A recording badge data shunting processing method based on an interaction area determination, a recording badge, a medium and a product

Abstract

The application provides a recording badge data shunting processing method based on interactive area judgment, a recording badge, a medium and a product, and relates to the technical field of communication and electronic data. The method comprises the following steps: before the recording badge is enabled, extracting a voiceprint feature from a voice input by a wearer and saving the voiceprint feature in a local device as a judgment basis. During work, multiple microphones on the badge simultaneously collect sound, and the sound source position is judged by analyzing the time difference and intensity difference of the sound reaching different microphones. A cone-shaped area in front of the badge is established as an effective interaction range. The collected audio is divided into short frames, and whether the sound source of each frame is in the effective area is judged, and whether the wearer speaks is confirmed by comparing the voiceprint. According to the two judgment results, a corresponding classification label is assigned to each audio frame to form shunting data. The technical problem that the recording badge cannot selectively process audio from different sources at the data collection source in the related art is solved.

Description

Technical Field

[0001] This application relates to the fields of communication and electronic data technology, and in particular to a method for data diversion processing of recording badges based on interaction area determination, a recording badge, a medium, and a product. Background Technology

[0002] Recording badges, as wearable audio capture devices, are widely used in areas such as enterprise service quality management and customer interaction analysis. Traditional recording badges use omnidirectional microphones for continuous audio capture, uploading all audio data indiscriminately to cloud servers. However, in open office environments, service halls, and other multi-person interaction scenarios, this indiscriminate recording mode simultaneously captures the wearer's work conversations, conversations of unrelated people around them, and environmental noise. This makes it difficult for the backend analysis system to accurately identify core work conversations, and indiscriminate recording of third-party conversations also poses a serious risk of privacy leaks.

[0003] To address the issue of mixed audio data, a speaker separation solution based on a backend server has been proposed. The solution works as follows: the employee ID card still collects all audio data and uploads it completely to a cloud server. Deep learning models such as Conv-TasNet or DPTNet are deployed on the server to separate speaker separation algorithms. By analyzing the time-frequency characteristics of the audio signal and the differences in the acoustic characteristics of the speakers, the different speakers' voices in the mixed audio are separated, generating multiple individual audio streams. Then, the server uses a pre-stored employee voiceprint template library to perform voiceprint matching on each separated audio stream, identifying the audio segments belonging to the ID card wearer. Finally, only the audio identified as belonging to the wearer and their direct conversation partner is processed for subsequent speech transcription and content analysis, while the audio data of other irrelevant personnel is deleted or archived. This solution, through complex audio separation and recognition processing in the cloud, can extract core work dialogue content to a certain extent, improving the accuracy of subsequent analysis.

[0004] However, with the above technical solution, since all raw audio data must be completely uploaded to the cloud server before separation and identification processing, this means that third-party conversations unrelated to the work scenario have already left the user's control before being identified and deleted, and have left data traces in the network transmission and server storage stages. Even if this data is subsequently deleted on the server side, it is impossible to completely eliminate the risk that the data may be intercepted, cached, or maliciously used in the transmission path and temporary storage. This "upload first, process later" architecture fundamentally cannot meet the technical requirements of data minimization principle and end-to-end privacy protection, thus causing the recording badge in the relevant technology to be unable to selectively process audio from different sources at the data collection source. Summary of the Invention

[0005] This application provides a method for data diversion processing of recording badges based on interactive area determination, as well as a recording badge, medium, and product, so as to enable the recording badge to selectively process audio from different sources at the data acquisition source.

[0006] Firstly, this application provides a data diversion processing method for recording badges based on interactive region determination, applied to the aforementioned recording badges. The method includes: during the initialization phase of the recording badge, acquiring a standard voice sample of the wearer through the recording badge's audio acquisition unit, and storing the voiceprint feature vector extracted from the standard voice sample as the wearer's voiceprint reference in the recording badge's local storage unit; during the operational acquisition phase of the recording badge, acquiring audio signals in real time through a microphone array integrated into the recording badge, and determining the audio source generating the audio signal based on the arrival time difference and energy intensity difference of the audio signals received by different microphones in the microphone array. The spatial location parameters are used to construct an effective interaction area, which is a cone-shaped spatial area extending in front of the wearer with the recording badge as the vertex. The audio signal is processed by short-time frame segmentation to obtain audio frame sequence data. Each audio frame in the audio frame sequence data is processed for region attribution determination based on the spatial location parameters to obtain the region attribution determination result. In addition, each audio frame is processed for voiceprint matching determination based on the wearer's voiceprint reference to obtain the voiceprint matching determination result. Based on the region attribution determination result and the voiceprint matching determination result, a corresponding classification label is assigned to each audio frame data to obtain the target audio distribution data.

[0007] By adopting the above technical solution, the voiceprint feature vector extracted in the initialization stage and the voiceprint matching judgment in the working acquisition stage form an identity recognition link. The arrival time difference and energy intensity difference of the microphone array can improve the spatial positioning accuracy through mutual verification. The spatial position parameters and the geometric constraints of the conical spatial region work together to accurately define the effective interaction range. Short-time frame segmentation allows each audio frame to independently undergo dual verification of region attribution judgment and voiceprint matching judgment. The spatial attributes provided by region attribution judgment and the identity attributes provided by voiceprint matching judgment complement each other to form a two-dimensional classification matrix. The wearer's voiceprint benchmark is used as a constant reference for real-time comparison with dynamically acquired audio frames. The continuous update of spatial position parameters combined with the fixed boundary of the effective interaction area can achieve accurate classification of dynamic audio streams. The classification tag allocation mechanism converts the two-dimensional judgment results into structured target audio stream data, enabling intelligent differentiation and selective processing of audio from different sources at the data acquisition source. This solves the technical problem in related technologies that recording badges cannot selectively process audio from different sources at the data acquisition source, achieving the technical effect of selective processing of audio from different sources by recording badges at the data acquisition source.

[0008] Optionally, each audio frame in the audio frame sequence data is processed for region attribution determination based on spatial location parameters to obtain a region attribution determination result. Additionally, each audio frame is processed for voiceprint matching determination based on the wearer's voiceprint reference to obtain a voiceprint matching determination result. Specifically, this includes: determining the relative spatial relationship between the frame sound source position and the effective interaction area of ​​each audio frame data based on spatial location parameters; marking the region attribution determination result corresponding to the frame sound source position located within the cone-shaped spatial area as within the area based on the relative spatial relationship; and marking the region attribution determination result corresponding to the frame sound source position located outside the cone-shaped spatial area as outside the area based on the relative spatial relationship; extracting voiceprint feature vectors from each audio frame data; comparing the similarity of the voiceprint feature vectors with the wearer's voiceprint reference to obtain a similarity comparison result; marking the voiceprint matching determination result corresponding to audio frame data with a similarity value greater than a preset similarity threshold as a match based on the similarity comparison result; and marking the voiceprint matching determination result corresponding to audio frame data with a similarity value less than or equal to a preset similarity threshold as a mismatch based on the similarity comparison result.

[0009] By adopting the above technical solution, the spatial coordinates of the frame sound source location are compared with the geometric boundaries of the effective interaction area, which can convert the continuous spatial relationship into a discrete region attribution label; the similarity value formed by the extraction of voiceprint feature vector and comparison with the wearer's voiceprint benchmark can be converted into a binary matching result through the judgment of a preset similarity threshold; the spatial judgment inside and outside the region and the identity judgment of matching or not matching are executed in parallel. The judgment results of the two dimensions are independent of each other but work together, which can avoid the one-sidedness of single-dimensional judgment; the determination of relative spatial position relationship provides the basis for region attribution, and the quantification of similarity comparison results provides the judgment benchmark for voiceprint matching. The combination of the two judgment mechanisms can form a four-quadrant classification system; the cross-combination of the spatial dimension binary classification and the identity dimension binary classification can cover all possible audio source scenarios and realize comprehensive classification management of audio data.

[0010] Optionally, the relative spatial relationship between the frame source position and the effective interaction area of ​​each audio frame data is determined based on the spatial location parameters. The region attribution result corresponding to the frame source position located within the cone-shaped spatial region is marked as within the region, and the region attribution result corresponding to the frame source position located outside the cone-shaped spatial region is marked as outside the region. Specifically, this includes extracting the azimuth angle, pitch angle, and spatial distance parameters of the frame source position from the spatial location parameters. The azimuth angle parameter represents the deviation angle of the frame source position on the horizontal plane relative to the forward central axis of the recording sign; the pitch angle parameter represents the deviation angle of the frame source position on the vertical plane relative to the forward central axis of the recording sign; and the spatial distance parameter represents the spatial distance between the frame source position and the recording sign. Distance; determine whether the azimuth parameter is within the preset horizontal angle range, whether the pitch parameter is within the preset vertical angle range, and whether the spatial distance parameter is less than the preset extension distance of the conical spatial region; when the azimuth parameter is within the horizontal angle range, the pitch parameter is within the vertical angle range, and the spatial distance parameter is less than the preset extension distance, determine that the relative spatial position is within the region, and mark the region attribution determination result of the audio frame data corresponding to the frame sound source position as within the region; when the azimuth parameter is not within the horizontal angle range, or the pitch parameter is not within the vertical angle range, or the spatial distance parameter is greater than or equal to the preset extension distance, determine that the relative spatial position is outside the region, and mark the region attribution determination result of the audio frame data corresponding to the frame sound source position as outside the region.

