Intelligent alarm monitoring and listening system

Abstract

The application discloses a kind of wisdom alarm monitoring listening system, belong to intelligent security technology field, including time delay calibration module, double-path alignment module, audio repair module, monitoring and early warning module and audio reconstruction module.Time delay calibration module obtains and updates baseline time delay difference by hop-by-hop forwarding calibration and periodic detection;Double-path alignment module calculates expected timestamp in combination with baseline time delay difference, and obtains double-path sequence by frame matching;Audio repair module locates misaligned broken section and analyzes feature similarity, to obtain continuous audio frame sequence;Monitoring and early warning module collects signal quality parameters, identifies link degradation trend and triggers switching warning;Audio reconstruction module identifies loss window and analyzes boundary audio time-domain evolution law, to obtain complete audio frame.The application effectively solves the problem of audio timing misalignment and switching window packet loss caused by asymmetric double-loop wiring, significantly improves the accuracy of voiceprint recognition and alarm reliability.

Description

Technical Field

[0001] This invention relates to the field of intelligent security technology, and in particular to an intelligent alarm monitoring and listening system. Background Technology

[0002] To improve communication reliability, existing intelligent alarm monitoring systems generally adopt a dual-ring self-healing redundancy architecture for transmitting audio data. However, this architecture has several hidden defects in practical applications. For example, the physical wiring lengths of the primary and backup rings are difficult to be completely consistent, resulting in inherent transmission delay differences. The receiving end simply discards delayed audio segments after deduplication, leading to misalignment of short distress call voiceprint segments and difficulty in fully extracting dangerous voiceprint features for voiceprint recognition. Another example is the existence of a millisecond-level timing window during the dual-ring self-healing switchover process. The distress call audio collected within the window is completely lost, and there is no instantaneous gap in packet loss log records, causing silent missed alarms.

[0003] The aforementioned defects stem from the fact that the dual-ring redundancy design focuses on ensuring link connectivity while neglecting the stringent requirements of audio signals for temporal continuity: the fixed delay difference caused by asymmetrical wiring leads to misalignment of arrival times for audio frames from the same source. When the receiver processes the deduplication according to the arrival order, short-duration distress call voiceprints are cut into discontinuous segments, directly causing voiceprint recognition failure; the hardware switching delay during loop switching creates a data vacuum window. If the emergency happens to occur within this window, the alarm system will have difficulty detecting the abnormality on site, ultimately leading to missed reports of critical emergencies and severely weakening the early warning reliability and emergency response capability of the security system. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent alarm monitoring and listening system to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides an intelligent alarm monitoring and listening system, including a time delay calibration module, a dual-channel alignment module, an audio repair module, a monitoring and early warning module, and an audio reconstruction module; The latency calibration module is used to obtain the baseline latency difference and update the baseline latency difference by adding timestamps and periodic probes through hop-by-hop forwarding of each monitoring terminal. The dual-path alignment module is used to construct a dual-path frame queue and, in conjunction with the updated baseline delay difference, calculate the expected acquisition timestamp. After frame matching, the dual-path frame queue is updated to obtain the dual-path sequence. The audio restoration module is used to locate misaligned and broken segments based on dual-channel sequences, and analyze the feature similarity of the audio at the boundary of the misaligned and broken segments to obtain a temporally continuous audio frame sequence. The monitoring and early warning module is used to collect signal quality parameters and analyze the changing trend of loop signal quality to identify link degradation trends. When a link degradation trend is present, a handover warning is triggered. The audio reconstruction module is used to receive switching warnings, identify audio frame loss windows, and analyze the temporal acoustic evolution of the audio at both boundaries of the audio frame loss window to obtain the completed audio frame, so as to realize audio alarm monitoring and waveform compensation.

[0006] The advantages and positive effects of the intelligent alarm monitoring and listening system described in this invention are: 1. This invention breaks through the traditional dual-ring redundancy design that only focuses on link connectivity. It deeply integrates dynamic latency difference calibration with audio frame timing splicing. Through a three-level mechanism of baseline calibration, real-time tracking, and drift correction, it continuously optimizes the dual-ring latency difference benchmark. Combining sequence number and timestamp dual-dimensional indexing with MFCC feature-based optimal splicing, it fundamentally solves the audio timing misalignment problem caused by asymmetrical dual-ring wiring. Unlike conventional solutions that simply deduplicate and discard lagging frames, this system uses feature matching to complete misaligned segments across paths, ensuring that short distress call voiceprints remain complete and continuous, significantly improving the voiceprint recognition accuracy in complex wiring scenarios.

[0007] 2. This invention achieves early warning of handover by constructing a collaborative mechanism of link quality gradient prediction and bidirectional LPC audio reconstruction. Then, it utilizes a weighted fusion of bidirectional linear prediction models of the preceding and following boundary frames to fill in the lost window. Unlike traditional passive handover solutions after a fault, this system predicts the handover tens of milliseconds in advance and initiates preparation for completion. It effectively mitigates the cumulative distortion problem of unidirectional prediction through bidirectional weighted fusion, smoothly filling in the audio gap at the moment of handover and significantly reducing the risk of missed alarms caused by loop handover.

[0008] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the structure of an intelligent alarm monitoring and listening system according to the present invention. Detailed Implementation

[0010] In the description of this invention, it should be noted that the terms "upper," "lower," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use. They are used only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," and "connect" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0011] In this application, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. In case of any inconsistency, the meaning set forth in this specification or derived from the content described herein shall prevail. Furthermore, the terminology used herein is for the purpose of describing embodiments of this application only and is not intended to limit the scope of this application.

