Audio full-link adaptive fusion implementation method of emergency broadcasting system

Abstract

The application discloses an audio full-link adaptive fusion implementation method of an emergency broadcasting system, and belongs to the technical field of petrochemical safety production. The method runs through three core links of explosion-proof talk station pickup, industrial IP network transmission and explosion-proof broadcasting terminal playing, and fuses four core algorithms of an improved IMCRA, an improved NLMS, an improved FEC and an improved dynamic jitter buffer. The pickup link solves the noise suppression problem through a pulse noise detection mechanism and dynamic smoothing factor adjustment, and introduces a noise type weight factor and a high-frequency compensation factor to reserve speech details. The transmission link adopts a network fluctuation coefficient to adaptively adjust the FEC redundancy, a dynamic grouping strategy and a hierarchical PLC packet loss compensation to improve the anti-packet loss capability. The application is implemented by using a pure software algorithm, can be directly transplanted to an existing petrochemical emergency broadcasting system, has a pulse noise suppression effect of more than 28 dB, an echo suppression amount of more than 25 dB, and an FEC decoding recovery rate of greater than or equal to 98%, and meets the core demand of petrochemical safety production.

Description

Technical Field

[0001] This invention belongs to the field of petrochemical safety production technology, and in particular, it is a method for implementing audio end-to-end adaptive fusion in an emergency broadcasting system. Background Technology

[0002] Petrochemical plant areas are high-risk explosion-proof environments, and emergency broadcasting systems, as core equipment for ensuring safe production, must deliver emergency instructions quickly and clearly in complex environments. However, existing petrochemical emergency broadcasting systems have many deficiencies in audio processing and transmission links, making them unsuitable for the special working conditions of petrochemical plants.

[0003] (1) Sound pickup link: There are various noises in the petrochemical plant area, such as steady-state mechanical noise (pump body and compressor operating noise), non-steady-state wind noise, and instantaneous impulse noise (equipment start-up and shutdown noise). Conventional IMCRA noise estimation algorithm judges noise frames and speech frames only by a single posterior signal-to-noise ratio, which is easily affected by impulse noise, resulting in incomplete noise suppression and speech distortion, especially loss of high-frequency speech details, which affects the recognition of emergency instructions;

[0004] (2) Transmission link: The petrochemical industry Ethernet has problems such as large fluctuations in packet loss rate and frequent sudden packet loss. The conventional FEC algorithm uses fixed redundancy or single parameter adjustment, which cannot balance the anti-packet loss capability and bandwidth utilization. The conventional dynamic jitter buffer only adjusts the depth according to the jitter value, which is prone to excessive delay or audio stuttering. The PLC packet loss compensation method is single, and voice interruption is prone to occur when multiple frames are lost.

[0005] (3) Playback link: The backflow of the explosion-proof terminal speaker is prone to cause howling. The conventional NLMS adaptive filtering algorithm has a contradiction between convergence speed and steady-state error, and cannot adapt to the time-varying characteristics of the echo path, which is prone to incomplete echo suppression or speech distortion. At the same time, the noise intensity varies greatly in different scenarios, and fixed gain playback cannot ensure that the speech is clearly audible.

[0006] (4) Poor end-to-end coordination: The three links of sound pickup, transmission and playback work independently without a closed-loop feedback mechanism. The algorithm parameters cannot be adaptively adjusted in real time according to the on-site environment and network status, resulting in unstable audio quality across the entire link, which makes it difficult to meet the core requirements of high reliability, high definition and low latency for petrochemical emergency broadcasting.

[0007] A search revealed a novel emergency broadcast self-testing loudspeaker based on nuclear power, marine engineering, and petrochemical industries, with publication number CN219555110U. This patent primarily focuses on structural improvements for loudspeakers in special scenarios such as petrochemical and nuclear power plants. It utilizes a die-cast nylon and fiberglass blend to create an explosion-proof shell, employs a diaphragm and sintered sheets to buffer sound pressure and prevent sparks from excessive instantaneous sound pressure, and uses a self-test module to identify the loudspeaker's electroacoustic signals to determine its health status. However, this technical solution only addresses the terminal playback device level of the emergency broadcast system, neglecting audio processing technology for the pickup and transmission links. It fails to address issues such as voice enhancement in the complex noise environment of petrochemical plants, reliable audio transmission under industrial Ethernet packet loss conditions, and end-to-end adaptive collaboration. It only achieves explosion-proof design and self-testing functions at the hardware level, lacking end-to-end audio processing capabilities.

[0008] A search revealed a desktop emergency radio with publication number CN118432647A (the same technical solution corresponds to utility model CN221553256U). This patent provides an emergency radio that integrates multiple network interfaces (fiber optic transmission network, 4G / 5G network, Ethernet interface, Mesh intelligent networking, etc.) to achieve network intercom, network broadcasting, and amplification. It connects to a network broadcasting module, Ethernet interface, VoIP module, wireless intelligent networking module, and U / V dual-frequency intercom module via a PLC processor. However, this technical solution primarily focuses on the network architecture design and multi-module integration of the emergency communication system, neglecting core audio processing aspects such as noise suppression algorithms at the pickup end, packet loss compensation mechanisms in the transmission link, and echo cancellation and adaptive gain control at the playback end. It only solves the integration problem of multiple network transmission methods and fails to achieve the end-to-end adaptive audio fusion processing required for petrochemical emergency broadcasting systems.

[0009] The aforementioned existing technologies demonstrate that, in practical applications, the audio processing technologies for each stage of the current petrochemical emergency broadcasting system—including sound pickup, transmission, and playback—are relatively independent and have significant shortcomings, failing to meet the core requirements of high reliability, high definition, and low latency for emergency broadcasting under complex operating conditions in petrochemical plants. Therefore, this invention proposes an adaptive fusion implementation method for the entire audio link of an emergency broadcasting system. This method aims to improve the overall performance of the petrochemical emergency broadcasting system by improving core algorithms such as noise estimation, packet loss compensation, and echo cancellation, and by introducing a closed-loop feedback mechanism across the entire link, thereby achieving adaptive collaborative optimization of each stage. Summary of the Invention

[0010] The purpose of this invention is to provide an adaptive fusion method for the entire audio link of an emergency broadcasting system to solve the problems mentioned in the background art.