[0011] By adopting the above technical solution, the spatial descriptions of azimuth, elevation, and spatial distance parameters work together to accurately quantify the location of sound sources in three-dimensional space. The constraints of the horizontal angular range on the azimuth parameter, the vertical angular range on the elevation parameter, and the preset extension distance on the spatial distance parameter, combined with these three constraints, accurately define the three-dimensional boundary of the conical spatial region. The joint judgment of the three parameters forms a strict admission mechanism through logical AND operation; if any parameter exceeds its corresponding range, an out-of-area judgment is triggered. This multi-verification mechanism improves the reliability of the judgment. The combination of azimuth and elevation angle judgments with spatial distance judgments ensures directional accuracy through angle constraints and guarantees the effectiveness of the effective range through distance constraints. The multi-parameter collaborative judgment reduces misjudgments in areas with ambiguous boundaries. The complete judgment chain formed by the independent extraction and joint judgment of the three spatial parameters can transform complex spatial relationships into clear regional attribution results.

[0012] Optionally, each audio frame is assigned a corresponding classification label based on the region attribution determination result and the voiceprint matching determination result to obtain target audio distribution data. Specifically, this includes: assigning a wearer's own audio label to audio frame data where the region attribution determination result is within the region and the voiceprint matching determination result is a match; assigning an interactive object audio label to audio frame data where the region attribution determination result is within the region and the voiceprint matching determination result is a mismatch; assigning a wearer's non-interactive audio label to audio frame data where the region attribution determination result is outside the region and the voiceprint matching determination result is a match; and assigning an environmental background audio label to audio frame data where the region attribution determination result is outside the region and the voiceprint matching determination result is a mismatch. Each audio frame is then differentiated based on the wearer's own audio label, the interactive object audio label, the wearer's non-interactive audio label, and the environmental background audio label to obtain target audio distribution data.

[0013] By adopting the above technical solution, the binary results of region attribution determination and voiceprint matching determination are combined by Cartesian product to generate four mutually exclusive and complete classification categories. The allocation condition of the wearer's own audio tag needs to satisfy both regional and voiceprint matching, while the audio tag of the interactive object needs to be within the region but the voiceprint does not match. This combination logic can accurately identify different interaction scenarios. In the case of the wearer's non-interactive audio tag corresponding to the region but the voiceprint matches, and the case of the environmental background audio tag corresponding to the region but the voiceprint does not match, the allocation rules of the four tags are mutually exclusive to ensure the uniqueness of the classification. The mapping relationship between the logical combination of the two-dimensional determination results and the differentiated processing strategy enables each type of audio to obtain a processing method that matches its characteristics. The four classification tags serve as the trigger conditions for subsequent differentiated processing. Different tags activate different processing flows, which can realize the refined and personalized management of audio data.

[0014] Optionally, each audio frame data is differentiated based on the wearer's own audio marker, the interactive object's audio marker, the wearer's non-interactive audio marker, and the ambient background audio marker to obtain target audio splitting data. Specifically, this includes: configuring a first priority parameter for audio frame data with the wearer's own audio marker to obtain first-class processed audio frame data, where the first priority parameter indicates that the first-class processed audio frame data has the highest processing priority; configuring a second priority parameter for audio frame data with the interactive object's audio marker to obtain second-class processed audio frame data, where the second priority parameter indicates that the second-class processed audio frame data has the second-highest processing priority; and performing volume attenuation processing on audio frame data with the wearer's non-interactive audio marker, and then processing the attenuated audio frame data accordingly. The audio frame data is configured with a third priority parameter to obtain the third type of processed audio frame data. The third priority parameter is used to indicate that the third type of processed audio frame data has the lowest processing priority. The audio frame data with environmental background audio tags is subjected to privacy desensitization processing to obtain the fourth type of processed audio frame data. The first type of processed audio frame data is encapsulated with the corresponding classification tag and corresponding audio data acquisition timestamp information, the second type of processed audio frame data is encapsulated with the corresponding classification tag and corresponding audio data acquisition timestamp information, the third type of processed audio frame data is encapsulated with the corresponding classification tag and corresponding audio data acquisition timestamp information, and the fourth type of processed audio frame data is encapsulated with the corresponding classification tag and corresponding audio data acquisition timestamp information to generate the target audio stream data.

[0015] By adopting the above technical solutions, the gradient setting of priority parameters corresponds to the decreasing law of audio importance. The decreasing configuration from the first priority parameter to the third priority parameter can guide the rational allocation of system resources. The combination of volume attenuation processing and the third priority parameter can maintain the processing order of low priority while reducing the amount of data, which can balance storage efficiency and information preservation. Privacy desensitization processing matches the special characteristics of the fourth type of audio, retaining necessary environmental information while protecting privacy, which can meet compliance requirements. Four different preprocessing methods form a mapping relationship with the corresponding classification tags. The differentiated processing of priority configuration, volume attenuation, and privacy desensitization can adapt to the characteristic needs of different types of audio. The joint encapsulation of classification tags, timestamp information, and processed audio data, and the structured combination of multi-dimensional information can ensure the integrity and traceability of data. The implementation of differentiated processing strategies and the generation of target audio diversion data form a complete processing link, which can realize the efficient conversion from raw audio to structured data.

[0016] Optionally, privacy-desensitizing processing is performed on the audio frame data with environmental background audio markers to obtain the fourth type of processed audio frame data. Specifically, this includes: performing time-frequency transformation processing on the audio frame data with environmental background audio markers to obtain time-frequency domain representation data; determining the energy density value of each time-frequency point based on the energy distribution in the time-frequency domain representation data, and determining the region of continuous time-frequency points with energy density values ​​greater than a preset density threshold as the region to be processed; performing phase randomization processing on the time-frequency domain representation data within the region to be processed to obtain phase-perturbed time-frequency data, which is time-frequency data with the same amplitude spectrum structure but a different phase spectrum structure as the time-frequency domain representation data; performing downsampling processing on the time-frequency domain representation data outside the region to be processed to obtain compressed time-frequency data; merging the phase-perturbed time-frequency data and the compressed time-frequency data to obtain desensitized time-frequency data; and performing inverse time-frequency transformation processing on the desensitized time-frequency data to obtain the fourth type of processed audio frame data.

[0017] By employing the above technical solutions, time-frequency transformation converts time-domain audio to the time-frequency domain, making energy distribution visible. The combination of energy density value calculation and threshold comparison can accurately locate high-energy regions. The identification of the region to be processed and the targeted implementation of phase randomization processing work together. Phase perturbation of only high-energy regions can preserve energy characteristics while compromising intelligibility. Downsampling processing outside the region to be processed and phase randomization processing within the region form a differentiated strategy. The selective application of the two processing methods can balance privacy protection and data compression. The preservation of the same amplitude spectrum structure and the generation of different phase spectrum structures are combined. The preservation of amplitude information can maintain statistical characteristics, while the randomization of phase information can destroy semantic content. The merging operation of phase-perturbed time-frequency data and compressed time-frequency data integrates the two processing results into unified desensitized time-frequency data. The application of inverse time-frequency transformation can restore the processing results to a usable audio format. The serial execution of the multi-step processing flow and the generation of the final fourth type of processed audio frame data can achieve an optimized balance between privacy protection and data usability.

[0018] Optionally, after encapsulating the first type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information, the second type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information, the third type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information, and the fourth type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information to generate target audio offloading data, the method further includes: maintaining the original audio data format of the first type of processed audio frame data, offloading the first type of processed audio frame data to the core working audio channel, adding an upload timestamp tag to the first type of processed audio frame data, and uploading it to the target audio stream through the core working audio channel. The system is as follows: First, it routes the second type of processed audio frame data to the interactive audio channel. Then, it adds an interactive object identifier and an interaction type marker to the original audio data format of the second type of processed audio frame data before uploading it to the server via the interactive audio channel. Second, it routes the third type of processed audio frame data to the low-priority cache channel. When the local cache space reaches a preset capacity threshold, it performs batch compression on the third type of processed audio frame data before uploading it to the target server via the low-priority cache channel. Third, it routes the fourth type of processed audio frame data to the local backup channel. The fourth type of processed audio frame data is stored in segments according to a preset time window. When the storage time of the fourth type of processed audio frame data exceeds the preset retention period, the fourth type of processed audio frame data is deleted.

[0019] By adopting the above technical solutions, the parallel setup of four dedicated channels matches the differentiated transmission requirements of four types of processed audio. Channel isolation can avoid mutual interference between data of different priorities. The combination of preserving the original format and adding upload timestamps is applied to the first type of data, ensuring the integrity and timing accuracy of the core audio. The addition of interactive object identifiers and interactive type tags is coordinated with the upload of the second type of data, and the addition of identifier information can enhance the retrieval of interactive content. The combined action of batch compression processing and capacity threshold triggering mechanism is applied to the third type of data, and the combination of compression and caching can optimize network bandwidth utilization. The cyclical execution of segmented storage and automatic deletion mechanism is applied to the fourth type of data, and the combination of time window management and retention period control can balance storage costs and review requirements. The differentiated processing strategies adopted by different channels are adapted to the characteristics of various types of audio, and the diversified transmission and storage solutions can meet different business needs, achieving efficient and stable operation of recording badges.

[0020] Secondly, embodiments of this application provide a recording badge, which includes: one or more processors and a memory; the memory is coupled to one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the recording badge to perform the method described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a recording badge, cause the recording badge to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a recording badge, cause the recording badge to perform the method described in the first aspect and any possible implementation thereof.

[0023] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0024] 1. By deploying a microphone array at the recording work badge for spatial positioning and voiceprint recognition, audio from different sources can be accurately classified at the first moment of audio acquisition. This avoids the problem in traditional solutions where all audio data must be uploaded to the cloud before it can be separated and processed. It fundamentally eliminates the risk of privacy leakage of irrelevant audio during network transmission and server storage, and achieves true end-to-end privacy protection.