[0012] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0013] Reference Figure 1 As shown, the present invention provides a technical solution: a smart alarm monitoring and listening system, including a time delay calibration module, a dual-channel alignment module, an audio repair module, a monitoring and early warning module, and an audio reconstruction module; The latency calibration module is used to obtain the baseline latency difference and update the baseline latency difference by adding timestamps and periodic probes through hop-by-hop forwarding of each monitoring terminal. The dual-path alignment module is used to construct a dual-path frame queue and, in conjunction with the updated baseline delay difference, calculate the expected acquisition timestamp. After frame matching, the dual-path frame queue is updated to obtain the dual-path sequence. The audio restoration module is used to locate misaligned and broken segments based on dual-channel sequences, and analyze the feature similarity of the audio at the boundary of the misaligned and broken segments to obtain a temporally continuous audio frame sequence. The monitoring and early warning module is used to collect signal quality parameters and analyze the changing trend of loop signal quality to identify link degradation trends. When a link degradation trend is present, a handover warning is triggered. The audio reconstruction module is used to receive switching warnings, identify audio frame loss windows, and analyze the temporal acoustic evolution of the audio at both boundaries of the audio frame loss window to obtain the completed audio frame, so as to realize audio alarm monitoring and waveform compensation.

[0014] In this embodiment, the baseline time delay difference of the dual-loop circuit is calibrated by the time delay calibration module and dynamically tracked and updated. Combined with the sliding window frame matching of the dual-path alignment module and the MFCC feature splicing of the audio repair module, the audio timing misalignment caused by the asymmetry of the dual-loop wiring is effectively eliminated. Simultaneously, the link quality gradient prediction of the monitoring and early warning module and the bidirectional LPC interpolation completion of the audio reconstruction module fill the millisecond-level data vacuum window caused by loop switching. Overall, the system forms a complete closed loop from delay calibration, frame alignment, misalignment repair to switching prediction and packet loss completion, significantly improving the audio timing continuity and voiceprint recognition reliability under the dual-ring redundant architecture.

[0015] For example, in underground utility tunnel security scenarios, when the main ring fiber deteriorates due to construction disturbances and the system triggers loop switching, traditional solutions will produce an audio gap of 2 to 5 milliseconds. If someone makes a short cry for help at this time, voiceprint recognition is very likely to miss it. However, this system can maintain the confidence level of audio reconstruction in the lost window at a high level by predicting the switch 20 to 50 milliseconds in advance and performing LPC interpolation to complete it, ensuring that the cry for help is complete and identifiable, and significantly reducing the false alarm rate.

[0016] Specifically, the delay calibration module includes: The message acquisition unit is used to synchronously send calibration messages to the dual-loop differential bus through the edge control host in the power-on initialization state. The calibration data stream is obtained by forwarding the calibration messages hop by hop through each listening terminal and adding local timestamps. This data stream is a complete time-series message sequence that is forwarded node by node in the dual loop, with node timestamps superimposed, and then transmitted back to the edge host along the loop. It is the data carrier for performing delay calculations.

[0017] The power-on initialization state refers to the edge control host and all monitoring terminals being powered on and started up. The hardware has completed register reset, clock phase-locking, and communication interface initialization. The system has not yet entered the pre-operation stage of normal audio acquisition and data transmission. The edge control host is an embedded main control device deployed at the central node of the security system. It is equipped with a dual-loop differential communication interface and an audio timing processing unit, and is responsible for bus message transmission and reception, latency calculation, audio frame timing alignment and overall analysis.

[0018] The dual-loop differential bus is a redundant communication bus consisting of two independent differential twisted-pair loops with opposite transmission directions. It uses differential levels to transmit audio and commands, providing redundant data channels for the host and all monitoring terminals. The calibration message is a dedicated test message issued by the edge host during the power-on initialization phase. It carries timestamp information and is used to calculate the dual-loop transmission delay. It is only used in the initialization calibration process and is different from the service audio data frame. The monitoring terminal is a front-end device deployed at the on-site monitoring point, responsible for collecting environmental audio and forwarding various messages on the bus; The baseline solving unit is used to extract an ordered set of timestamps from the calibration data stream to calculate the total transmission delay of the primary and backup loops. After difference calculation, the baseline delay difference is obtained and stored in a dedicated register. The baseline delay difference is used as a timing alignment reference parameter to compensate for the audio frame arrival time deviation caused by the inconsistent wiring length of the primary and backup loops and to eliminate audio timing misalignment. Specifically, the edge control host's dual differential receiving ports receive calibration messages returned from the main loop and the backup loop respectively, parse the timestamp sequence of each node in each message, and then process them according to the formula. , Calculate the total transmission delay of the primary and backup loops respectively; and... and The absolute difference is used as the initial reference value. The calibration process is executed ten times consecutively, and the median of the ten calculation results is taken as the final initial baseline value. Single abnormal fluctuation data are eliminated to improve the stability and anti-interference ability of the baseline value.

[0019] , These represent the total transmission delay of the primary and backup loops, respectively. , These are the timestamps for receiving the primary and backup loop messages back to the host, respectively. A record of the instant the edge control host sends a calibration message; The hop-by-hop parsing unit is used to record the listening terminal as a node, and extract the hop-by-hop delay sequence of each node based on the receiving timestamp of each node in the calibration message and by calculating the difference between the timestamps of adjacent nodes.