[0011] To achieve the above objectives, this invention provides the following technical solution: an audio end-to-end adaptive fusion implementation method for an emergency broadcasting system, encompassing three core stages: explosion-proof microphone pickup, industrial IP network transmission, and explosion-proof broadcasting terminal playback. It integrates four core algorithms: improved IMCRA, improved NLMS, improved FEC, and improved dynamic jitter buffering, to achieve end-to-end closed-loop adaptive optimization. The specific steps are as follows:

[0012] S1. Explosion-proof voice pickup link (signal acquisition and adaptive noise reduction): Acquires emergency voice signals, suppresses various noises in the petrochemical plant area through improved algorithms, and outputs a clean voice signal, laying the foundation for subsequent transmission; specifically including:

[0013] (1) Audio signal acquisition: The explosion-proof microphone is built into the explosion-proof telephone station to acquire mixed audio signals on site. The signal includes emergency voice messages (such as emergency instructions and evacuation notices) and typical noises in petrochemical plants (steady-state mechanical noise of pumps / compressors, outdoor non-steady-state wind noise, and instantaneous pulse noise during equipment start-up and shutdown). The pickup range covers 0.5-5m and is suitable for explosion-proof conditions on site.

[0014] (2) Preprocessing of the audio signal: preprocessing the acquired mixed audio signal Framing and windowing are performed, following unified basic parameters across the entire link: frame length N=256 points, frame shift L=128 points (50% overlap, balancing latency and audio quality), Hanning window is used to suppress spectral leakage, and the windowed signal is...

[0015] ,

[0016] The Hanning window formula is: After framing, the signal of the m-th frame is .in, For the Hanning window function, For the first frame after segmentation Frame time domain signal, For the first time after adding a window Frame time domain signal, Index of signal sampling points within a single frame ( ), For frame index ( ).

[0017] (3) STFT time-frequency transformation and power spectrum calculation: Perform FFT short-time Fourier transform on each frame of the windowed signal to convert the time-domain signal into a complex frequency-domain spectrum. Calculate the power spectrum at each frequency point As the core foundational data for noise identification and denoising.

[0018] (4) Improved IMCRA Adaptive Noise Power Spectrum Estimation: To address the vulnerability of the conventional IMCRA algorithm to transient impulse noise, an improved IMCRA algorithm is adopted to track time-varying noise in real time and dynamically update the noise power spectrum. Specific implementation details are as follows:

[0019] Calculate the a posteriori signal-to-noise ratio at each frequency point ,in For the previous frame (the first frame) (Frame) Estimated noise power spectrum at frequency points For the first Frame number Posterior signal-to-noise ratio at a frequency point (unit: dB);

[0020] Impulse noise detection: Calculate the energy of the current frame. Multi-frame average energy ,like If the frame is identified as a transient impulse noise frame, the noise spectrum is updated using a "neighboring frame interpolation" strategy. To avoid pulse noise contaminating the noise substrate; among which, For the first The instantaneous energy of the signal after frame windowing. For the first Average energy of multiple frames (smoothed energy). For the first Frame number Estimated noise power spectrum at frequency points For the first Frame number Estimated noise power spectrum at frequency points;

[0021] Dynamic smoothing factor adjustment: The smoothing factor is dynamically adjusted according to the noise type. Steady-state mechanical noise scenarios (such as pump areas) Pick This improves noise spectrum stability; suitable for non-steady-state wind noise scenarios (such as outdoor factory areas). Pick Improve noise spectrum tracking speed;

[0022] Soft VAD decision: If it is a steady-state noise frame ( and ),according to Recursively smooth the noise spectrum update;

[0023] If it contains audio frames Freezing noise spectrum .

[0024] (5) Improved prior signal-to-noise ratio estimation and noise reduction gain calculation: To address the problem of large estimation bias in conventional decision-directed recursive methods under low signal-to-noise ratio scenarios, a collaborative improvement is made to ensure that noise reduction does not lose speech details.

[0025] Introduce a noise type weighting factor ω, dynamically adjusted based on the noise type identified in step 4: steady-state noise ω=0.96, non-steady-state noise ω=0.98, impulse noise ω=0.94; optimize the prior signal-to-noise ratio estimation formula: ,in This is the prior signal-to-noise ratio smoothing factor. For the previous frame (the first frame) The frequency domain amplitude after denoising (frame). For the first Frame number Prior signal-to-noise ratio at a frequency point This indicates taking the maximum value (ensuring the prior signal-to-noise ratio is non-negative).

[0026] High-frequency compensation optimization: Introducing a high-frequency compensation factor ,when (High-frequency band, concentrated frequency band of speech consonants) The final noise reduction gain formula is: To avoid losing high-frequency speech details. Among them, For the first High-frequency compensation factor at frequency point For the first Frame number Noise reduction gain at frequency points.

[0027] (6) Frequency domain reconstruction and time domain synthesis: The frequency domain spectrum is corrected by noise reduction gain to obtain the clean speech spectrum. Perform an inverse IFFT transform on the corrected spectrum to restore the time-domain signal. The system employs an Overlapping Addition (OLA) algorithm to concatenate continuous time-domain audio, outputting a denoised, clean emergency voice signal x̂(n), which is then sent to an industrial IP audio transmission module. For the first Frequency domain spectrum after frame denoising For the first The time-domain signal after frame denoising, This represents the inverse short-time Fourier transform. This is the final output of a clean emergency voice time-domain signal.

[0028] S2: Industrial IP network transmission link (signal encoding and adaptive transmission) stably transmits clean emergency voice messages via industrial Ethernet. Combined with improved algorithms, it addresses network packet loss, latency jitter, and bandwidth fluctuations, ensuring smooth, uninterrupted transmission and meeting emergency latency requirements (≤300ms). Specifically, it includes:

[0029] (1) Audio encoding and RTP encapsulation: The pure voice x̂(n) output by the explosion-proof telephone station is encoded using OPUS (balancing sound quality and bandwidth), and the encoding parameters are dynamically adjusted according to the real-time network status to generate an audio payload frame after encoding; it is encapsulated according to the RTP standard, with the sequence number Seq, timestamp, and SSRC identifier added, and the encapsulation format is as follows. Redundant data, adapted to industrial IP transmission protocols. Among them, This is the encapsulated RTP data packet. This is the RTP protocol header (containing information such as sequence number and timestamp). The audio payload encoded with OPUS. Redundant data for forward error correction.

[0030] (2) Real-time monitoring and classification of network status: The sending end receives network status parameters reported by the terminal through the RTCP protocol (feedback period of 500ms) and calculates the core network indicators in real time: packet loss rate. latency jitter ,in Packet loss rate (unit: %) The number of lost data packets. This represents the total number of data packets sent. Jitter (unit: ms) The number of data packets to be counted. Let be the end-to-end delay of the i-th data packet. The average end-to-end latency for all statistical data packets; This represents the real-time available bandwidth (unit: kbps), which is the maximum bandwidth that the current network can stably transmit.