[0025] 2. The cone-shaped spatial region-based judgment mechanism can identify audio within the effective interaction range in real time. Combined with the differentiated processing strategies for four types of audio and dedicated channel transmission, core working audio can be processed and transmitted first, interactive audio is appropriately protected, non-interactive audio is compressed and cached, and environmental audio is locally desensitized. This hierarchical processing mechanism can reduce the amount of invalid data transmitted, reduce network bandwidth usage and server processing load.

[0026] 3. Environmental background audio is stored locally after being anonymized through phase randomization in the time and frequency domain. Unrelated third-party conversations will not leave the user's control, thus meeting the principle of data minimization. At the same time, the clear labeling and structured encapsulation of the four audio types ensure the traceability and auditability of the data, thus protecting the privacy of irrelevant personnel while meeting the service quality management requirements. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating a method for processing recording ID card data diversion based on interaction area determination in an embodiment of this application.

[0028] Figure 2 This is a schematic diagram of the physical device structure of a recording badge in the embodiments of this application. Detailed Implementation

[0029] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to any or all possible combinations including one or more of the listed items.

[0030] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0031] This application provides a method for splitting and processing recording ID card data based on interaction area determination, see reference. Figure 1 , Figure 1 This is a flowchart illustrating a method for data diversion of recording work badges based on interaction area determination in this application, comprising the following steps:

[0032] Step S101: During the initialization phase of the recording badge, the standard voice sample of the wearer is obtained through the audio acquisition unit of the recording badge, and the voiceprint feature vector extracted from the standard voice sample is used as the voiceprint reference of the wearer and stored in the local storage unit of the recording badge.

[0033] Step S102: During the recording work card acquisition phase, audio signals are acquired in real time through the microphone array integrated into the recording work card, and the spatial location parameters of the audio source that generates the audio signal are determined based on the arrival time difference and energy intensity difference of the audio signals received by different microphones in the microphone array.

[0034] Step S103: Construct an effective interaction area based on spatial location parameters. The effective interaction area is a cone-shaped spatial area extending in front of the wearer with the recording badge as the vertex.

[0035] Step S104: Perform short-time frame segmentation processing on the audio signal to obtain audio frame sequence data, and perform region attribution determination processing on each audio frame data in the audio frame sequence data according to the spatial location parameters to obtain the region attribution determination result. Also, perform voiceprint matching determination processing on each audio frame according to the wearer's voiceprint reference to obtain the voiceprint matching determination result.

[0036] Step S105: Assign a corresponding classification label to each audio frame data according to the region attribution determination result and the voiceprint matching determination result to obtain the target audio split data.

[0037] In the above embodiment, taking a bank branch lobby as an example, bank tellers wear recording badges for daily customer service. The lobby has an open layout with multiple service windows adjacent to each other. The environment contains various audio sources, including conversations between other tellers and customers, conversations of customers waiting in line, and lobby announcements. The recording badges need to be initialized before the teller wears them for the first time. In a relatively quiet environment, the teller reads a pre-set text, approximately 30 seconds long, containing sentences with different phoneme combinations, through the audio acquisition unit of the recording badge. The collected standard speech samples are preprocessed, including pre-emphasis, framing, and windowing. Each frame is set to 25 milliseconds in length and has a frame shift of 10 milliseconds. Mel-frequency cepstral coefficients are extracted from each frame of audio, and combined with the fundamental frequency and formant parameters to construct multi-dimensional acoustic features. These acoustic features are modeled using a Gaussian mixture model to generate a voiceprint feature vector containing the teller's voice characteristics, including personalized features such as timbre, pitch, and rhythm. The extracted voiceprint feature vector is used as the voiceprint reference for the teller and stored in the local storage unit of the recording badge. The storage format adopts binary encoding to reduce storage space usage.

[0038] In the above embodiment, during the daily work data collection phase, the recording badge continuously collects ambient audio through an integrated quad microphone array. The four microphones are arranged in a cross shape on the badge surface, with a 3-centimeter spacing between adjacent microphones. This layout effectively captures the spatial information of the sound source. When a customer speaks with a teller, the customer's voice propagates as sound waves. Due to the varying distances between each microphone and the sound source, the arrival times of the sound waves at each microphone differ slightly. The time delay between each microphone pair is calculated using a generalized cross-correlation algorithm; for example, the time difference between the left and right microphones receiving the same sound is 0.08 milliseconds, and the time difference between the top and bottom microphones is 0.05 milliseconds. Simultaneously, due to the energy attenuation characteristics of sound wave propagation, the microphone closer to the sound source receives a stronger signal. The signal power spectral density of each microphone is measured, and the energy intensity difference is calculated. Combining the arrival time difference and the energy intensity difference, a triangulation algorithm is used to calculate the azimuth, pitch, and distance of the audio source relative to the recording badge. These parameters collectively constitute the spatial location parameters of the audio source.

[0039] In the above embodiment, an effective interaction area is constructed based on the obtained spatial location parameters. This area is defined as a cone-shaped space extending forward from the recording badge as its apex. The central axis of the cone-shaped area is aligned with the direction directly in front of the wearer, with a horizontal angle of 120 degrees, covering the area directly in front of the teller and 60 degrees to the left and right; a vertical angle of 90 degrees, covering 45 degrees above and below the horizontal plane; and an extension distance of 2 meters, which covers the normal communication range inside and outside the counter. This cone-shaped space effectively covers the face-to-face interaction area between the teller and the customer, while eliminating interference from other windows to the side and rear. The continuously acquired audio signals are processed by short-time frame segmentation, dividing the continuous audio stream into discrete audio frame sequences. Each audio frame has a duration of 20 milliseconds, with a 10-millisecond overlap between adjacent frames. This overlap ensures the continuity of audio information. For each audio frame, two parallel decision processing flows are executed. In the area attribution decision flow, the spatial location parameters of the sound source at the corresponding time of the audio frame are extracted to determine whether the sound source is located within the preset cone-shaped space area. For example, the sound source location parameters of one audio frame show an azimuth angle of 30 degrees, an elevation angle of 10 degrees, and a distance of 1.5 meters. These parameters are all within the cone-shaped area, so the area classification result for this frame is marked as within the area. However, the sound source location of another audio frame shows an azimuth angle of 150 degrees, which clearly exceeds the 120-degree horizontal angle range, indicating that the sound source is coming from the side and rear of the teller. This frame is marked as outside the area.

[0040] In the above embodiments, during the voiceprint matching determination process, real-time voiceprint feature vectors are extracted from audio frames using the same feature extraction method as in the initialization phase. The extracted real-time voiceprint feature vectors are then compared with the stored wearer voiceprint baseline using cosine similarity calculation. The similarity value ranges from 0 to 1, and a similarity threshold of 0.75 is set. When the voiceprint similarity calculation result for an audio frame is 0.82, it exceeds the threshold and is determined to be a voiceprint match; when the similarity is 0.45, it is below the threshold and is determined to be a voiceprint mismatch. After completing the two-dimensional determination, a classification label is assigned to each audio frame based on the combination of the two determination results. When a teller introduces services to a customer, the audio frame generated by their voice is within the designated area and matches the voiceprint, and is labeled as the wearer's own audio. The audio frame when a customer asks a question is within the designated area but the voiceprint does not match, and is labeled as the interaction audio. The audio frame when a teller turns to briefly communicate with a colleague, although the voiceprint matches, is outside the designated area and is labeled as the wearer's non-interaction audio. The audio frames of conversations between other tellers and customers at adjacent windows are both outside the designated area and do not match the voiceprint, and are labeled as ambient background audio. After all audio frames are classified and labeled, they form structured target audio distribution data. Each data unit contains the original data of the audio frame, classification label, timestamp information, and spatial location parameters. This structured data allows the recording badge to perform differentiated processing on different types of audio locally. The wearer's own audio and the interaction audio can be prioritized for transmission for service quality analysis, while ambient background audio can be stored or discarded locally after privacy protection processing, preventing the leakage of irrelevant personnel's privacy information at the source, while ensuring the complete recording of core business audio.

[0041] Through the above steps, the voiceprint feature vector extracted in the initialization phase and the voiceprint matching judgment in the data acquisition phase form an identity recognition link. The arrival time difference and energy intensity difference of the microphone array are mutually verified to improve spatial positioning accuracy. The spatial position parameters and the geometric constraints of the conical spatial region work together to accurately define the effective interaction range. Short-time frame segmentation allows each audio frame to independently undergo dual verification of region attribution judgment and voiceprint matching judgment. The spatial attributes provided by region attribution judgment and the identity attributes provided by voiceprint matching judgment complement each other to form a two-dimensional classification matrix. The wearer's voiceprint benchmark is used as a constant reference for real-time comparison with dynamically acquired audio frames. The continuous update of spatial position parameters combined with the fixed boundary of the effective interaction area can achieve accurate classification of dynamic audio streams. The classification tag allocation mechanism converts the two-dimensional judgment results into structured target audio stream data, enabling intelligent differentiation and selective processing of audio from different sources at the data acquisition source. This solves the technical problem in related technologies where recording badges cannot selectively process audio from different sources at the data acquisition source, achieving the technical effect of selective processing of audio from different sources by recording badges at the data acquisition source.

[0042] The entity performing the above steps may be a system with data diversion and processing capabilities, or a device with data diversion and processing capabilities, or a controller or processor in the device or system, or a standalone controller or processor, or other processing devices or processing units with similar processing functions, but is not limited to these.