[0020] Specifically, after parsing the timestamp sequence of each node in the calibration message, the edge control host calculates the hop-by-hop delay of each link segment based on the difference in received timestamps between adjacent nodes. The main loop direction calculates the main loop hop-by-hop delay sequence, and the backup loop direction calculates the backup loop hop-by-hop delay sequence. The hop-by-hop delay sequence is stored in a node delay table, which includes three fields: node number, main loop hop-by-hop delay, and backup loop hop-by-hop delay. This table records the transmission delay data of the corresponding link segment for each node, used to locate specific nodes with abnormal delay fluctuations.

[0021] The dynamic tracking unit is used to obtain the real-time dynamic delay difference based on the baseline delay difference and the hop-by-hop delay sequence through bidirectional acquisition of periodic probe frames and smoothing with first-order low-pass filtering. In practice, after the system enters normal operation, the edge host sends a probe frame to the dual-loop circuit at fixed time intervals. The probe frame carries an incrementing sequence number and a timestamp of the sending time. Upon receiving the probe frame, the monitoring terminal does not perform any additional processing and forwards it directly along the original path without appending a node timestamp, in order to reduce the error introduced by forwarding processing latency.

[0022] The edge host's dual receiving ports record the arrival times of the primary loop and backup loop probe frames respectively, according to the formula... Calculate the first Real-time delay difference of the second detection ,in Main ring road Frame arrival time, To prepare for the ring road Frame arrival time; For the first The detection period sequence number of the next detection frame; Delay difference samples are continuously collected for N detection cycles. A sliding window first-order low-pass filter algorithm is used to smooth the real-time delay difference. The filter formula is as follows: , where the filter coefficients The value ranges from 0.1 to 0.3 and is dynamically adjusted based on link stability. , The first , The time delay difference after secondary filtering and smoothing, i.e. the real-time dynamic time delay difference, is used to perform exponential moving average filtering on the original time delay difference to filter out sudden changes in time delay pulses caused by instantaneous disturbances in the cable and sampling jitter. The probe frame transmission interval is set to 200 milliseconds to balance tracking sensitivity and bus bandwidth usage. All raw probe arrival time data are stored in a latency difference history buffer with a capacity of 24 hours for the self-calibration module to perform backtesting statistics.

[0023] The drift correction unit is used to perform a difference operation between the real-time dynamic delay difference and the baseline delay difference to obtain the drift amount, and compare it with the preset drift interval to obtain the drift level flag. Through the baseline dynamic approximation write-back mechanism, the updated baseline delay difference is obtained, which serves as the latest reference for subsequent dual-loop audio frame timing matching.

[0024] In practice, the filtered value in the dynamic delay difference register and the initial value in the baseline delay difference register are continuously compared to obtain the drift amount. The system presets three drift thresholds: slight drift threshold, moderate drift threshold, and severe drift threshold. For example, the three thresholds are set proportionally to 1ms, 3ms, and 5ms.

[0025] The drift amount is compared sequentially with three threshold levels, and the corresponding drift level flag is output. When the drift amount is below the slight drift threshold, a stable flag is output, and the baseline value remains unchanged. When the drift amount is between the slight and moderate thresholds, a slow drift flag is output, and the baseline value approaches the real-time value in fixed steps, with the step size set to 0.1ms per detection cycle. This approach continues until the difference between the two falls within the slight drift threshold range, at which point the adjustment stops. After the approach is completed, the baseline delay difference register is immediately written back, and the register update flag is triggered. When the drift amount exceeds the moderate threshold and persists for more than five detection cycles, a significant drift flag is output, triggering the baseline recalibration process. If the recalibration result is still a significant drift, the baseline is no longer iteratively updated. The system locks the baseline value obtained from this recalibration to maintain timing alignment and continuously outputs link anomaly alarms. When the drift amount exceeds the severe threshold, an anomaly flag is output, and the link quality alarm register is set.

[0026] The drift sequence, threshold comparison results, and baseline adjustment records generated during the drift determination process are all stored in the time delay difference history buffer for the self-calibration module to perform long-term drift trend analysis.

[0027] Baseline value refers to the baseline delay difference; real-time value refers to the real-time dynamic delay difference. In this embodiment, the delay calibration module obtains the complete timing data stream by synchronously sending calibration messages through the message acquisition unit during the power-on initialization phase. The initial baseline delay difference is calculated by taking the median after ten consecutive calibrations by the baseline solving unit. Combined with the hop-by-hop parsing unit, the delay sequence of each node is extracted to locate abnormal nodes. Then, the dynamic tracking unit sends probe frames at a period of 200 milliseconds and smooths the real-time delay difference through a first-order low-pass filter. Finally, the baseline adaptive update is achieved through the three-level threshold judgment and dynamic approximation write-back mechanism of the drift correction unit.

[0028] Overall, the module forms a complete time delay calibration closed loop from initial calibration and hop-by-hop parsing to real-time tracking and drift correction. This ensures the stability and anti-interference capability of the baseline value, while also sensitively capturing the slow drift of link delay, providing a precise benchmark for subsequent audio frame timing alignment.