[0031] The network is divided into three levels based on the indicators:

[0032] High-quality network: , ;

[0033] Medium network: , ;

[0034] Weak network congestion: , .

[0035] (3) Adaptive bitrate and sampling rate adjustment: Based on the network classification results, the coding bitrate and sampling rate are dynamically adjusted, and a first-order smoothing filter is used to avoid bitrate jumps. The formula is as follows: , λ∈[0.9,0.95]), where, For the first Adaptive adjustment bandwidth of the frame (unit: kbps). For the first Adaptive adjustment of frame bandwidth For the first Real-time available bandwidth of the frame This is a bandwidth smoothing factor used to avoid bitrate jumps caused by excessively rapid bandwidth adjustments.

[0036] The relationship between bit rate and sampling rate:

[0037] High-quality network: 128kbps bitrate, 16kHz sampling, enhancing sound quality;

[0038] Medium network: 64kbps bitrate, 16kHz sampling, balancing audio quality and bandwidth;

[0039] Weak network congestion: 32kbps bitrate, 8kHz sampling, prioritizing transmission stability.

[0040] (4) Improved Adaptive FEC Forward Error Correction Coding: In view of the characteristics of large packet loss fluctuations and frequent burst packet loss in the petrochemical industry Ethernet, the conventional FEC algorithm is improved to enhance the packet loss resistance capability. The specific implementation is as follows:

[0041] Introducing a network volatility coefficient σ to dynamically adjust the redundancy weighting coefficient

[0042] ,in, The standard deviation of the packet loss rate over the past 10 frames (in %) reflects the degree of fluctuation in network packet loss. These are redundancy weighting coefficients. The larger, The larger the value, the higher the redundancy, and the better it is at handling sudden packet loss.

[0043] Dynamic grouping strategy: Adjust the number of original audio packets k in each group according to the network level. For high-quality networks, k=10 (to improve bandwidth utilization), for medium-quality networks, k=8 (to balance redundancy and bandwidth), and for weak networks with congestion, k=6 (to improve the probability of packet loss recovery).

[0044] Redundant packet priority allocation: Redundant packets are divided into core redundant packets (60%) and supplementary redundant packets (40%). Core redundant packets are transmitted synchronously with the original packets, while supplementary redundant packets are dynamically inserted according to bandwidth to avoid preempting the transmission resources of the original packets.

[0045] Redundancy calculation: Each group Generating one original audio package A redundant packet; as long as the receiver receives it... With one valid packet (original packet + core redundant packet prioritized), the entire audio set can be recovered via RS decoding. The number of redundant packets corresponding to each group of original audio packets (rounded up). This represents the floor function. The number of original audio packets in each group.

[0046] (5) Improved dynamic jitter buffer and PLC packet loss compensation: Optimize the conventional jitter buffer and packet loss compensation algorithm to achieve a balance between jitter suppression and delay control. Specific implementation:

[0047] Dynamic cache depth adjustment: The formula is as follows ,in Frame (base jitter buffer). (Jitter weighting factor) (Latency-Jitter Balance Factor) This represents the current end-to-end latency (in milliseconds). For the first The dynamic jitter buffer depth (unit: frames) is used to avoid excessive latency due to an oversized buffer or stuttering due to an undersized buffer.

[0048] Cache status warning: When the cache utilization rate is >80%, reduce the cache depth in advance and speed up the audio reading speed; when the utilization rate is <20%, increase the cache depth and slow down the reading speed to avoid overflow or underflow.

[0049] Hierarchical PLC packet loss compensation: Time-domain interpolation is used for single-frame packet loss. For 2-3 consecutive frames of packet loss, "interpolation + speech feature prediction" (combining the speech amplitude trend of the previous 10 frames to predict the lost frames) is used. For ≥4 consecutive frames of packet loss, a silent transition + subsequent frame fade-in is used to avoid speech interruption.

[0050] (6) Receiver timing synchronization and decoding: The receiver reorders out-of-order packets according to the RTP sequence number and timestamp, reads audio frames at a constant speed through a jitter buffer, and performs smooth fade-in and fade-out splicing to avoid splicing distortion; the spliced ​​audio frames are then OPUS decoded, and the decoded audio signal is output. The signal is then sent to the explosion-proof broadcast terminal playback module. This is the time-domain audio signal of the playback source after decoding at the receiving end.

[0051] S3: Explosion-proof terminal playback link (signal noise reduction, echo suppression, and gain control) cancels speaker backflow echo, suppresses ambient noise, and adaptively adjusts playback gain according to noise level to ensure emergency voices are clearly audible in different scenarios without howling or popping sounds. Specifically, it includes:

[0052] (1) Dual signal synchronous acquisition: The explosion-proof broadcast terminal synchronously acquires two signals: one is the audio signal y(n) after IP transmission decoding (playback source signal), and the other is the mixed signal z(n) acquired by the built-in microphone of the terminal (ambient noise + speaker feedback echo); z(n) is preprocessed by framing and windowing (same as step 2 of the sound pickup link) to prepare for echo suppression and noise detection.

[0053] (2) Improved NLMS adaptive filtering echo suppression: To address the contradiction between the convergence speed and steady-state error of the conventional NLMS algorithm, and the time-varying characteristics of the echo path in petrochemical terminals, a triple improvement is made to counteract the backfeed echo from the loudspeaker:

[0054] Variable step size adaptive strategy: Dynamically adjust the step size μ(n) based on the echo error e(n). When the echo is strong... ), μ(n) is set to 0.04-0.05 to accelerate the convergence speed;

[0055] When the echo is weak ), μ(n) is set to 0.01-0.02 to reduce steady-state error;

[0056] Echo path abrupt change detection: Calculate the Euclidean distance of the filter coefficients of adjacent frames. When d(n) > 0.05, the filter coefficients are reset (initialized using the average coefficients of the previous 5 frames) to prevent echo suppression from failing; among which, The Euclidean distance between the filter coefficients of adjacent frames. For the first The adaptive filter coefficient vector at time t. For the first The adaptive filter coefficient vector at time t. Represents the Euclidean distance between vectors.

[0057] Voice protection mechanism: When the playback source signal y(n) is detected to be a voice frame (judged by short-time energy), the filter coefficient update rate is multiplied by a weight of 0.5 to avoid the voice signal being over-filtered and causing distortion.