[0043] In an optional embodiment, each audio frame in the audio frame sequence data is processed for region attribution determination based on spatial location parameters to obtain a region attribution determination result. Furthermore, each audio frame is processed for voiceprint matching determination based on the wearer's voiceprint reference to obtain a voiceprint matching determination result. Specifically, this includes: determining the relative spatial relationship between the frame sound source position and the effective interaction area of ​​each audio frame data based on spatial location parameters; marking the region attribution determination result corresponding to the frame sound source position located within the cone-shaped spatial area as within the area based on the relative spatial relationship; and marking the region attribution determination result corresponding to the frame sound source position located outside the cone-shaped spatial area as outside the area based on the relative spatial relationship; extracting voiceprint feature vectors from each audio frame data; comparing the similarity of the voiceprint feature vectors with the wearer's voiceprint reference to obtain a similarity comparison result; marking the voiceprint matching determination result corresponding to audio frame data with a similarity value greater than a preset similarity threshold as a match based on the similarity comparison result; and marking the voiceprint matching determination result corresponding to audio frame data with a similarity value less than or equal to the preset similarity threshold as a mismatch based on the similarity comparison result.

[0044] In the above embodiment, it is assumed that in the claims service center of a large insurance company, claims specialists wearing recording badges handle customer claims inquiries. The service center adopts an open workstation layout, with adjacent workstations spaced approximately 1.5 meters apart. Multiple claims specialists may simultaneously communicate with different customers by phone or in person. When a claims specialist is processing a visiting customer's claim, the microphone array of the recording badge continuously collects audio signals from the environment. The collected audio stream has been segmented into a 20-millisecond audio frame sequence. Each audio frame carries spatial location parameters calculated using the microphone array, including the azimuth, pitch, and distance information of the sound source relative to the recording badge. For a particular audio frame being processed, a region attribution determination is first performed. The spatial location parameters of this audio frame show that the sound source's azimuth is 15 degrees directly in front, the pitch is 5 degrees upward, and the distance is 1.2 meters. These parameters are compared with a predefined effective interaction area, which is a cone-shaped space with the recording badge as its vertex, a horizontal angle range of 60 degrees to the left and right, a vertical angle range of 45 degrees to the right and up, and an extension distance of 2 meters. Geometric calculations determine that the sound source location of this frame is within the boundary of the cone-shaped space, thus establishing an internal containment relationship in its relative spatial position. The region attribution result of this audio frame is then marked as within the region.

[0045] In the above embodiment, at the same time, another claims specialist at an adjacent workstation is answering a phone call, and their voice is also captured by the recording badge. The spatial location parameters of the corresponding audio frame show that the azimuth angle of the sound source is 95 degrees to the left, which exceeds the 60-degree left boundary of the cone-shaped region. It is determined that the relative spatial position of the sound source and the cone-shaped spatial region is an external separation relationship, so the region attribution determination result of this audio frame is marked as outside the region. While performing the region attribution determination, the voiceprint matching determination process is executed in parallel. For each audio frame, preprocessing is first performed, including removing the DC component and pre-emphasis filtering to enhance high-frequency components. Then, the audio frame is windowed using a Hamming window function to reduce spectral leakage. Next, a fast Fourier transform is performed to convert the time-domain signal to the frequency domain and extract the power spectrum. Based on the power spectrum, the Mel frequency cepstral coefficients are calculated, and the spectral envelope features reflecting the characteristics of human hearing are extracted using a Mel filter bank. At the same time, the fundamental period is extracted using autocorrelation, and the fundamental frequency features are calculated. The formant frequency and bandwidth parameters are extracted through linear predictive analysis. These multi-dimensional acoustic features combine to form the voiceprint feature vector of the audio frame. After extraction, the voiceprint feature vector is compared with the locally stored wearer voiceprint baseline. Cosine similarity is used as the metric to calculate the cosine of the angle between the two vectors in the multi-dimensional feature space. For the audio frame generated by the claims specialist speaking, its voiceprint feature vector is highly similar to the wearer's voiceprint baseline, and the calculated similarity value is 0.86. The preset similarity threshold is 0.75. This preset similarity threshold is an empirical value derived from statistical analysis of a large amount of experimental data, which can reduce the false positive rate while ensuring recognition accuracy. Since 0.86 is greater than the threshold of 0.75, the voiceprint matching result of this audio frame is marked as a match.

[0046] In the above embodiments, for audio frames generated by a customer's speech, due to significant differences in the speaker's vocal tract structure, pronunciation habits, and other physiological and behavioral characteristics compared to the claims specialist, the similarity calculation result between the extracted voiceprint feature vector and the wearer's voiceprint baseline is 0.42, significantly lower than the threshold of 0.75, thus the voiceprint matching result is marked as a mismatch. In actual processing, some special cases may arise. For example, when a claims specialist turns to a colleague behind them to inquire about a certain clause, although it is the claims specialist speaking, the similarity value between the voiceprint feature vector and the wearer's voiceprint baseline is 0.83, higher than the threshold, and the voiceprint matching result is a match. However, because the sound source azimuth angle reaches 165 degrees, far exceeding the boundary of the cone-shaped region, the region attribution result is outside the region. Similarly, when a customer's family member provides supplementary explanations, the sound source location is within the cone-shaped region, and the region attribution is within the region, but the voiceprint similarity is only 0.38, resulting in a mismatch. The above two-dimensional judgment process is performed on each frame in the audio frame sequence, generating two independent judgment result sequences. The region attribution determination result sequence records whether each frame is within or outside the region, while the voiceprint matching determination result sequence records whether each frame is a match or a non-match. The combination of these two binary determination results provides a clear basis for subsequent four-category labeling, enabling accurate differentiation between the wearer's work conversations, customer inquiries, the wearer's non-work communication, and irrelevant noise in the environment. This achieves precise audio distribution at the audio acquisition end, avoiding privacy risks and resource waste caused by uploading all mixed audio to the server.

[0047] In an optional embodiment, the relative spatial relationship between the frame source position and the effective interaction area of ​​each audio frame data is determined based on spatial location parameters. The region attribution determination result corresponding to the frame source position located within the cone-shaped spatial region is marked as within the region based on the relative spatial location relationship, and the region attribution determination result corresponding to the frame source position located outside the cone-shaped spatial region is marked as outside the region based on the relative spatial location relationship. Specifically, this includes extracting the azimuth angle parameter, pitch angle parameter, and spatial distance parameter of the frame source position from the spatial location parameters. The azimuth angle parameter represents the deviation angle of the frame source position on the horizontal plane relative to the forward central axis of the recording sign; the pitch angle parameter represents the deviation angle of the frame source position on the vertical plane relative to the forward central axis of the recording sign; and the spatial distance parameter represents the distance between the frame source position and the recording sign. The spatial distance is determined by: 1) determining whether the azimuth parameter is within a preset horizontal angle range, and 2) determining whether the elevation parameter is within a preset vertical angle range, and 3) determining whether the spatial distance parameter is less than a preset extension distance of the conical spatial region; 4) determining whether the relative spatial position is within the region when the azimuth parameter is within the horizontal angle range, the elevation parameter is within the vertical angle range, and the spatial distance parameter is less than the preset extension distance, and marking the region attribution result of the audio frame data corresponding to the frame sound source position as within the region; 5) determining whether the relative spatial position is outside the region when the azimuth parameter is not within the horizontal angle range, or the elevation parameter is not within the vertical angle range, or the spatial distance parameter is greater than or equal to the preset extension distance, and marking the region attribution result of the audio frame data corresponding to the frame sound source position as outside the region.

[0048] In the above embodiment, assume a service center's lobby where staff wearing recording badges process various procedures for customers. The lobby uses a semi-open window system with multiple service windows side-by-side, separated only by low partitions. The environment is filled with conversations between staff at different windows, customers chatting in the waiting area, and announcements from the queuing system. When a staff member serves a customer, the recording badge needs to accurately identify which audio segments fall within the valid interaction range. The spatial location parameters of each audio frame have already been obtained using a microphone array positioning algorithm; now, it's necessary to determine whether these positions are within a predefined valid interaction area. For the currently processed audio frame, three key values ​​are extracted from its spatial location parameters. The azimuth parameter is displayed as -12 degrees, calculated by analyzing the phase difference between the signals received by the left and right microphones in the microphone array; a negative value indicates the sound source is located 12 degrees to the left of the recording badge's forward-facing central axis. The pitch parameter is displayed as +8 degrees, obtained through signal difference analysis of the upper and lower microphones; a positive value indicates the sound source is slightly above the recording badge's horizontal plane. The spatial distance parameter is 0.9 meters, which is the straight-line distance from the sound source to the recording badge calculated based on the sound intensity attenuation law and the ratio of signal energy received by multiple microphones.

[0049] In the above embodiment, a triple condition judgment is initiated. First, it is judged whether the azimuth angle parameter is within the preset horizontal angle range. The horizontal angle range of the recording badge is set to 60 degrees to the left and right, that is, a fan-shaped range from -60 degrees to +60 degrees. This range can cover the service window area directly in front of the staff while eliminating lateral interference. The azimuth angle of the current audio frame is -12 degrees, which is clearly within this range, satisfying the first condition. Next, it is judged whether the pitch angle parameter is within the preset vertical angle range. The vertical angle range is set to 45 degrees up and down, that is, from -45 degrees to +45 degrees. This range takes into account the height difference between standing and sitting customers communicating with staff. The pitch angle of the current audio frame is +8 degrees, which is within the vertical angle range, satisfying the second condition. Then, it is judged whether the spatial distance parameter is less than the preset extension distance of the cone-shaped spatial area. The extension distance is set to 2 meters. This distance is based on the actual scenario of window service, which can cover the normal communication distance inside and outside the window without extending into the waiting area in the back. The current audio frame's spatial distance is 0.9 meters, less than the 2-meter extension distance limit, thus satisfying the third condition. Since all three conditions are met, the relative spatial relationship between the audio frame's sound source location and the effective interaction area is determined to be within the area. Therefore, the area attribution result for this audio frame data is marked as within the area. This indicates that the audio likely originated from the customer or staff member currently conducting business.