[0029] For example, in outdoor utility tunnels where there are large temperature differences between day and night in winter, the thermal expansion and contraction of cables causes the transmission delay to change slowly. Traditional fixed baseline solutions can have a deviation of several milliseconds after a week, causing audio frame misalignment. This module, however, can control the baseline deviation by probing every 200 milliseconds and approximating with a step size of 0.1 milliseconds, thus ensuring the timing alignment accuracy of dual-channel audio.

[0030] Specifically, the dual-path alignment module includes: The receiving unit is used to collect audio data through the monitoring terminal, transmit it through dual-loop differential independent transmission, and obtain a dual-channel frame queue and a dual-channel audio index table. The dual-channel frame queue includes a main loop receiving queue and a backup loop receiving queue. In practice, the audio acquisition module of the monitoring terminal acquires ambient audio at a fixed sampling rate. Each acquired sample is encapsulated into a frame of audio data, and each frame carries two timing identifier fields. The first identifier field is the frame sequence number, generated by a locally incrementing counter on the terminal. The second identifier field is the acquisition timestamp, generated by a timer corresponding to the terminal's local audio sampling clock when the first sample of the frame is acquired.

[0031] After the audio frame is encapsulated, the terminal's dual-loop differential transmission module simultaneously writes the frame into both the main loop transmission buffer and the backup loop transmission buffer. The two buffers are independently scheduled for transmission without blocking each other. The edge control host's dual differential receiving ports receive audio frames transmitted from the main loop and backup loop respectively. Upon receiving each frame, its sequence number and acquisition timestamp are parsed and stored in the main loop receiving queue and backup loop receiving queue respectively. Each receiving queue maintains an index table, using the sequence number as the primary key and the acquisition timestamp as the secondary key, recording the storage address of each frame in the queue and the loop source identifier. The index table is implemented using a hash table structure, with the lower eight bits of the sequence number as the hash bucket index. Collisions are resolved using a linked list method. The index tables of both queues are continuously updated; when a new frame is enqueued, the index is inserted, and when an old frame is dequeued, the index is deleted, providing fast lookup capabilities for subsequent alignment and matching.

[0032] The dual-channel frame queue independently stores audio frames arriving from the main loop and the backup loop, arranged in order according to the acquisition timestamp. It is used to buffer the two audio frames and, together with the index table, enables cross-loop audio frame timing matching and lookup. The timestamp derivation unit is used to take the main loop receiving queue in the dual-path frame queue as the reference queue, extract the acquisition timestamp of the current frame to be processed in the reference queue, and perform joint calculation with the updated baseline delay difference and drift level flag to calculate the expected acquisition timestamp of the corresponding frame in the backup loop. According to the formula Obtain the expected acquisition timestamp of the corresponding frame in the backup loop. ,in The timestamp of the currently pending frame in the baseline queue. This is the offset corrected based on the baseline delay difference and drift level; this offset is the loop transmission delay compensation value used for audio frame timing matching, representing the inherent time difference between the main ring and the backup ring when the same audio frame is sent from the front-end terminal and arrives at the edge host under ideal conditions. When the drift level is a stable indicator This is the updated baseline delay difference. When the drift level is slight, moderate, or severe, the updated baseline delay difference is used as the base, and a correction factor is introduced based on the drift level to slightly adjust the compensation value, so that the matching window adapts to the slowly changing transmission delay of the link. The correction factor is used to scale and adjust the baseline delay difference, generating a timing correction offset to adapt to link fluctuations. It is pre-stored in the parameter register and retrieved according to the drift level flag. The dual-path alignment module also includes: The frame matching unit is used to perform the closest frame matching search based on the expected acquisition timestamp to obtain the backup loop candidate audio frame that is from the same source as the current frame of the main loop and its minimum time difference. Search the index table of the backup loop receiving queue for the frame whose acquisition timestamp is closest to the expected value; The minimum time difference refers to the absolute difference between the backup loop candidate audio frame and the expected acquisition timestamp, which is used to reflect the degree of timing matching between the backup loop candidate audio frame and the main loop reference frame. The sequence encapsulation unit is used to determine a match if the minimum time difference is less than the preset frame matching threshold, generate an audio frame pair and construct a dual-channel sequence, that is, to pair the acquisition timestamp of the current frame to be processed with the candidate audio frame of the backup loop; otherwise, it is determined that the match is not found, and the acquisition timestamp of the current frame to be processed is moved into the waiting buffer to wait for subsequent frames to participate in the secondary matching. Audio frame pairs are used to eliminate the inherent transmission delay deviation of dual loops, providing effective frame material for misaligned audio splicing and reconstruction.

[0033] The dual-path sequence includes the main loop sequence and the backup loop sequence. It is an ordered subset of frames that have completed timing alignment and are ready for splicing from the dual-path frame queue. It is the ordered frame stream that enters the splicing and reconstruction module.

[0034] In this embodiment, the dual-path alignment module constructs a primary and backup dual-path frame queue and a two-dimensional hash index of sequence number and timestamp through the receiving unit to achieve fast search and location of audio frames. Then, the backup loop expected timestamp is calculated by combining the timestamp deduction with the baseline delay difference and drift level flag. The closest frame matching search is performed by the frame matching unit. Finally, the sequence encapsulation unit generates the same source audio frame pair and the dual-path sequence.

[0035] Overall, the module forms a complete time-series alignment link from dual-channel reception and index construction to timestamp deduction and frame matching. It reduces the frame lookup complexity to constant level through hash index and improves matching accuracy by dynamically correcting the offset through drift level, providing high-quality source frame materials for subsequent audio splicing and reconstruction.