[0058] Echo cancellation calculation: The filtering formula is as follows The formula for iteratively updating the filter coefficients is: (∥y(n)∥²+δ)( (Regularization factor); signal after echo cancellation An improved IMCRA algorithm for the pickup link is used to further suppress ambient noise and output a clean noise reference signal. .in, The filter coefficient vector transpose, This is a regularization factor (to avoid a denominator of 0). The clean noise reference signal after echo cancellation and noise suppression.

[0059] (3) Environmental noise intensity detection and classification:

[0060] Calculate the average noise power of a single frame ;in, For the first Average noise power of the frame The number of frequency points per frame (consistent with the frame length). ), For the first Frame number Estimated noise power spectrum at frequency points;

[0061] Multi-frame smoothing: ;in This is a noise intensity smoothing factor to avoid misjudgment of instantaneous noise. For the first The average noise power after frame smoothing. For the first Average noise power after frame smoothing;

[0062] Three-level classification:

[0063] Low noise level ( (such as office areas)

[0064] Medium noise level ( (such as a regular workshop)

[0065] High noise level ( (e.g., pump area, compressor workshop).

[0066] (4) Adaptive AGC gain control: Establish an adaptive mapping relationship between ambient noise level and playback gain, and use a first-order recursive smoothing algorithm to avoid gain jumps. Specific strategies:

[0067] The reference gain is set to Gain adjustment step size , ;in, Baseline playback gain (default value) ), The gain adjustment step size for low-noise and medium-noise scenes. The gain adjustment step size for medium and high noise scenarios;

[0068] Gain allocation: at low noise levels, (To avoid disturbing others with excessive volume); at medium noise level, (Maintain clear sound quality); at high noise levels, (Compensation for noise masking to ensure clear audibility); among which, The raw gain (unsmoothed) is obtained based on the noise level.

[0069] Smooth adjustment: ,in This is the gain smoothing factor. For the first Frame smoothing followed by adaptive playback gain. For the first Frame smoothing followed by adaptive playback gain. For the first The original gain of the frame.

[0070] (5) Playback output and closed-loop feedback:

[0071] Amplitude limiting processing: Playback source signal With adaptive gain Multiply to apply amplitude limiting and anti-trimming, the formula is as follows: ,in The maximum amplitude value of 16-bit PCM audio ( ), This is the final playback time-domain audio signal after clipping. A limiting function (limiting the signal amplitude to a certain value) Within the range, avoid popping sounds);

[0072] Audio playback: The signal is sent to the explosion-proof speaker at the terminal to ensure... The sound was clearly audible inside, without any popping or whistling.

[0073] Closed-loop feedback: The terminal feeds back the collected ambient noise level to the sending end in real time via the RTCP protocol, and links it with network status parameters to synchronously adjust the sound pickup and noise reduction intensity, transmission bit rate, FEC redundancy and playback gain, so as to achieve end-to-end adaptive optimization.

[0074] S4: End-to-end closed-loop collaborative logic, specifically including:

[0075] (1) Data linkage: The noise reduction effect of the sound pickup link and the environmental noise level of the terminal link are fed back to the transmission link in real time, serving as the core basis for adjusting the transmission parameters;

[0076] (2) Parameter coordination: In high noise and weak network scenarios, the intensity of sound pickup and noise reduction, FEC redundancy and playback gain are increased simultaneously, and the coding bit rate is reduced; in high quality network and low noise scenarios, the coding bit rate is increased, FEC redundancy and playback gain are reduced, and the noise reduction intensity is weakened, taking into account both sound quality and transmission efficiency.

[0077] (3) Real-time update: After each frame of data is processed, the network status and noise status are fed back through RTCP to realize the real-time adaptive update of all improved algorithm parameters, ensuring that the emergency broadcast is stable and clear throughout the entire link and adaptable to the complex working conditions of the petrochemical plant.

[0078] S5: End-to-end process deployment guarantee, ensuring stable deployment of the entire process, adapting to special working conditions such as explosion-proof, high-temperature, and high-humidity environments in petrochemical plants, ensuring efficient algorithm operation, and avoiding process interruption or performance degradation. Specifically, it includes:

[0079] (1) Hardware compatibility guarantee:

[0080] Explosion-proof compatibility: The explosion-proof telephone station and explosion-proof broadcast terminal meet the petrochemical explosion-proof level requirements (ExdIIBT6Gb). The microphone and speaker adopt an explosion-proof structure to avoid generating electrical sparks during equipment operation, making them suitable for explosion-proof scenarios such as refining workshops and pump areas.

[0081] Hardware performance support: The explosion-proof microphone station has a built-in high-performance audio processing chip that supports real-time processing at a sampling rate of 16kHz, ensuring the efficient operation of the improved IMCRA algorithm; the terminal has a built-in low-noise microphone to improve echo suppression and noise acquisition accuracy; the industrial IP transmission module supports hardware acceleration of RTP / RTCP protocol to reduce transmission latency.

[0082] Anti-interference design: All hardware devices are equipped with shielded shells to avoid strong electromagnetic interference in the petrochemical plant area (such as electromagnetic pulses generated by equipment start-up and shutdown) from affecting audio signal acquisition, transmission and playback.

[0083] (2) Algorithm operation guarantee:

[0084] Parameter initialization and calibration: Upon system startup, the system automatically collects ambient noise and network status data to initialize the core parameters of the improved algorithm (such as the noise smoothing factor). FEC redundancy weighting coefficient NLMS step size (etc.) to ensure that the algorithm can quickly adapt to the current scenario after startup;

[0085] Algorithm fault tolerance mechanism: When a link algorithm malfunctions (such as IMCRA algorithm noise spectrum estimation failure or FEC decoding failure), the system automatically switches to the backup algorithm logic (such as switching to basic Wiener filtering when IMCRA malfunctions, and strengthening PLC packet loss compensation when FEC decoding fails) to avoid interruption of the entire link process;

[0086] Real-time monitoring and calibration: The algorithm output effect is calibrated every 100 frames of data. The speech recognition, transmission stuttering rate and playback echo suppression are compared after noise reduction. If the preset thresholds are not met (speech recognition ≥ 95%, stuttering rate ≤ 1%, echo suppression ≥ 25dB), the algorithm parameters are automatically fine-tuned to ensure the quality of process operation.

[0087] (3) Environmental adaptability guarantee:

[0088] Temperature and humidity adaptation: The algorithm parameters support dynamic adjustment according to the ambient temperature and humidity. In high temperature scenarios (≥40℃), the audio frame length is reduced to 200 points to improve processing speed; in high humidity scenarios (humidity ≥85%), the noise smoothing factor λ_d is increased to avoid noise acquisition distortion caused by humidity.