[0050] In the above embodiment, another audio frame processed subsequently presented a different situation. The azimuth parameter of this audio frame was -85 degrees, indicating that the sound source was located far to the left of the staff member. Although the pitch parameter was -5 degrees, still within the vertical range, and the spatial distance parameter of 1.3 meters was also less than the extension distance, the azimuth parameter of -85 degrees exceeded the left boundary of -60 degrees, failing to meet the condition of the horizontal angle range. It was determined that if any condition was not met, the relative spatial position was outside the area, and therefore the area classification result of this audio frame was marked as outside the area. This audio likely came from other staff members or customers in adjacent windows. Other typical situations were also encountered during continuous processing. One audio frame had an azimuth of +5 degrees and a pitch of -3 degrees, both within the corresponding ranges, but the spatial distance parameter reached 3.2 meters, exceeding the 2-meter extension distance limit. This situation usually occurs when citizens waiting in the back row loudly inquire or converse; although the direction is directly facing the window, the distance is too far, not constituting direct service interaction, and it is marked as outside the area. Another scenario is where the azimuth of the audio frame is +30 degrees and the spatial distance is 1.5 meters, both meeting the requirements, but the pitch angle parameter is -55 degrees, exceeding the lower boundary of -45 degrees. This could be the sound of an object falling on the ground or other abnormally low-lying sound sources, and it is also marked as outside the area. This comprehensive judgment mechanism based on three-dimensional spatial parameters can accurately delineate the boundaries of the effective interaction area. Only when the sound source is located within a 120-degree fan-shaped area in front of the staff in the horizontal direction, within a reasonable communication height range in the vertical direction, and the distance does not exceed the normal window service range, is it considered audio within the effective interaction area. This judgment method fully considers the spatial characteristics of the actual service scenario, effectively filtering out interfering audio from the side, rear, too far away, or abnormal locations, providing a reliable spatial dimension judgment basis for subsequent audio classification processing. Using the results of regional attribution determination as one of the important bases for audio classification, combined with the results of voiceprint matching, can accurately identify the business dialogue content between staff and citizens, while excluding irrelevant environmental noise and dialogue from other windows. This solves the privacy leakage risk and data processing burden caused by the indiscriminate collection of recording badges in traditional recording methods from the source.

[0051] In an optional embodiment, each audio frame is assigned a corresponding classification label based on the region attribution determination result and the voiceprint matching determination result to obtain target audio distribution data. Specifically, this includes: assigning a wearer's own audio label to audio frame data whose region attribution determination result is within the region and whose voiceprint matching determination result is a match; assigning an interactive object audio label to audio frame data whose region attribution determination result is within the region and whose voiceprint matching determination result is a mismatch; assigning a wearer's non-interactive audio label to audio frame data whose region attribution determination result is outside the region and whose voiceprint matching determination result is a match; assigning an environmental background audio label to audio frame data whose region attribution determination result is outside the region and whose voiceprint matching determination result is a mismatch; and performing differentiated processing on each audio frame data based on the wearer's own audio label, the interactive object audio label, the wearer's non-interactive audio label, and the environmental background audio label to obtain target audio distribution data.

[0052] In the above embodiment, assume that at the outpatient service window of a large hospital, registration and payment staff wearing recording badges process patient registrations. The outpatient hall is crowded, with a complex audio environment including multiple service windows operating simultaneously, conversations between patients' families, and communication between medical staff. The recording badges have already completed the region attribution and voiceprint matching determination of the audio frame sequence. Now, based on the results of these two dimensions, an accurate classification label needs to be assigned to each audio frame. When processing a patient's registration, the payment staff asks the patient which department they need to register for. This speech is segmented into multiple audio frames, each of which has undergone spatial localization and voiceprint analysis. For a specific audio frame generated by the payment staff's speech, its region attribution determination is within the region because the sound source is located within the cone-shaped interaction space in front of the badge; simultaneously, the voiceprint matching determination is a match because the voiceprint features are consistent with the pre-stored wearer's voiceprint reference height. Based on the combination of these two determination results, the wearer's own audio label is assigned to this audio frame. This type of audio represents the content of the payment staff's business communication with patients during their normal work, and is core data for service quality assessment.

[0053] In the above embodiment, when a patient replies that they want to register for the respiratory department, a certain audio frame of this response is determined to be within the designated area because the patient is standing in front of the window and communicating face-to-face with the cashier. However, the voiceprint matching result is a mismatch, as the patient's voiceprint characteristics differ significantly from the cashier's voiceprint baseline. Based on this, an interactive object audio tag is assigned to this audio frame. This type of audio records the service recipient's expressed needs and feedback information, which is of great value for understanding the complete service process. During the process, the cashier turns to ask a colleague behind them if there are any available appointments for the respiratory department. Although the voiceprint matching result of this audio frame is a match, confirming it is the cashier's voice, the sound source location deviates from the forward cone area because the speaker is facing backward, resulting in an outside-area classification. A wearer-non-interactive audio tag is assigned to this type of audio frame. This tag indicates that although it is the staff member's voice, it does not belong to the direct interaction with the current service recipient and may involve internal communication or other matters. Another cashier at the adjacent window is processing payments for other patients, and their conversation is also captured by the recording badge. These audio frames were determined to be outside the designated area because the sound source originated from other windows to the side; the voiceprint matching result was a mismatch, indicating it was not the voice of the wearer of this badge. Environmental background audio tags were assigned to these audio frames. This type of audio falls under the category of environmental noise and contains private conversations of other individuals, requiring special processing to protect privacy. After classification and tagging, audio frames with different tags were processed differently. For frames tagged as the wearer's own audio, the original sampling rate and bit depth were maintained to ensure clear sound quality, and a priority transmission identifier was added. These audio frames were assembled into continuous audio segments, with timestamps and employee ID information appended, forming a core data stream for service quality analysis.

[0054] In the above embodiments, for frame data marked as interactive audio, high audio quality is maintained, but mild noise reduction is performed to remove subtle background interference and improve speech clarity. These audios are paired and associated with the wearer's own audio to form a complete service dialogue record. High transmission and storage priorities are assigned to these two types of audio data. For frame data marked as non-interactive audio for the wearer, audio compression is performed, reducing the sampling rate to 8kHz to decrease data volume. This type of audio is mainly used for work status monitoring and does not require high fidelity. These audio frames are cached locally and selectively uploaded only when a complete work record is needed. For frame data marked as ambient background audio, privacy protection processing is performed. First, spectral blurring is performed, using random phase perturbations to destroy the intelligibility of the speech, making it impossible to reconstruct the specific dialogue content. Then, energy statistics are performed, retaining only environmental noise level information for work environment assessment. The processed data only contains noise intensity statistics and does not contain any identifiable speech content, fundamentally avoiding the risk of privacy leakage. After differential processing, structured target audio stream data is generated. Each data unit contains the processing result of the audio frame, classification label, processing method identifier, timestamp, and associated information. The wearer's own audio and the audio of interacting objects are packaged into high-priority data packets, which can be transmitted to the management platform in real-time or near real-time for service quality analysis. The wearer's non-interactive audio is compressed and stored in a local cache, periodically cleaned up or uploaded on demand according to the storage policy. Environmental background audio undergoes privacy processing, retaining only statistical information, while the original audio content is securely deleted. This four-category labeling mechanism based on dual-dimensional judgment, combined with differentiated processing strategies, achieves precise audio data routing. Unlike traditional solutions that upload all audio indiscriminately, audio type identification and targeted processing are completed at the recording badge end. Core business audio is prioritized, ensuring the data requirements for service quality management; simultaneously, by protecting the privacy of environmental audio, the risk of privacy leaks by unauthorized personnel is eliminated at the source, effectively reducing network transmission burden and server computing resource consumption compared to backend server processing solutions.

[0055] In an optional embodiment, each audio frame data is differentiated based on the wearer's own audio marker, the interactive object's audio marker, the wearer's non-interactive audio marker, and the ambient background audio marker to obtain target audio splitting data. Specifically, this includes: configuring a first priority parameter for audio frame data with the wearer's own audio marker to obtain first-class processed audio frame data, where the first priority parameter indicates that the first-class processed audio frame data has the highest processing priority; configuring a second priority parameter for audio frame data with the interactive object's audio marker to obtain second-class processed audio frame data, where the second priority parameter indicates that the second-class processed audio frame data has the second-highest processing priority; and performing volume attenuation processing on audio frame data with the wearer's non-interactive audio marker, and... The attenuated audio frame data is configured with a third priority parameter to obtain the third type of processed audio frame data. The third priority parameter is used to indicate that the third type of processed audio frame data has the lowest processing priority. The audio frame data with environmental background audio tags is subjected to privacy desensitization processing to obtain the fourth type of processed audio frame data. The first type of processed audio frame data is encapsulated with the corresponding classification tag and corresponding audio data acquisition timestamp information, the second type of processed audio frame data is encapsulated with the corresponding classification tag and corresponding audio data acquisition timestamp information, the third type of processed audio frame data is encapsulated with the corresponding classification tag and corresponding audio data acquisition timestamp information, and the fourth type of processed audio frame data is encapsulated with the corresponding classification tag and corresponding audio data acquisition timestamp information to generate the target audio stream data.