[0036] For example, in a subway tunnel security scenario, dozens of monitoring terminals along the line simultaneously collect audio. Traditional linear traversal search methods require several milliseconds to match each frame, while this module uses a hash bucket index with the lower eight bits of the sequence number, which can compress the single-frame matching time to the microsecond level. At the same time, combined with the expected timestamp for drift correction, the frame matching accuracy is significantly improved, ensuring the temporal continuity of subsequent voiceprint recognition input.

[0037] Specifically, the audio repair module includes: The fracture localization unit is used to locate misaligned fracture segments based on audio frame pairs and the continuity of sequence numbers, and to extract audio data of several consecutive frames before and after the fracture segment in the dual-channel sequence and record them as boundary audio samples. The measurement unit is used to perform a fast Fourier transform on each frame boundary audio sample, and after filtering by the Mel filter bank and discrete cosine transform, obtain the Mel frequency cepstral coefficient feature vector to calculate the cosine distance between the Mel frequency cepstral coefficient feature vectors of two adjacent frames in the dual-channel sequence. The filtering unit is used to select the loop with the smallest cosine distance as the reference path for the corresponding misaligned and broken segment. This path serves as the main link for audio splicing. Subsequently, missing frames are filled in from another loop to maintain the auditory continuity of the audio. Specifically, all frame sequence numbers of the main loop frame sequence and the backup loop frame sequence are extracted and arranged in ascending order of value; The sequence number difference between adjacent frames is compared sequentially along the arranged sequence number. When the difference is greater than 1, it is determined that there is a gap between the two frames where the sequence numbers are discontinuous. The audio interval corresponding to this gap is marked as a misaligned break segment. The starting sequence number, ending sequence number and gap length of each misaligned break segment are recorded.

[0038] For each misaligned break segment, three consecutive frames of audio data before the break segment and three consecutive frames of audio data after the break segment are extracted from the main loop sequence and the backup loop sequence, for a total of six boundary audio samples. Within the boundary frames of the primary and backup loop sequences, the cosine distance between the feature vectors of adjacent frames is calculated sequentially according to the frame order, using the following formula: ,in The cosine distance reflects the smoothness of the continuity of the acoustic features between two adjacent audio frames. The larger the value, the greater the difference in acoustic features between adjacent frames and the more abrupt the audio transition. and For the first , Acoustic feature vectors corresponding to frame audio; and for and The modulus length; For the first in the boundary area Frame audio; Audio acoustic features, namely Mel-frequency cepstral coefficients (MFCC) feature vectors, are multi-dimensional acoustic vectors obtained by converting a single frame of audio time-domain signal. They are 24-dimensional and represent the audio spectral envelope, timbre, and voiceprint information. They are used to quantify audio auditory features and to evaluate the continuity between audio segments. The calculated cosine distance values ​​are arranged in frame order to form a sequence of characteristic distances of the main and backup loop fracture boundaries.

[0039] Then, the average values ​​of the characteristic distance sequences of the main loop and the backup loop are calculated respectively. The two average values ​​are compared, and the loop with the smaller average value is selected as the reference path of the misaligned fracture segment.

[0040] The synthesis unit is used to obtain a temporally continuous audio frame sequence and splicing marker field based on the baseline path by extracting missing frame audio samples across paths and performing cosine window fade-in and fade-out weighted processing.

[0041] In practice, for missing frames in the baseline path, audio samples are extracted from frames with the corresponding sequence numbers from another path and inserted into the corresponding positions in the baseline path according to the sampling point order. During the insertion process, a fade-in / fade-out weighted processing is performed on the first and last 1ms intervals of the inserted frame. The weighting coefficient uses a cosine window function, and the formula is as follows: ,in These are the cosine window weighting coefficients, also known as the Hanning window coefficients; they are used to gradually weight the amplitude of sampling points within the transition interval, enabling audio fade-in and fade-out and eliminating pulse noise at splicing breakpoints. This represents the total number of sampling points in the transition interval. The sampling point number within the transition interval; Pi is a constant. The independent variable is the perspective of normalization; The value of the cosine function corresponding to the angle; The base gradually varying amplitude term is used to construct a non-linearly smooth base curve, ensuring that the weighting coefficients change gradually and continuously, and avoiding sudden changes in slope in the transition interval. This is a scaling factor, which constrains the range of weighted coefficient values ​​to between 0 and 1. After splicing, a continuous audio frame sequence is obtained, and a splicing mark field is output to mark the source path and insertion operation record of each frame for later reference.

[0042] In this embodiment, the audio restoration module uses the fracture localization unit to detect and locate misaligned fracture segments based on the continuity of sequence numbers and extracts six frame boundary audio samples. The measurement unit performs Fast Fourier Transform, Mel Filter, and Discrete Cosine Transform on the boundary samples to generate 24-dimensional MFCC feature vectors. The cosine distance between adjacent frames is calculated to evaluate continuity. Then, the filtering unit selects the loop with the smallest mean cosine distance as the reference path. Finally, the synthesis unit achieves smooth splicing by cross-path frame interpolation and cosine window fade-in / fade-out weighting, outputting a temporally continuous audio frame sequence and splicing markers.