[0089] Scene switching adaptation: When the terminal detects a scene switch (such as personnel entering the pump area from a regular workshop, or the noise level changing from low noise to high noise), it quickly feeds back to the sending end via RTCP. Within 100ms, it completes the coordinated adjustment of the sound pickup and noise reduction intensity, playback gain, and FEC redundancy to ensure stable audio quality during scene switching.

[0090] S6: End-to-end anomaly handling process. For sudden anomalies that may occur in petrochemical plants (such as network interruption, equipment failure, and sudden noise changes), a hierarchical anomaly handling mechanism is established, and an improved algorithm is linked to ensure that emergency broadcasts are not interrupted and instructions can be transmitted.

[0091] (1) Level 1 anomaly (minor anomaly): minor network fluctuations (packet loss rate 3%-5%, jitter 50-70ms), instantaneous noise changes (such as brief device start-up and shutdown); processing flow: the improved FEC algorithm automatically increases the redundancy weighting coefficient β to 1.3, and increases the dynamic jitter buffer depth by 1 frame; the improved IMCRA algorithm temporarily adjusts the smoothing factor. Reduce by 0.02 to improve noise tracking speed; temporarily increase AGC gain by 3dB to ensure clear speech, and automatically restore default parameters after the anomaly is resolved.

[0092] (2) Level 2 anomaly (moderate anomaly): moderate network congestion (packet loss rate 5%-10%, jitter 70-100ms), terminal echo suppression effect decay, and noise reduction distortion; processing flow: reduce the encoding bit rate to 64kbps and maintain the sampling rate at 16kHz; the improved NLMS algorithm triggers step size adjustment, μ(n) is increased to 0.05 to accelerate echo suppression convergence; the improved IMCRA algorithm enables impulse noise detection enhancement mode to avoid noise pollution; the terminal increases the feedback frequency to 250ms to synchronize the abnormal status in real time.

[0093] (3) Level 3 anomalies (serious anomalies): network interruption, explosion-proof telephone station / terminal failure, sudden strong noise (such as equipment explosion, pipeline leakage and instantaneous strong noise); processing flow: when the network is interrupted, the terminal automatically switches to the local emergency voice backup (pre-stored emergency instructions), adopts basic noise reduction and fixed gain playback, and quickly synchronizes the link parameters after the network is restored, seamlessly connecting to the normal process; when the equipment fails, the system automatically switches to the backup equipment, and the algorithm parameters are migrated synchronously; when sudden strong noise occurs, the improved IMCRA algorithm freezes the noise spectrum, the AGC gain is temporarily increased by 15dB, and the sending end strengthens the sound pickup and noise reduction intensity to ensure that the emergency instructions penetrate the strong noise.

[0094] S7: Core Indicators and Verification Standards for the Entire Process. Based on the core needs of petrochemical emergency broadcasting, the core indicators for the entire process operation are defined, the fusion effect of the improved algorithm is verified, and the process is ensured to meet the requirements of field applications.

[0095] (1) Sound pickup link indicators: noise suppression ratio ≥30dB, emergency voice recognition ≥95%, impulse noise suppression effect ≥28dB, high-frequency voice detail retention rate ≥90%;

[0096] (2) Transmission link indicators: end-to-end latency ≤300ms, stuttering rate ≤1% in weak network scenarios (packet loss rate 12%, jitter 120ms), FEC decoding recovery rate ≥98%, and no obvious voice interruption after PLC packet loss compensation;

[0097] (3) Playback link indicators: echo suppression ≥25dB, no howling, no popping, playback volume adaptation error ≤3dB under different noise scenarios, speech clarity ≥90% within 50m;

[0098] (4) Closed-loop adaptive indicators: parameter adjustment response time ≤ 100ms, scene switching adaptation time ≤ 200ms, and stable operation time ≥ 72 hours without human intervention throughout the entire chain.

[0099] Compared with the prior art, the technical effects and advantages of the present invention are as follows:

[0100] The audio end-to-end adaptive fusion implementation method of the emergency broadcasting system:

[0101] (1) The algorithm improvement is original: the conventional IMCRA, NLMS, FEC and jitter caching algorithms are improved in a targeted manner, and dynamic parameter adjustment, scene adaptation strategy and multi-dimensional feedback mechanism are introduced. Unlike the existing conventional algorithms, it can accurately adapt to the special scene of steady-state mechanical noise, non-steady-state wind noise, instantaneous impulse noise and large fluctuations and frequent sudden packet loss in the petrochemical plant area, and avoid the problems of incomplete noise suppression, voice distortion, echo residue and transmission stuttering that occur in the petrochemical scene.

[0102] (2) Full-link adaptive fusion: Integrating the three links of explosion-proof telephone station pickup, industrial IP transmission, and explosion-proof terminal playback, realizing algorithmic collaborative linkage and closed-loop feedback, solving the problem of independent operation and poor adaptability of existing system modules, and significantly improving the audio quality and transmission reliability of the entire link;

[0103] (3) Adapt to special petrochemical scenarios: In view of the characteristics of strong mechanical noise, large network fluctuations, varied playback scenarios and high explosion-proof requirements in petrochemical plants, the noise estimation algorithm, transmission strategy and gain control are optimized to ensure clear and stable emergency broadcasts and adapt to various petrochemical scenarios such as refineries, oil depots and pumping areas.

[0104] (4) High level of intelligence: No manual intervention is required throughout the process. The algorithm parameters are updated in real time according to environmental noise and network status, which reduces operation and maintenance costs and improves emergency response efficiency.

[0105] (5) High reliability: By using technologies such as improved FEC forward error correction, hierarchical PLC packet loss compensation, and improved NLMS echo suppression, combined with three major guarantee mechanisms of hardware adaptation, algorithm fault tolerance, and anomaly handling, the anti-interference capability of emergency broadcasting is improved, ensuring that emergency instructions are transmitted in a timely and accurate manner, and guaranteeing safe production in the petrochemical plant area.

[0106] (6) Strong compatibility: It is implemented by pure software algorithm and can be directly ported to the existing petrochemical emergency broadcasting system without the need for additional hardware equipment. The transformation cost is low and it is easy to promote and apply. Attached Figure Description

[0107] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0108] Figure 1 This is an overall architecture diagram of the audio end-to-end adaptive fusion implementation method for the petrochemical emergency broadcasting system of the present invention;

[0109] Figure 2 This is a detailed flowchart of the microphone pickup link of the explosion-proof telephone station of the present invention;

[0110] Figure 3 This is a detailed flowchart of the industrial IP network transmission link of the present invention;

[0111] Figure 4 This is a detailed flowchart of the playback link of the explosion-proof terminal of the present invention;

[0112] Figure 5This is a flowchart of the improved IMCRA adaptive noise power spectrum estimation algorithm of the present invention;

[0113] Figure 6 This is a block diagram of the end-to-end closed-loop collaboration and anomaly handling logic of the present invention. Detailed Implementation

[0114] In the following description, numerous specific details are set forth in order to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention can be practiced without one or more of these details. In other instances, certain technical features well-known in the art have not been described in order to avoid obscuring the invention.