[0056] In the above embodiment, assume a bank's wealth management service center where financial advisors wearing recording badges provide investment consultation services to clients. The service center employs a semi-open consultation area, with multiple advisors simultaneously serving different clients, resulting in a large amount of cross-talk and background noise. The recording badges have already categorized and labeled the collected audio; now, differentiated processing of different categories of audio frames is needed to generate structured, distributed data. When a financial advisor introduces the returns of a wealth management product to a client, the audio frames generated by their speech are labeled as the wearer's own audio. A first priority parameter is configured for these audio frames, represented internally as a binary identifier sequence, with the highest bit set to 1, indicating the highest processing priority. After configuring the first priority parameter, these audio frames become the first category of processed audio frame data. This highest priority ensures that the financial advisor's professional explanations are transmitted and processed preferentially, without delay due to network congestion or processing queue backlog, guaranteeing the integrity and timeliness of service records. When a client inquires about the product's risk level and redemption conditions, these audio frames are labeled as interactive audio. These audio frames are configured with a second priority parameter, where the second-highest bit in the internal identifier sequence is set to 1, and the other bits are 0, indicating the second-highest processing priority. These audio frames configured with the second priority parameter become the second type of processed audio frame data. This second-highest priority ensures that customer questions and needs can be recorded and analyzed in a timely manner. Although the priority is slightly lower than the consultant's explanation, it is still higher than other non-core audio, guaranteeing the complete recording of the two-way dialogue.

[0057] In the above embodiment, during the service process, financial advisors occasionally turn their heads to briefly exchange work arrangements with passing colleagues; these audio frames are marked as non-interactive audio from the wearer. Volume attenuation is performed on these audio frames, specifically by multiplying the amplitude value of the audio signal by an attenuation coefficient. The attenuation coefficient is set to 0.3, meaning the audio volume is reduced to 30% of its original level. After attenuation, the audio still retains identifiable speech content, but the volume is significantly reduced, minimizing interference with the main business audio. A third priority parameter is configured for the attenuated audio frame data, with only the least significant bit in the identifier sequence set to 1, indicating the lowest processing priority. These become third-class processed audio frame data, which are placed at the end of the queue during transmission and storage, and processed only when system resources are sufficient. Other financial advisors at nearby consultation booths are serving another client; their conversations are collected and marked as ambient background audio. Privacy anonymization processing is performed on these audio frame data. Desensitization can be achieved using spectral scrambling techniques. First, the audio frame undergoes a Fast Fourier Transform (FFT) to obtain its frequency domain representation. Then, a pseudo-random sequence is used to perturb the phase of the spectral components, disrupting the temporal relationships of the frequency components. The perturbed spectrum is then inversely transformed back to the time domain. The resulting audio signal retains the original energy distribution characteristics and can be used for environmental noise level assessment, but the speech content is completely unrecognizable, making it impossible to reconstruct any meaningful dialogue. This processed audio frame becomes the fourth type of processed audio frame data, which is not configured with transmission priority parameters because this type of data is primarily used for local environmental analysis and does not require real-time uploading.

[0058] In the above embodiments, after completing the differentiation processing, data encapsulation begins. For the first type of processed audio frame data, the audio frame in which the financial advisor explains the product features is associated with the classification tag of the wearer's own audio corresponding to that audio frame. Simultaneously, audio data acquisition timestamp information is added, accurate to the millisecond level, recorded in the format YYYY-MM-DD HH:MM:SS.mmm. This information is organized into a structured data packet, containing a metadata area in the audio frame header and an audio data area. The metadata area stores priority parameters, classification tags, and timestamps, while the data area stores processed audio samples. The encapsulation process for the second type of processed audio frame data is similar; the audio frame in which the customer asks a question is associated with the interactive object's audio tag, and the timestamp accurately records the moment the customer asks the question. A special association identifier is added during encapsulation to establish a logical connection between the customer's question audio and the preceding and following financial advisor's answer audio, facilitating the subsequent reconstruction of the complete dialogue scenario. For the third type of processed audio frame data, in addition to including the attenuated audio data, the wearer's non-interactive audio tag, and the timestamp, an attenuation processing identifier and attenuation coefficient value are also added during encapsulation, facilitating the attempt to restore the original volume if necessary. Because they have the lowest priority, these packets are placed at the end of the buffer queue.

[0059] In the above embodiments, the encapsulation of the fourth type of processed audio frame data places greater emphasis on privacy protection. The desensitized audio data is associated with environmental background audio markers, and a timestamp records the collection time of the environmental sound. A desensitized status indicator is specifically marked during encapsulation, clearly indicating that the audio has undergone privacy processing and does not contain identifiable speech content. After all four types of processed audio frame data are encapsulated, they together constitute the target audio distribution data. This data structure is a multi-layered organizational system. The top layer contains the overall information of this data packet, such as the data packet sequence number, total number of frames, and statistics on the number of audio frames of each type. The middle layer organizes different categories of audio data according to priority and classification, with each category maintaining an independent data queue. The bottom layer contains specific audio frame data packets, each packet completely containing audio content, processing status, classification information, and time information. The generated target audio distribution data has a clear structure and a well-defined processing strategy. The first and second types of processed audio frames are preferentially transmitted to the management platform via the network for real-time or near-real-time service quality monitoring and analysis. The third type of processed audio frames are selectively uploaded or periodically cleaned up based on local storage space and network conditions. The fourth type of processed audio frames are primarily stored locally for work environment assessment. The original audio information has been securely processed, completely eliminating the risk of privacy leaks. This differentiated processing and structured encapsulation not only solves the problems caused by the indiscriminate collection of recordings from work badges, but also improves data processing efficiency and the level of audio privacy protection.

[0060] In an optional embodiment, privacy desensitization processing is performed on audio frame data with environmental background audio tags to obtain fourth-type processed audio frame data. Specifically, this includes: performing time-frequency transformation processing on the audio frame data with environmental background audio tags to obtain time-frequency domain representation data; determining the energy density value of each time-frequency point based on the energy distribution in the time-frequency domain representation data, and determining the region of continuous time-frequency points with energy density values ​​greater than a preset density threshold as the region to be processed; performing phase randomization processing on the time-frequency domain representation data within the region to be processed to obtain phase-perturbed time-frequency data, wherein the phase-perturbed time-frequency data is time-frequency data with the same amplitude spectrum structure but a different phase spectrum structure as the time-frequency domain representation data; performing downsampling processing on the time-frequency domain representation data outside the region to be processed to obtain compressed time-frequency data; merging the phase-perturbed time-frequency data and the compressed time-frequency data to obtain desensitized time-frequency data; and performing inverse time-frequency transformation processing on the desensitized time-frequency data to obtain fourth-type processed audio frame data.

[0061] In the above embodiment, it is assumed that in the inpatient nursing station of a general hospital, nurses wearing recording badges conduct daily ward rounds and execute medical orders. The nursing station is an open work area, surrounded by various sound environments such as other nurses handing over shifts, doctors discussing patients' conditions, and patients' families consulting. A certain audio frame has been identified as ambient background audio, specifically the conversation between patients' families in the next ward discussing treatment plans. This type of audio, containing private information, requires thorough desensitization. First, time-frequency transformation is performed on this audio frame data marked with ambient background audio. The short-time Fourier transform method is used to convert the time-domain audio signal to the time-frequency domain. The transformation process uses the Hanning window function, with a window length of 512 sampling points and an overlap of 256 sampling points between adjacent windows. After transformation, the time-frequency domain representation data is obtained, which is a two-dimensional complex matrix. The horizontal axis represents time frames, and the vertical axis represents frequency components. Each matrix element contains the amplitude and phase information of that time-frequency point. Energy distribution characteristics are analyzed in the time-frequency domain representation data. Energy density values ​​were calculated for each time-frequency point by taking the modulus squared of a complex number. Observations revealed that the speech energy of conversations between patients' family members was concentrated in specific time-frequency regions, particularly high in the fundamental frequency and formant range, while lower in other frequency regions. A preset density threshold was set at 1.5 times the overall average energy density. This threshold was determined based on statistical analysis of a large amount of environmental audio data and effectively distinguishes speech components from background noise.

[0062] In the above embodiments, multiple time-frequency points with energy density values ​​greater than a preset density threshold are identified through point-by-point comparison. These high-energy points often exhibit a continuous distribution, forming several connected regions. Using a connected component analysis algorithm, adjacent high-energy time-frequency points are merged to form regions to be processed. For example, within the frequency range of 300Hz to 3400Hz, from the 10th to the 45th time frame, a large continuous region is formed, corresponding to a complete sentence. Additionally, within the range of 500Hz to 2500Hz from the 60th to the 80th time frames, there is another region to be processed, corresponding to the main energy distribution of another sentence. For the identified regions to be processed, phase randomization is performed. Specifically, the amplitude value of each time-frequency point remains unchanged, but the phase value is replaced with a randomly generated phase. The random phase is generated using a pseudo-random number generator, uniformly distributed between negative π and positive π. For example, the complex representation of a certain time-frequency point is originally 3.5 × e (j×0.8π) After phase randomization, it may become 3.5×e [j×(－0.3π)]The amplitude remains 3.5, but the phase changes from 0.8π to -0.3π. This processing preserves the energy characteristics of the original audio, making the processed audio similar to the original in terms of perceived loudness. However, due to the randomization of the phase, the waveform of the time-domain signal is completely altered, and the intelligibility of the speech is utterly destroyed, making it impossible to identify the specific language content. The processed data is phase-perturbed time-frequency data, which has the same amplitude spectrum structure as the original time-frequency domain representation data but a completely different phase spectrum structure.

