when the mean value of the energy envelope exceeds a preset mean value threshold, calculating a steady-state gain suppression amplitude parameter based on the mean value; and fusing the instantaneous gain suppression amplitude parameter and the steady-state gain suppression amplitude parameter to generate a noise suppression gain parameter; Determining a speech enhancement response sensitivity parameter according to a speech communication behavior density value of each speech enhancement demand area; Noise data of each area requiring speech enhancement is obtained, and a noise feature matrix is ​​constructed by combining the noise data with noise suppression frequency band parameters and noise suppression gain parameters. The noise feature matrix is ​​input into a BP neural network for parameter learning to construct a regionalized speech enhancement model. The response sensitivity of the speech enhancement processing parameters output by the regionalized speech enhancement model in different areas is determined according to the speech enhancement response sensitivity parameters.

5. The method for voice enhancement processing of a dual-mode intercom based on AI recognition according to claim 1, characterized in that: The method of acquiring incremental data of voice communication behaviors at different locations in the target operation area in real time and determining the emergency status of different locations based on the incremental data and the regionalized voice enhancement model is as follows: Acquire voice communication behavior increment data at different locations in the target operation area in real time, mark areas where voice communication behavior increments are greater than a preset increment value as emergency risk areas, and acquire real-time voice communication data for the emergency risk areas; extracting real-time noise data of the real-time voice communication data, inputting the real-time noise data into a regionalized voice enhancement model for matching, and determining voice enhancement processing parameters of the real-time voice communication data under the current real-time noise data; Performing voice enhancement processing on the real-time voice communication data according to the voice enhancement processing parameters, performing semantic analysis on the real-time call voice data after the voice enhancement processing, and converting it into voice text data; The emergency status of the emergency risk area is determined based on the voice and text data.

6. The method for voice enhancement processing of a dual-mode intercom based on AI recognition according to claim 1, characterized in that: The speech processing strategy for the target operation area constructed according to the emergency state is specifically: Obtaining an emergency state calibration result of the emergency risk area within a preset time period according to the emergency state, calibrating the emergency risk area where the emergency state exists as an emergency area according to the emergency state calibration result, and determining the emergency state level of the emergency area; Generate estimated concurrent communication instruction quantity data based on the emergency state level and incremental voice communication behavior data of the emergency area, compare the estimated concurrent communication instruction quantity with a historical concurrency threshold, and when the estimated concurrent communication instruction quantity exceeds the historical concurrency threshold, determine that there is a command output conflict risk in the current emergency area, and extract real-time voice communication data of the emergency area under the command output conflict risk state; When the estimated number of concurrent communication instructions does not exceed the historical concurrency threshold, the cumulative probability of potential conflict is calculated based on the emergency state level and the incremental data of voice communication behavior. If the cumulative probability of potential conflict exceeds the preset conflict probability threshold, the area is determined to be a high-incidence area of ​​potential conflict, and the incremental data of voice communication behavior in the subsequent time window of the area is obtained in real time; Constructing a conflict risk assessment matrix based on the real-time voice communication data in the command output conflict risk state or the incremental voice communication behavior data in the potential conflict high-incidence area, and determining the command output conflict risk value of the emergency area based on the conflict risk assessment matrix; The command output priority areas within the target operation area are divided according to the command output conflict risk value and the emergency state level, and the resource allocation ratio of the regionalized speech enhancement model to the emergency area is adjusted to obtain the speech processing strategy of the target operation area.

7. A dual-mode intercom voice enhancement processing system based on AI recognition, characterized in that: The dual-mode intercom voice enhancement processing system based on AI recognition includes a memory and a processor. The memory includes a dual-mode intercom voice enhancement processing method program based on AI recognition. When the dual-mode intercom voice enhancement processing method program based on AI recognition is executed by the processor, the following steps are implemented: Acquiring working environment acoustic data and historical communication acoustic data of the dual-mode walkie-talkie in the target operating area, and determining the communication quality distribution of the target operating area based on the working environment acoustic data and the historical communication acoustic data; Acquire communication behavior data of the dual-mode walkie-talkie within a preset time period, perform clustering operations on the communication behavior data based on a clustering algorithm, and construct a voice communication behavior density map; Determine and construct a regionalized speech enhancement model for the target operation area based on the communication quality distribution and the speech communication behavior density map; Acquire incremental data on voice communication behavior at different locations in the target operation area in real time, and determine the emergency status of different locations based on the incremental data and a regionalized voice enhancement model; A speech processing strategy for the target operation area is constructed according to the emergency state.

Images