[0115] Unless otherwise defined, the directions mentioned herein, such as up, down, left, right, front, back, inside, and outside, are based on the directions shown in the figures of this invention, and are explained here together.

[0116] This embodiment provides, for example Figures 1 to 6 The present invention illustrates an audio end-to-end adaptive fusion implementation method for an emergency broadcasting system, applied to an emergency broadcasting scenario in a pumping area of ​​a refinery. This scenario involves steady-state mechanical noise from the pump body (noise power P_avg≈0.06), outdoor non-steady-state wind noise, and slight fluctuations in the industrial IP network. The specific implementation steps are as follows:

[0117] In this embodiment, the explosion-proof microphone pickup link is implemented as follows:

[0118] Audio signal acquisition: The ExdIIBT6Gb explosion-proof microphone is used to acquire mixed audio signals x(n) on site, including emergency command voice, pump mechanical noise, and wind noise. The pickup distance is 3m, the sampling rate is 16kHz, and the quantization bit is 16bit.

[0119] Preprocessing: Frame length N = 256 points, frame shift L = 128 points, Hanning window is used for windowing. The Hanning window formula is as follows: The signal of the mth frame after framing Signal after windowing Effectively suppresses spectrum leakage;

[0120] STFT Time-Frequency Transform and Power Spectrum Calculation: Perform a 256-point FFT transform on each windowed signal frame to obtain the complex spectrum in the frequency domain. Calculate the power spectrum at each frequency point , serving as the basic data for noise estimation and denoising;

[0121] Improved IMCRA Adaptive Noise Power Spectrum Estimation: This scenario involves both steady-state mechanical noise and non-steady-state wind noise, and the smoothing factor is dynamically adjusted. When pump body mechanical noise dominates When the wind noise has a significant impact, take 0.94. Take 0.91; calculate the a posteriori signal-to-noise ratio at each frequency point. Simultaneously calculate the energy of the current frame. Multi-frame average energy ,like If a frame is identified as an instantaneous impulse noise frame (such as pump start-up / shutdown noise), the noise spectrum is updated using a neighboring frame interpolation strategy. ; By using soft VAD decision, when <3dB and At that time, according to Update the noise spectrum; freeze the noise spectrum when γ(m,k)≥3dB.

[0122] Improved prior signal-to-noise ratio estimation and noise reduction gain calculation: Introducing a noise type weighting factor ω, where ω = 0.96 for steady-state mechanical noise, ω = 0.98 for non-steady-state wind noise, and ω = 0.94 for impulse noise; taking α = 0.96 as the prior signal-to-noise ratio smoothing factor, and using the formula... Estimate the prior signal-to-noise ratio; introduce a high-frequency compensation factor. Noise reduction gain To avoid losing high-frequency speech details;

[0123] Frequency domain reconstruction and time domain synthesis:

[0124] Correcting the frequency domain spectrum through noise reduction gain ;

[0125] Perform an inverse IFFT transform on the corrected spectrum to obtain the time-domain signal. The system uses the Overlapping Addition (OLA) algorithm to splice continuous time-domain audio and outputs a clean emergency voice x̂(n) after noise reduction. This voice is then sent to an industrial IP audio transmission module. Testing shows that the link has a noise suppression ratio of 32dB and an emergency voice recognition rate of 96%, meeting the preset targets.

[0126] In this embodiment, the industrial IP network transmission link is implemented as follows:

[0127] Audio encoding and RTP encapsulation: for clean emergency voice recordings It adopts OPUS encoding, with initial encoding parameters configured for medium network conditions (64kbps bitrate, 16kHz sampling); it is encapsulated according to the RTP standard, and RTPeader contains sequence number Seq (starting value is 0, incrementing by 1 for each frame sent), timestamp (incrementing synchronously at 16kHz sampling rate), and SSRC identifier (uniquely identifying the sender). The encapsulation format is RTPPkt=RTTPeader+AudioPayload+FEC redundant data, which is compatible with industrial IP transmission protocols.

[0128] Real-time network status monitoring and classification: The sending end receives network parameters reported by the terminal via the RTCP protocol (feedback period 500ms) and calculates the packet loss rate in real time. Latency jitter, real-time available bandwidth The measured average packet loss rate of the industrial IP network in this pumping area was 4%, and the jitter was 65ms, which was determined to be a medium-sized network.

[0129] Adaptive bitrate and sampling rate adjustment: First-order smoothing filtering is used to avoid bitrate jumps. λ=0.92 is taken as the bandwidth smoothing factor, according to the formula... Adjust the bandwidth adaptation parameters to maintain a bitrate of 64kbps and a sampling rate of 16kHz, balancing audio quality and bandwidth.

[0130] Improved Adaptive FEC Forward Error Correction Coding: Calculate the standard deviation of packet loss rate over the past 10 frames σ = 0.8, and dynamically adjust the redundancy weighting coefficient β = 1.2 + 0.3 × 0.8 = 1.44; under medium network conditions, the number of original audio packets per group k = 8, and the redundancy... That is, each group of 8 original packets generates 1 redundant packet, of which the core redundant packet accounts for 60% (0.6 packets, take 1), and the supplementary redundant packet accounts for 40% (0.4 packets, not inserted for the time being). The core redundant packet is transmitted synchronously with the original packets, and the receiving end decodes it through RS to ensure that it can be quickly recovered in case of packet loss.

[0131] Improved dynamic jitter buffer and PLC packet loss compensation: =4 frames, γ=0.01, η=0.005, current end-to-end delay Delay(m)=220ms, according to the formula Cache utilization is maintained at 40%-60%, no adjustment to read speed is needed; when single-frame packet loss occurs, temporal interpolation is used. For compensation, when two consecutive frames are lost, "interpolation + prediction of the amplitude trend of the previous 10 frames" is used to ensure that there is no voice interruption.