[0043] Overall, the module forms a complete audio repair chain from fracture localization and feature calculation to path optimization and smooth splicing. It ensures auditory continuity at the splicing point through MFCC feature matching and eliminates splicing abrupt noise through a 1-millisecond cosine window transition, providing complete and continuous audio input for subsequent voiceprint recognition.

[0044] For example, in the security scenario of an underground parking lot, the time delay difference between the two rings causes the distress call to be cut into discontinuous segments. The traditional direct splicing method will produce obvious splicing noise, and the accuracy of voiceprint recognition will drop significantly. However, this module can greatly reduce the spectral abrupt change at the splicing point by using the feature selection path and cosine window to ensure that the distress call voiceprint features are complete and recognizable.

[0045] Specifically, the monitoring and early warning module includes: The indicator acquisition unit includes acquiring signal quality parameters of the dual-loop differential receiving port, including differential voltage amplitude, signal-to-noise ratio, bit error rate, and frame loss rate. The evaluation unit performs a weighted summation operation on each feature within the signal quality parameters to obtain a quality score sequence. This quality score is used to quantify the overall health of the dual-loop communication link and serves as the core basis for calculating the link quality change gradient, determining whether the link has deteriorated, and triggering loop switching prediction.

[0046] Specifically, the differential voltage amplitude is obtained by sampling through the peak detection circuit of the analog front-end at the receiving end, with a sampling frequency of 100 times per second, and the sampled values ​​are stored in the amplitude history queue. The signal-to-noise ratio (SNR) is obtained by calculating the ratio of effective signal power to noise power, with noise power measured during time slots with no effective data transmission. The bit error rate (BER) is calculated as the ratio of the number of frames that failed CRC checks to the total number of received frames, with a statistical window set to 1 second. The frame loss rate is calculated based on sequence number continuity detection, with the statistical window also set to 1 second.

[0047] Calculate the quality score every 10ms and store it in the quality score sequence; The early warning discrimination unit calculates the quality change gradient based on the quality score sequence using a first-order differential gradient. If the quality change gradient exceeds a preset degradation threshold several times consecutively, and the current quality score is lower than a preset normal threshold, it indicates that the loop signal quality is continuously and rapidly deteriorating, not due to transient interference. Based on this, it is predicted that the current loop will experience a short-term loss of communication reliability, triggering a handover warning and prompting the system to initiate the handover preparation process in advance. The warning lead time is set to 20 to 50 ms, allowing edge hosts to prepare for loop handover and audio receiving link handover in advance, minimizing the audio loss window caused by the handover interval. The specific lead time is dynamically adjusted based on the absolute value of the degradation gradient; the larger the gradient, the smaller the lead time. The quality score sequence, degradation gradient sequence, and warning records are all stored in a link quality history buffer with a capacity of 7 days for subsequent long-term link stability statistics and threshold iteration optimization. "Several times" here includes at least two instances. In this embodiment, the monitoring and early warning module synchronously collects four signal quality parameters—differential voltage amplitude, signal-to-noise ratio, bit error rate, and frame loss rate—through the indicator acquisition unit. The evaluation unit calculates a comprehensive quality score every 10 milliseconds using a weighted fusion algorithm. Then, the early warning discrimination unit uses first-order differential gradient calculation and continuous threshold to jointly determine the persistent link degradation trend and trigger a handover warning 20 to 50 milliseconds in advance.

[0048] Overall, the module breaks through the traditional passive response mode of switching after a fault. It achieves proactive early warning through multi-dimensional quality fusion and gradient trend prediction. It avoids misjudgment of a single parameter by weighting four indicators and distinguishes between instantaneous interference and real degradation by continuous gradient judgment, which significantly improves the accuracy and lead time of switching prediction.

[0049] For example, in strong electromagnetic interference scenarios in industrial parks, instantaneous surges often cause single-frame bit errors. Traditional solutions are prone to accidentally triggering handover, causing unnecessary audio jitter. However, this module can effectively filter instantaneous interference by making joint judgments on multiple consecutive gradient over-limits. At the same time, it issues warnings tens of milliseconds in advance when the link actually deteriorates, which greatly shortens the handover loss window and significantly reduces the risk of missed alarms.

[0050] Specifically, the audio reconstruction module includes: Due to hardware transmission and reception switching delays and message routing switching delays during dual-loop switching, a brief time gap is formed between the disconnection of the old loop and the completion of stable reception of the new loop. As a result, audio frames cannot be received normally, resulting in data gaps. The specific loss window exists on the audio frame receiving side of the edge host, in the audio frame timing interval corresponding to the switching gap between the main loop and the backup loop. The positioning unit is used to receive switching warnings in order to locate the audio frame loss window and its preceding and following boundary frames; The front and back boundary frames include the front boundary frame and the back boundary frame. The front boundary frame is the last audio frame that is completely received before the handover moment, and the back boundary frame is the first audio frame that is completely received after the handover moment.