[0063] In the above embodiments, for time-frequency data outside the processing area, which mainly consists of low-energy background noise components, downsampling is performed to reduce the data volume. Downsampling is performed simultaneously in the frequency and time dimensions. In the frequency dimension, one out of every four frequency points is sampled, reducing the frequency resolution to one-quarter of the original. In the time dimension, one out of every two time frames is sampled, reducing the time resolution to one-half of the original. For example, a 16×8 time-frequency data block outside the processing area becomes a 4×4 data block after downsampling, reducing the data volume to one-eighth of the original. This compression does not affect privacy protection because these areas do not contain clear voice information, and it also reduces subsequent storage and transmission overhead, resulting in compressed time-frequency data. The phase-perturbed time-frequency data and the compressed time-frequency data are merged to reconstruct a complete time-frequency representation. During merging, attention needs to be paid to data alignment and boundary handling. For the downsampled area, interpolation methods are used to restore the original time-frequency resolution, but interpolation is only to maintain the integrity of the data structure and does not restore the information content. After merging, the desensitized time-frequency data is obtained. This data retains the basic energy distribution characteristics of the original audio and can be used to assess the level of environmental noise, but the speech content in it is completely unrecoverable.

[0064] In the above embodiments, the desensitized time-frequency data undergoes inverse time-frequency transform processing to convert it back into a time-domain audio signal. The inverse transform employs an inverse short-time Fourier transform, using the same window function and overlap parameters as the forward transform to ensure consistency. Since the phase of the region to be processed has been randomized, the time-domain signal after the inverse transform retains the approximate energy envelope of the original audio, but the waveform details are completely different. The originally clear sentence, "How much does this treatment plan cost?" becomes, after processing, a sound with similar energy fluctuations but completely incomprehensible, sounding like some kind of modulated noise. The inverse transform yields the fourth type of processed audio frame data. This audio data effectively protects the privacy of other people in the environment; even if this data is intercepted or leaked, no meaningful dialogue content can be extracted from it. Simultaneously, because the energy distribution characteristics are preserved, this data can still be used for applications that do not involve specific content, such as environmental noise assessment and work environment analysis. Through this refined privacy desensitization process, complete desensitization of environmental background audio is achieved. Compared to simple muting or deletion, this method preserves the statistical characteristics of the audio, facilitating environmental analysis. Compared to traditional encrypted storage, this method fundamentally destroys the voice information, making it impossible to recover the original content even if the data is illegally obtained. The entire process is completed locally on the recording ID, avoiding the uploading of original audio containing others' privacy to the cloud, thus solving the privacy leakage risks inherent in traditional solutions at the source.

[0065] In an optional embodiment, after encapsulating the first type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information, the second type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information, the third type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information, and the fourth type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information to generate target audio stream data, the method further includes: maintaining the original audio data format of the first type of processed audio frame data, streaming the first type of processed audio frame data to the core working audio channel, adding an upload timestamp tag to the first type of processed audio frame data, and then streaming it through the core working audio channel. Upload to the target server; divert the second type of processed audio frame data to the interactive audio channel, and add interactive object identifiers and interaction type tags to the original audio data format of the second type of processed audio frame data before uploading it to the server through the interactive audio channel; divert the third type of processed audio frame data to the low-priority cache channel, and when the local cache space reaches the preset capacity threshold, perform batch compression processing on the third type of processed audio frame data before uploading it to the target server through the low-priority cache channel; divert the fourth type of processed audio frame data to the local backup channel, and store the fourth type of processed audio frame data in segments according to the preset time window. When the storage time of the fourth type of processed audio frame data exceeds the preset retention period, delete the fourth type of processed audio frame data.

[0066] In the above embodiment, assume an insurance consultant wearing a recording badge is handling insurance claims for a customer at a customer service center of an insurance company. The service center has an open-plan layout, with multiple consultants simultaneously serving different customers, creating an environment filled with various conversations and office noise. The audio data has already been classified, processed, and packaged, generating target audio stream data containing four types of processed audio frames. Now, this data needs to be transmitted and stored in a differentiated manner through different channels. The insurance consultant is explaining the required materials list for a claim to a customer; this audio has been processed into the first type of processed audio frame data. The original audio data format of these audio frames is maintained, i.e., 16-bit quantization precision and 16kHz sampling rate PCM encoding format, ensuring the integrity of the sound quality. This data is streamed to the core working audio channel, a transmission channel specifically designed for high-priority business audio, with independent buffers and transmission queues. Before uploading, an upload timestamp is added to each audio frame, recorded in the format YYYY-MM-DD HH:MM:SS.mmm. This timestamp differs from the audio acquisition timestamp; it records the moment the data is ready to be uploaded and is used to monitor transmission latency. These audio data, with double timestamps, are immediately uploaded to the target server via the core working audio channel. Because the original format is preserved and a dedicated channel is used, the audio quality received by the server is completely consistent with that acquired locally, enabling accurate speech recognition and content analysis.

[0067] In the above embodiment, when a customer inquires about the review timeframe and payment method for a claim, this audio has been processed into second-type processed audio frame data. This data is then diverted to the interactive audio channel, a dedicated transmission channel for handling customer interactions. Before data transmission, an interaction object identifier is added to the metadata area of ​​the audio frame. This identifier is generated using a simple sequence number, such as CUST_001, indicating an interaction with the first customer. An interaction type marker is also added, determining whether the interaction is a claims consultation based on the business scenario. This identifier information is embedded in the extended fields of the audio data format, providing important contextual information without affecting the audio playback itself. For example, a custom INFO block is added after the original WAV format audio file, including the CUST_001 identifier and the claims consultation marker. After being uploaded through the interactive audio channel, the server can quickly locate and associate relevant business records based on these identifiers, improving the efficiency of subsequent analysis. During breaks in processing claims, insurance advisors briefly exchange information about other customers with passing colleagues; this audio has been processed into third-type processed audio frame data and its volume has been attenuated. This data is diverted to a low-priority cache channel and initially stored in a local circular buffer. The buffer size is set to 50MB. When the accumulated third-type audio data approaches this size, a preset capacity threshold is reached. At this point, batch compression processing is triggered. An audio encoder converts the cached PCM format audio into a compressed format. The compression algorithm uses adaptive differential pulse code modulation, which can reduce the data size to about one-quarter of the original size. The compressed audio data is packaged into a batch file, with an appended batch number and time range information, and then uploaded to the target server through a low-priority cache channel. This batch processing method avoids frequent network connections, reduces transmission overhead, and because it is low-priority data, the upload process does not affect the real-time transmission of core business audio.

[0068] In the above embodiment, the conversations between other consultants and clients in the surrounding area have been anonymized for privacy, becoming the fourth type of processed audio frame data. This data is diverted to a local backup channel, a purely local storage channel without network transmission. The data is segmented and stored according to a preset time window, with each 5-minute segment of audio data stored as a separate file. The file name includes start and end time information, such as ENV_16650720_142300_142800.dat representing environmental audio data from 2:23 PM to 2:28 PM. These files are stored in the local storage space of the recording badge, using a cyclic overwrite strategy. The preset retention period is set to 24 hours, determined based on actual business needs, satisfying the need for daily work environment review without consuming excessive storage space. When the storage time of an environmental audio file exceeds 24 hours, it is automatically deleted. The deletion process uses a secure erasure method, not only deleting the index in the file system but also overwriting the storage area, ensuring that the data cannot be recovered by technical means. For example, the file ENV_16650719_142300_142800.dat, generated at 2:23 PM the previous day, was identified as an expired file and immediately deleted securely at 2:23 PM the following day. This automatic cleanup mechanism ensures efficient use of local storage space, and because these audio files have been anonymized, privacy will not be compromised even if the device is lost during storage. This multi-channel split-transmission mechanism enables precise management of different types of audio data. Core working audio is uploaded in real-time through a dedicated channel, ensuring timely service quality monitoring. Interactive audio carries rich identification information, facilitating business correlation and analysis. Non-interactive audio is batch compressed and uploaded with delays, preserving complete work records without affecting core business operations. Environmental audio is stored locally only for a short period, meeting environmental monitoring needs while completely avoiding privacy risks.

[0069] This application's embodiments achieve intelligent identification and selective processing of audio from different sources at the data collection source for recording badges. Through a dual-dimensional determination using microphone array spatial positioning and voiceprint recognition, it can accurately distinguish four types of data: wearer's work conversations, customer interaction content, non-interactive audio, and environmental background audio. Differentiated processing strategies ensure priority transmission of core business audio, while local anonymization of environmental audio fundamentally avoids the risk of leakage of irrelevant personnel's privacy information. A multi-channel split transmission mechanism precisely manages audio based on type and priority, significantly reducing network bandwidth consumption and server processing load. Compared to traditional cloud processing solutions, it not only improves real-time performance and reliability but also significantly reduces system costs. This end-to-end privacy protection solution meets the principle of data minimization, effectively protecting the privacy and security of all personnel in the work environment while ensuring service quality management requirements.

[0070] The recording badge in the embodiments of this invention is described below from the perspective of hardware processing. (See attached document.) Figure 2 , Figure 2 This is a schematic diagram of the physical device structure of a recording badge in the embodiments of this application.

[0071] It should be noted that, Figure 2 The structure of the recording badge shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0072] like Figure 2 As shown, the recording badge includes a Central Processing Unit (CPU) 201, which can perform various appropriate actions and processes according to a program stored in Read-Only Memory (ROM) 202 or a program loaded from storage section 208 into Random Access Memory (RAM) 203, such as performing the methods described in the above embodiments. The RAM 203 also stores...

[0073] It contains various programs and data required for system operation. CPU 201, ROM 202, and RAM 203 are interconnected via bus 204. Input / output (I / O) interface 205 is also connected to bus 204.

[0074] The following components are connected to I / O interface 205: input section 206 including audio input devices, push-button switches, etc.; output section 207 including a liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 208 including a hard disk, etc.; and communication section 209 including a network interface card such as a LAN (Local Area Network) card, modem, etc. Communication section 209 performs communication processing via a network such as the Internet. Drive 210 is also connected to I / O interface 205 as needed. Removable media 211, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., are installed on drive 210 as needed so that computer programs read from them can be installed into storage section 208 as needed.