[0132] Receiver timing synchronization and decoding: The receiver reorders out-of-order packets according to the RTP sequence number, reads audio frames at a constant speed through a jitter buffer, and performs smooth fade-in and fade-out splicing to avoid splicing distortion; the spliced ​​audio frames are OPUS decoded, and the decoded audio signal y(n) is output and sent to the explosion-proof broadcast terminal playback module. After testing, the end-to-end latency is 240ms and the stuttering rate is 0.3%, which meets the requirements of ≤300ms latency and ≤1% stuttering rate.

[0133] In this embodiment, the explosion-proof terminal playback link is implemented as follows:

[0134] Dual signal synchronous acquisition: The explosion-proof broadcast terminal synchronously acquires and decodes the audio signal y(n) (playback source signal) and the mixed signal z(n) (pump mechanical noise + speaker backflow echo) acquired by the terminal's built-in microphone.

[0135] The z(n) is preprocessed by framing and windowing according to the pickup link parameters to prepare for echo suppression and noise detection;

[0136] Improved NLMS adaptive filtering echo suppression: The step size μ(n) is dynamically adjusted according to the echo error e(n). When the echo is strong, μ(n) is 0.045, and when the echo is weak, μ(n) is 0.015. The Euclidean distance between the filter coefficients of adjacent frames is calculated. When d(n) > 0.05, the average value of the coefficients from the previous 5 frames is used to reset the filter coefficients; when y(n) is detected as a speech frame, the filter coefficient update rate is multiplied by a weight of 0.5 to avoid over-filtering of the speech signal; according to the formula The formula for iteratively updating the filter coefficients to cancel echoes is as follows: The signal e(n) after echo cancellation is further suppressed by the improved IMCRA algorithm, and a clean noise reference signal ê(n) is output. The echo suppression is 28dB with no echo residue.

[0137] Environmental noise intensity detection and classification:

[0138] Calculate the average noise power of a single frame ,Pick =0.9 is the noise intensity smoothing factor, according to Smoothing was performed on the pump area, as measured in actual measurements. It was determined to be at a high noise level.

[0139] Adaptive AGC gain control: Setting the reference gain Gain adjustment step size , High noise level ;Pick The gain smoothing factor is calculated according to the formula. Smoothly adjust the gain to avoid abrupt gain changes;

[0140] Playback output and closed-loop feedback: This refers to the playback source signal. With adaptive gain Multiply and then apply amplitude limiting; the formula is as follows: ;

[0141] (The maximum amplitude of 16-bit PCM audio is 32768); The signal was played through the terminal's explosion-proof speaker. Actual testing showed a 92% voice clarity rate within 50 meters, with no popping or feedback. The terminal then switched to a high-noise level (…). The signal is fed back to the sending end in real time via the RTCP protocol, which adjusts the noise reduction intensity of the audio pickup link accordingly. Adjusted to 0.93), transmission link FEC redundancy (r adjusted to 2), to achieve full-link closed-loop coordination.

[0142] In this embodiment, the implementation of end-to-end closed-loop collaboration, landing assurance, and anomaly handling specifically includes:

[0143] End-to-end closed-loop coordination: The transmitting end receives high-noise level information from the receiving terminal and, combined with the network status (medium network), synchronously adjusts the parameters of each link: sound pickup link. The noise reduction strength was improved by adjusting the value from 0.94 to 0.93; the FEC redundancy r of the transmission link was increased from 1 to 2 to improve the ability to resist packet loss; the playback link was maintained. =12dB, ensuring clear speech; after each frame of data is processed, the status is fed back through RTCP to realize real-time updates of algorithm parameters;

[0144] Implementation and Support: In terms of hardware compatibility, both the explosion-proof base stations and terminals adopt an ExdIIBT6Gb-level explosion-proof design, the microphones and speakers use explosion-proof structures, and the equipment casing employs a shielded design to avoid electromagnetic interference. Regarding algorithm operation, the system automatically collects on-site noise and network status upon startup and initializes... =0.94, β=1.44, μ(n)=0.045 and other core parameters; the algorithm output is calibrated every 100 frames of data to ensure speech recognition ≥95%, stuttering rate ≤1%, and echo suppression ≥25dB; in terms of environmental adaptation, the pump area is at room temperature (25℃) and humidity 60%, so there is no need to adjust the frame length and When switching scenes (e.g., from the pump area to an adjacent ordinary workshop), (≈0.03, medium noise level), parameter coordination adjustment is completed within 100ms;

[0145] Anomaly Handling Implementation: During the operation of this scenario, a Level 1 anomaly occurred (a brief start-stop of the pump caused a sudden change in noise, E(m) > 3·Ē(m)). The system automatically executed the following processing steps: the improved FEC algorithm β was increased to 1.3, the dynamic jitter buffer depth was increased by 1 frame (becoming 5 frames), and the improved IMCRA algorithm... The value was reduced by 0.02 (becoming 0.91), and the AGC gain was temporarily increased by 3dB (becoming 15dB). After the anomaly was resolved (the pump resumed stable operation), the default parameters were automatically restored. No level 2 or level 3 anomalies occurred, and the entire chain operated without interruption, meeting the requirement of stable operation without human intervention for 72 hours.

[0146] This embodiment was applied to an emergency broadcast scenario in a petrochemical plant's pump area. Field testing showed that all indicators across the entire link met the preset standards: noise suppression ratio of the pickup link was 32dB, emergency voice recognition was 96%, impulse noise suppression was 30dB, and high-frequency voice detail retention was 92%; end-to-end latency of the transmission link was 240ms, stuttering rate in a weak network simulation scenario (packet loss rate 12%, jitter 120ms) was 0.8%, and FEC decoding recovery rate was 99%; echo suppression of the playback link was 28dB, volume adaptation error for different noise scenarios was 2dB, and voice clarity within 50m was 92%; closed-loop adaptive parameter adjustment response time was 80ms, and scene switching adaptation time was 150ms. It fully adapts to the complex operating conditions of a petrochemical pump area, solving the pain points of incomplete noise suppression, transmission stuttering, and echo residue in existing technologies, ensuring clear and timely transmission of emergency commands, and guaranteeing safe production in the plant area.