[0051] The coefficient solving unit is used to perform linear predictive coding analysis on the temporal samples of the preceding and following boundary frames respectively. It uses the autocorrelation method to obtain the autocorrelation sequence, and uses the Levinson-Dubin recursive algorithm to solve the linear prediction coefficients, i.e., LPC coefficients, to obtain two sets of linear prediction coefficients, one forward and one backward. This set includes 12 linear prediction coefficients. The autocorrelation function is calculated using the formula... Calculation, where For delay The autocorrelation function value; This is the autocorrelation delay, representing the number of interval points between two sampling points. ; The sampling point number; This represents the total number of sampling points in the current audio frame, i.e., the frame length. It is a discrete time-domain sampling sequence of a single frame of audio, that is, a sequence of discrete sampling values ​​at each moment within a frame of audio. It is a time-domain digital audio sample obtained after analog-to-digital conversion and frame-by-frame processing. For distance from the first Sampling points shifted backward The sequence of audio sample values ​​corresponding to each position; upper limit of summation In order to ensure It will not exceed the maximum index of a frame's sampling points to prevent access to invalid sampling positions; Next, two sets of autocorrelation sequences are calculated for the front boundary frame and the back boundary frame, respectively, and input into the Levinson-Dubin recursive equation system. Through recursive iteration, the equation system is solved to obtain the forward linear prediction coefficient set and the backward linear prediction coefficient set, which are used for bidirectional prediction interpolation to reconstruct the audio of the lost window. The waveform reconstruction unit is used to obtain the interpolated sample sequence within the audio frame loss window and repackage it into a complete audio frame based on two sets of linear prediction coefficients, forward and backward, through bidirectional prediction interpolation operation with time proportional linear weighting, so as to realize audio alarm monitoring and waveform compensation.

[0052] In practice, the interpolated samples within the missing window are weighted and synthesized with the forward and backward predicted values ​​according to a time ratio, with the weighting coefficients linearly transitioning from 1 to 0 over time. Each interpolated sample is processed according to the formula... Calculate, where, For the lost window The audio prediction sampling points to be generated reflect the audio amplitude at the corresponding moment of the lost interval, and are used to reconstruct the missing audio samples within the lost window; this value represents the instantaneous amplitude of the audio, and the larger the absolute value, the greater the loudness of the sound at that moment; The weighting coefficient for the forward forecast term smoothly decreases from 1 to 0 over time; The sequence number of the predicted sampling point within the lost window; For LPC prediction order index; The first frame obtained by solving for the front boundary frame Forward LPC prediction coefficients; and For the lost window and One audio prediction sampling point to be generated; The weighting coefficients for the backward prediction term smoothly increase from 0 to 1 over time, satisfying the following conditions: ; The solution obtained for the back boundary frame backward LPC prediction coefficients; To weight the forward prediction components, audio samples are inferred from front to back based on the front boundary frame model; To weight the backward prediction components, audio samples are inferred backward from the back boundary frame model; The formula uses a bidirectional LPC prediction fusion method to generate missing samples. Near the beginning of the missing window, the forward prediction component dominates; near the end of the missing window, the backward prediction component dominates, thus avoiding the distortion of unidirectional prediction increasing over time.

[0053] The front boundary frame model is a 12th-order all-pole LPC model obtained by solving the audio samples of the front boundary frame of the switching window. It is characterized by a set of forward LPC coefficients and can generate subsequent audio samples in the forward time direction based on historical sampling points, thus achieving prediction from front to back. The back boundary frame model is a 12th-order all-pole LPC model obtained by back analysis of the audio samples of the back boundary frame of the switching window. It is characterized by a set of backward LPC coefficients and can generate the previous audio samples in the backward time direction based on future sampling points, thus achieving prediction from back to front.

[0054] In this embodiment, the audio reconstruction module breaks through the traditional approach of unidirectional prediction or simple repetition for packet loss compensation. The positioning unit accurately locks the front and back boundary frames of the lost switching window. The coefficient solving unit uses the autocorrelation method and the Levinson-Dubin recursive algorithm to solve the 12th-order forward and backward LPC coefficient sets respectively. Finally, the waveform reconstruction unit weights and fuses the bidirectional prediction results according to the time ratio to generate the complete audio frame.

[0055] Overall, the module innovatively applies bidirectional LPC prediction fusion to dual-ring switching audio completion scenarios. Forward prediction extrapolates from the beginning to the end, while backward prediction extrapolates from the end to the beginning. A smooth transition is achieved through time-weighted scaling, avoiding the problem of unidirectional prediction causing increasing distortion over time. It also fully utilizes the spectral characteristics of the real audio at both ends, ensuring a high degree of continuity in acoustic features between the completed segment and the original audio. For example, in a campus perimeter security scenario, when an emergency switch is triggered by the main ring fiber optic cable being severed during construction, traditional solutions simply use repeated filling of the preceding boundary frame or simple linear interpolation, leading to abrupt changes in voiceprint features at the millisecond gap, significantly reducing the accuracy of distress call recognition. This module, through bidirectional LPC weighted fusion, can significantly reduce the spectral distortion of the supplemented audio, maintain a high level of voiceprint recognition accuracy, and ensure that no dangerous situations during switching are missed.

[0056] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A smart alarm monitoring and listening system, characterized in that: The system includes a time delay calibration module, a dual-channel alignment module, an audio restoration module, a monitoring and early warning module, and an audio reconstruction module; The latency calibration module is used to obtain the baseline latency difference and update the baseline latency difference by adding timestamps and periodic probes through hop-by-hop forwarding of each monitoring terminal. The dual-path alignment module is used to construct a dual-path frame queue and, in conjunction with the updated baseline delay difference, calculate the expected acquisition timestamp. After frame matching, the dual-path frame queue is updated to obtain the dual-path sequence. The audio restoration module is used to locate misaligned and broken segments based on dual-channel sequences, and analyze the feature similarity of the audio at the boundary of the misaligned and broken segments to obtain a temporally continuous audio frame sequence. The monitoring and early warning module is used to collect signal quality parameters and analyze the changing trend of loop signal quality to identify link degradation trends. When a link degradation trend is present, a handover warning is triggered. The audio reconstruction module is used to receive switching warnings, identify audio frame loss windows, and analyze the temporal acoustic evolution of the audio at both boundaries of the audio frame loss window to obtain the completed audio frame, so as to realize audio alarm monitoring and waveform compensation.