[0075] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 209, and / or installed from removable medium 211. When the computer program is executed by central processing unit (CPU) 201, it performs the various functions defined in the present invention.

[0076] It should be noted that specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0077] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0078] Specifically, the recording badge in this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, it implements the recording badge data diversion processing method based on interactive area determination provided in the above embodiment.

[0079] In another aspect, the present invention also provides a computer-readable storage medium, which may be included in the recording badge described in the above embodiments; or it may exist independently and not assembled into the recording badge. The storage medium carries one or more computer programs, which, when executed by a processor of the recording badge, cause the recording badge to implement the recording badge data diversion processing method based on interaction area determination provided in the above embodiments.

[0080] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0081] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for data diversion processing of recording ID cards based on interaction area determination, characterized in that, include: During the initialization phase of the recording badge, the wearer's standard voice sample is acquired through the audio acquisition unit of the recording badge, and the voiceprint feature vector extracted from the standard voice sample is used as the wearer's voiceprint reference and stored in the local storage unit of the recording badge. During the recording work card's work acquisition phase, audio signals are acquired in real time through a microphone array integrated into the recording work card. Based on the arrival time difference and energy intensity difference of the audio signals received by different microphones in the microphone array, the spatial location parameters of the audio source that generates the audio signal are determined. An effective interaction area is constructed based on the spatial location parameters. The effective interaction area is a cone-shaped spatial area extending in front of the wearer with the recording badge as the vertex. The audio signal is subjected to short-time frame segmentation to obtain audio frame sequence data. Each audio frame in the audio frame sequence data is subjected to region attribution determination processing according to the spatial location parameters to obtain region attribution determination result. Furthermore, each audio frame is subjected to voiceprint matching determination processing according to the wearer's voiceprint reference to obtain voiceprint matching determination result. Based on the region attribution determination result and the voiceprint matching determination result, a corresponding classification label is assigned to each audio frame data to obtain the target audio split data.

2. The method according to claim 1, characterized in that, The process of performing region attribution determination on each audio frame in the audio frame sequence data based on the spatial location parameters to obtain a region attribution determination result, and performing voiceprint matching determination on each audio frame based on the wearer's voiceprint reference to obtain a voiceprint matching determination result, specifically includes: The relative spatial relationship between the frame sound source position of each audio frame data and the effective interaction area is determined according to the spatial position parameters. The region attribution determination result corresponding to the frame sound source position located within the cone-shaped spatial area is marked as within the region according to the relative spatial position relationship. The region attribution determination result corresponding to the frame sound source position located outside the cone-shaped spatial area is marked as outside the region according to the relative spatial position relationship. Voiceprint feature vectors are extracted from each audio frame data, and the voiceprint feature vectors are compared with the wearer's voiceprint reference to obtain the similarity comparison results; Based on the similarity comparison results, the voiceprint matching determination results corresponding to audio frame data with similarity values ​​greater than the preset similarity threshold are marked as matched, and based on the similarity comparison results, the voiceprint matching determination results corresponding to audio frame data with similarity values ​​less than or equal to the preset similarity threshold are marked as mismatched.

3. The method according to claim 2, characterized in that, The step of determining the relative spatial relationship between the frame sound source position of each audio frame data and the effective interaction area based on the spatial position parameters, and marking the area attribution determination result corresponding to the frame sound source position located within the cone-shaped spatial area as within the area based on the relative spatial position relationship, and marking the area attribution determination result corresponding to the frame sound source position located outside the cone-shaped spatial area as outside the area based on the relative spatial position relationship, specifically includes: The azimuth angle, pitch angle, and spatial distance parameters of the frame sound source position are extracted from the spatial position parameters. The azimuth angle parameter represents the deviation angle of the frame sound source position on the horizontal plane relative to the forward central axis of the recording sign. The pitch angle parameter represents the deviation angle of the frame sound source position on the vertical plane relative to the forward central axis of the recording sign. The spatial distance parameter represents the spatial distance between the frame sound source position and the recording sign. Determine whether the azimuth parameter is within a preset horizontal angle range, and determine whether the pitch angle parameter is within a preset vertical angle range, and determine whether the spatial distance parameter is less than a preset extension distance of the conical spatial region; When the azimuth parameter is within the horizontal angle range, the pitch angle parameter is within the vertical angle range, and the spatial distance parameter is less than the preset extension distance, the relative spatial position relationship is determined to be within the region, and the region attribution determination result of the audio frame data corresponding to the frame sound source position is marked as within the region; When the azimuth parameter is not within the horizontal angle range, or the pitch angle parameter is not within the vertical angle range, or the spatial distance parameter is greater than or equal to the preset extension distance, the relative spatial position relationship is determined to be outside the region, and the region attribution determination result of the audio frame data corresponding to the frame sound source position is marked as outside the region.

4. The method according to claim 2, characterized in that, The step of assigning a corresponding classification label to each audio frame data based on the region attribution determination result and the voiceprint matching determination result to obtain target audio stream data specifically includes: For audio frame data where the region attribution determination result is within the region and the voiceprint matching determination result is a match, assign the wearer's own audio tag; For audio frame data where the region attribution determination result is within the region and the voiceprint matching determination result is not a match, an interactive object audio tag is assigned; For audio frame data where the region attribution determination result is outside the region and the voiceprint matching determination result is a match, assign a wearer non-interactive audio tag; For audio frame data where the region attribution determination result is outside the region and the voiceprint matching determination result is not a match, an environmental background audio tag is assigned; The target audio stream data is obtained by differentiating each audio frame data based on the wearer's own audio tag, the interactive object's audio tag, the wearer's non-interactive audio tag, and the environmental background audio tag.

5. The method according to claim 4, characterized in that, The step of differentiating each audio frame data based on the wearer's own audio marker, the interactive object's audio marker, the wearer's non-interactive audio marker, and the environmental background audio marker to obtain the target audio splitting data specifically includes: A first priority parameter is configured for audio frame data with the wearer’s own audio tag to obtain a first type of processed audio frame data. The first priority parameter is used to indicate that the first type of processed audio frame data has the highest processing priority. Configure a second priority parameter for audio frame data with the audio tag of the interactive object to obtain second-class processed audio frame data. The second priority parameter is used to indicate that the second-class processed audio frame data is the second highest processing priority. The audio frame data with the wearer's non-interactive audio tag is subjected to volume attenuation processing, and a third priority parameter is configured for the attenuated audio frame data to obtain third-class processed audio frame data. The third priority parameter is used to indicate that the third-class processed audio frame data is the lowest processing priority. The audio frame data with the aforementioned environmental background audio tags are subjected to privacy desensitization processing to obtain the fourth type of processed audio frame data; The first type of processed audio frame data, along with its corresponding classification tag and corresponding audio data acquisition timestamp information, the second type of processed audio frame data, along with its corresponding classification tag and corresponding audio data acquisition timestamp information, the third type of processed audio frame data, along with its corresponding classification tag and corresponding audio data acquisition timestamp information, and the fourth type of processed audio frame data, along with their corresponding classification tags and corresponding audio data acquisition timestamp information, are encapsulated to generate target audio stream data.

6. The method according to claim 5, characterized in that, The process of performing privacy-desensitizing processing on the audio frame data with the environmental background audio marker to obtain the fourth type of processed audio frame data specifically includes: The audio frame data with the aforementioned environmental background audio markers are subjected to time-frequency transformation processing to obtain time-frequency domain representation data; The energy density value of each time-frequency point is determined based on the energy distribution in the time-frequency domain representation data, and the continuous time-frequency point region with the energy density value greater than the preset density threshold is determined as the region to be processed. Phase randomization processing is performed on the time-frequency domain representation data within the region to be processed to obtain phase-perturbed time-frequency data. The phase-perturbed time-frequency data is time-frequency data with the same amplitude spectrum structure but a different phase spectrum structure as the time-frequency domain representation data. Downsampling is performed on the time-frequency domain representation data outside the region to be processed to obtain compressed time-frequency data; The phase perturbation time-frequency data and the compressed time-frequency data are merged to obtain desensitized time-frequency data; The desensitized time-frequency data is subjected to inverse time-frequency transformation to obtain the fourth type of processed audio frame data.

7. The method according to claim 5, characterized in that, After encapsulating the first type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information, the second type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information, the third type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information, and the fourth type of processed audio frame data with its corresponding classification tag and corresponding audio data acquisition timestamp information to generate target audio streaming data, the method further includes: Maintaining the original audio data format of the first type of processed audio frame data, the first type of processed audio frame data is diverted to the core working audio channel, and after adding an upload timestamp to the first type of processed audio frame data, it is uploaded to the target server through the core working audio channel. The second type of processed audio frame data is diverted to the interactive audio channel, and after adding an interactive object identifier and an interactive type marker to the original audio data format of the second type of processed audio frame data, it is uploaded to the server through the interactive audio channel. The third type of processed audio frame data is diverted to a low-priority cache channel, and when the local cache space reaches a preset capacity threshold, the third type of processed audio frame data is batch compressed and then uploaded to the target server through the low-priority cache channel. The fourth type of processed audio frame data is diverted to a local backup channel and stored in segments according to a preset time window. When the storage time of the fourth type of processed audio frame data exceeds the preset retention period, the fourth type of processed audio frame data is deleted.

8. A recording work badge, characterized in that, The recording badge includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the recording badge to perform the method as described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, When the instruction is executed on the recording badge, the recording badge causes the recording badge to perform the method as described in any one of claims 1-7.

10. A computer program product, characterized in that, When the computer program product is run on the recording badge, the recording badge performs the method as described in any one of claims 1-7.

Images