[0147] It should be noted that, in this document, relational terms such as "one" and "two" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, the phrase "comprising an element defined as..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0148] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for adaptive fusion of the entire audio link in an emergency broadcasting system, characterized in that, Includes the following steps: The explosion-proof voice station pickup link collects mixed audio signals from the scene, and after preprocessing, STFT time-frequency transformation, improved IMCRA adaptive noise power spectrum estimation, improved prior signal-to-noise ratio estimation and noise reduction gain calculation, frequency domain reconstruction and time domain synthesis, it outputs a clean emergency voice signal. The industrial IP network transmission link performs OPUS encoding and RTP encapsulation on the pure emergency voice signal, monitors the network status in real time and classifies it, dynamically adjusts the encoding bit rate and sampling rate, and completes stable signal transmission and decoding through improved adaptive FEC forward error correction coding, improved dynamic jitter buffer and PLC packet loss compensation, and outputs the decoded audio signal. The explosion-proof terminal playback link synchronously collects the decoded audio signal and the mixed signal at the terminal site. It uses an improved NLMS adaptive filter to cancel the speaker backflow echo, detects and classifies the ambient noise level, performs adaptive AGC gain control, and plays the signal after amplitude limiting. At the same time, it feeds back the ambient noise level to the transmitting end in real time. The end-to-end closed-loop collaboration and assurance mechanism enables adaptive adjustment of algorithm parameters through data linkage, parameter coordination, and real-time updates. Combined with hardware adaptation, algorithm fault tolerance, and environment adaptation, it ensures stable implementation of the process and establishes a hierarchical anomaly handling mechanism to deal with sudden anomalies.

2. The method according to claim 1, characterized in that, The improved IMCRA adaptive noise power spectrum is equipped with an impulse noise detection mechanism. It calculates the ratio of the current frame energy E(m) to the average energy Ē(m) across multiple frames. When E(m) > 3·Ē(m), it is determined to be an instantaneous impulse noise frame. A neighboring frame interpolation strategy is then used to update the noise spectrum. To avoid pulse noise contaminating the noise substrate.

3. The method according to claim 2, characterized in that, The improved IMCRA adaptive noise power spectrum also includes a dynamic smoothing factor adjustment mechanism and a soft VAD decision mechanism. The dynamic smoothing factor adjustment mechanism dynamically adjusts the smoothing factor according to the noise type. Steady-state mechanical noise scenario Using a value of 0.93-0.95, for non-steady-state wind noise scenarios. Take 0.90-0.92; The soft VAD decision mechanism has a post-hoc signal-to-noise ratio. and The frame is identified as a steady-state noise frame and execution is performed. Recursively smooth the noise spectrum when The system is then identified as containing a speech frame and the noise spectrum is frozen.

4. The method according to claim 3, characterized in that, The improved adaptive FEC forward error correction coding introduces a network fluctuation coefficient. The standard deviation of packet loss rate over the past 10 frames, dynamically adjusted redundancy weighting coefficient. ; The number of original audio packets k in each group is dynamically adjusted according to the network level. For high-quality networks, k=10; for medium-quality networks, k=8; and for weak networks with congestion, k=6. The redundant packets are divided into core redundant packets (60%) and supplementary redundant packets (40%). Core redundant packets are transmitted synchronously with the original packets, and supplementary redundant packets are dynamically inserted according to the bandwidth to avoid preempting the transmission resources of the original packets. The formula for calculating redundancy is: .

5. The method according to claim 4, characterized in that, The improved dynamic jitter buffer and PLC packet loss compensation, the dynamic buffer depth calculation formula is as follows: ， in =4 frames, γ=0.01, η=0.005, establish a cache status early warning mechanism, adjust the reading speed according to the cache usage rate, and adopt a hierarchical PLC packet loss compensation strategy. For single frame packet loss, time domain interpolation is used; for 2-3 consecutive frame packet loss, interpolation + voice feature prediction is used; for ≥4 consecutive frame packet loss, a silent transition + subsequent frame fade-in is used.

6. The method according to claim 5, characterized in that, The improved NLMS adaptive filtering echo suppression adopts a variable step size adaptive strategy, which dynamically adjusts the step size μ(n) according to the echo error e(n). When |e(n)| is large, μ(n) takes 0.04 to 0.05, and when |e(n)| is small, μ(n) takes 0.01 to 0.

02. Abrupt changes in the echo path are detected by the Euclidean distance between the filter coefficients of adjacent frames, triggering a reset of the filter coefficients. A voice protection mechanism is introduced to avoid excessive filtering and distortion of the voice signal. The iterative update formula for the filter coefficients is as follows: ,in .

7. The method according to claim 6, characterized in that, The explosion-proof terminal playback link is equipped with an environmental noise detection and adaptive AGC gain control mechanism, which calculates the average noise power of a single frame and obtains it after multi-frame smoothing. The noise levels are divided into low noise level, medium noise level and high noise level. Gain is allocated based on noise level using a first-order recursive smoothing algorithm. Low noise level: Medium noise level: High noise level: ; Adaptive AGC gain control employs a first-order recursive smoothing algorithm, allocating gain according to noise level, with a low noise level... , Medium noise level ; High noise level , ; The smoothing adjustment formula is: ,in ∈[0.92,0.98].

8. The method according to claim 1, characterized in that, The full-link closed-loop coordination and protection link uses data linkage to feed back the noise reduction effect of the sound pickup link and the environmental noise level of the terminal link to the transmission link in real time, so that the encoding parameters of the transmission link can be adjusted in coordination according to the noise environment of the terminal, and the noise reduction parameters of the sound pickup link can be optimized according to the actual reception effect of the terminal. By coordinating parameters, the intensity of sound pickup and noise reduction, FEC redundancy, and playback gain are simultaneously improved and the encoding bitrate is reduced in high-noise and weak network scenarios, while the encoding bitrate is improved and the FEC redundancy and playback gain are reduced in high-quality network and low-noise scenarios. By updating in real time, the parameters of all improved algorithms are updated adaptively in real time through RTCP feedback status after each frame of data is processed in the entire link.

9. The method according to claim 1, characterized in that, The graded anomaly handling includes: Level 1 anomalies are handled by temporary parameter adjustment and automatic recovery; Level 2 anomalies are handled by encoding rate adjustment, algorithm parameter optimization and feedback frequency increase; Level 3 anomalies are handled by backup operation strategy. When the network is interrupted, the terminal automatically switches to local emergency voice backup, uses basic noise reduction and fixed gain playback, and quickly synchronizes link parameters after the network is restored, seamlessly connecting to the normal process.

10. The method according to claim 1, characterized in that, The safeguards include: The hardware compatibility guarantee mechanism ensures that both the explosion-proof telephone station and the explosion-proof broadcast terminal meet the petrochemical explosion-proof level requirements, and the equipment shell adopts a shielded design. The algorithm fault tolerance mechanism automatically switches to backup algorithm logic when the algorithm malfunctions, and performs output effect calibration every 100 frames of data. The environmental adaptation guarantee mechanism supports dynamic adjustment of parameters based on ambient temperature and humidity, and completes parameter coordination adjustment within a preset time when switching scenes.

Images