2. The intelligent alarm monitoring and listening system according to claim 1, characterized in that: The delay calibration module includes: The message acquisition unit is used to synchronously send calibration messages to the dual-loop differential bus through the edge control host in the power-on initialization state. The calibration data stream is obtained by forwarding the messages hop by hop through each listening terminal and appending a local timestamp. The baseline solving unit is used to extract an ordered set of timestamps from the calibration data stream to calculate the total transmission delay of the primary and backup loops, and to obtain the baseline delay difference through difference calculation. The hop-by-hop parsing unit is used to record the listening terminal as a node, and extract the hop-by-hop delay sequence of each node based on the receiving timestamp of each node in the calibration message and by calculating the difference between the timestamps of adjacent nodes.

3. The intelligent alarm monitoring and listening system according to claim 2, characterized in that: The delay calibration module also includes: The dynamic tracking unit is used to obtain the real-time dynamic delay difference based on the baseline delay difference and the hop-by-hop delay sequence through bidirectional acquisition of periodic probe frames and smoothing with first-order low-pass filtering. The drift correction unit is used to perform a difference calculation between the real-time dynamic delay difference and the baseline delay difference to obtain the drift amount, and compare it with the preset drift range to obtain the drift level flag. Through the baseline dynamic approximation write-back mechanism, the updated baseline delay difference is obtained.

4. The intelligent alarm monitoring and listening system according to claim 3, characterized in that: Dual-path alignment module, including: The receiving unit is used to collect audio data through the monitoring terminal, transmit it through dual-loop differential independent transmission, and obtain a dual-channel frame queue and a dual-channel audio index table. The dual-channel frame queue includes a main loop receiving queue and a backup loop receiving queue. The timestamp derivation unit is used to take the main loop receiving queue in the dual-path frame queue as the reference queue, extract the acquisition timestamp of the current frame to be processed in the reference queue, and perform joint calculation with the updated baseline delay difference and drift level flag to calculate the expected acquisition timestamp of the corresponding frame in the backup loop.

5. The intelligent alarm monitoring and listening system according to claim 4, characterized in that: The dual-path alignment module also includes: The frame matching unit is used to perform the closest frame match search based on the expected acquisition timestamp to obtain the candidate audio frames of the backup loop and their minimum time difference. The sequence encapsulation unit is used to determine a match if the minimum time difference is less than a preset frame matching threshold, generate audio frame pairs, and construct a dual-channel sequence.

6. The intelligent alarm monitoring and listening system according to claim 5, characterized in that: The audio repair module includes: The fracture localization unit is used to locate misaligned fracture segments based on audio frame pairs and the continuity of sequence numbers, and to extract audio data of several consecutive frames before and after the fracture segment in the dual-channel sequence and record them as boundary audio samples. The measurement unit performs a fast Fourier transform on each frame boundary audio sample, filters it using a Mel filter bank and performs a discrete cosine transform to obtain the Mel frequency cepstral coefficient eigenvector, and then calculates the cosine distance between the Mel frequency cepstral coefficient eigenvectors of two adjacent frames in the dual-channel sequence.

7. The intelligent alarm monitoring and listening system according to claim 6, characterized in that: The audio repair module also includes: The filtering unit is used to select the loop with the smallest cosine distance as the reference path for the corresponding misaligned fracture segment. The synthesis unit is used to obtain an audio frame sequence based on the baseline path by extracting missing frame audio samples across paths and performing weighted processing with cosine window fade-in and fade-out.

8. The intelligent alarm monitoring and listening system according to claim 7, characterized in that: The monitoring and early warning module includes: The indicator acquisition unit includes acquiring signal quality parameters of the dual-loop differential receiving port, including differential voltage amplitude, signal-to-noise ratio, bit error rate, and frame loss rate. The evaluation unit is used to perform a weighted summation operation on each feature within the signal quality parameters to obtain a quality score sequence. The early warning judgment unit is used to calculate the quality change gradient based on the quality score sequence through the first-order difference gradient. If the quality change gradient exceeds the preset degradation threshold for several consecutive times, and the current quality score is lower than the preset normal threshold, a switching warning is triggered.

9. The intelligent alarm monitoring and listening system according to claim 8, characterized in that: The audio reconstruction module includes: The positioning unit is used to receive switching warnings in order to locate the audio frame loss window and its preceding and following boundary frames; The coefficient solving unit is used to perform linear predictive coding analysis on the temporal samples of the preceding and following boundary frames respectively. It uses the autocorrelation method to obtain the autocorrelation sequence, and uses the Levinson-Dubin recursive algorithm to solve the linear prediction coefficients from the autocorrelation sequence to obtain two sets of linear prediction coefficients, namely the forward and backward sets. The waveform reconstruction unit is used to obtain the interpolated sample sequence within the audio frame loss window and repackage it into a complete audio frame based on two sets of linear prediction coefficients, forward and backward, through bidirectional prediction interpolation operation with time proportional linear weighting, so as to realize audio alarm monitoring and waveform compensation.

Images