A built-in prison intelligent voice risk early warning system

Abstract

The present application relates to the technical field of intelligent security monitoring, and specifically discloses a built-in prison intelligent voice risk early warning system, which comprises a plurality of wearable thin card terminals and an intranet early warning receiving terminal; the thin card terminal can be sewn into the lining of prison clothes, and the terminal is internally integrated with a power supply unit, a voice collection unit, a local offline voice operation unit, a local storage unit and a radio frequency early warning transmission unit; the overall size of the matched thin card terminal in the present application is 75mm*60mm*25mm, the overall weight is controlled to be less than 88g, a sewing fixing hole is reserved on the side of the shell, the thin card terminal can be directly sewn into the chest and shoulder lining of the prison clothes by cotton thread, the device is completely hidden between the two layers of cloth, there is no exposed hard protrusion, and the device will not interfere with the daily behaviors of the prisoners, such as living, working, clothes washing and the like.

Description

Technical Field

[0001] This invention relates to the field of intelligent security monitoring technology, and in particular to an embedded intelligent voice risk early warning system for prisons. Background Technology

[0002] Currently, domestic prisons, detention centers, compulsory drug rehabilitation centers and other supervisory facilities are equipped with traditional security monitoring systems. These systems rely solely on wall-mounted fixed cameras and fixed external audio pickup equipment in conjunction with manual on-site inspections for risk management. The only core flaw is that the existing audio monitoring equipment cannot follow the movement of detainees for monitoring.

[0003] Traditional audio monitoring equipment consists of cabinet-style or wall-mounted fixed devices, which are large, conspicuous, and easily obscured or damaged by detainees. These devices rely on wired power or high-capacity hard batteries, requiring regular on-site maintenance and wiring repairs, resulting in significant manpower investment. Furthermore, existing equipment utilizes a cloud-based voice recognition architecture, uploading all raw human voice data to an external server in real time. Given that prisons are classified and controlled areas, transmitting audio data over the internet poses a serious risk of information leakage. Conventional audio pickup equipment lacks hardware noise reduction optimization for the complex noise generated by machinery in the prison workshops, looped broadcasts, and simultaneous conversations by multiple people. In noisy environments, keyword recognition accuracy is less than 70%, leading to frequent false alarms and missed alarms. Prolonged ineffective warnings can cause complacency among supervisory personnel, negating the value of early risk warnings. The fixed-point deployment mode has a large number of monitoring blind spots, and private spaces such as cell corners and work stations cannot be covered. It is difficult to detect high-risk conversations such as collusion, fighting, and escape planning in advance. After a security incident, clues can only be found by reviewing video footage, resulting in a serious delay in early warning. Therefore, in order to solve these problems, an embedded intelligent voice risk early warning system for prisons is proposed. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an embedded intelligent voice risk warning system for prisons.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] An embedded intelligent voice risk warning system for prisons includes several wearable thin card terminals and an intranet warning receiving terminal;

[0007] The thin-card terminal can be sewn into the lining of prison uniforms. The terminal integrates a power supply unit, a voice acquisition unit, a local offline voice processing unit, a local storage unit, and a radio frequency early warning transmission unit.

[0008] The entire voice recognition and risk assessment process is completed offline on the local thin card terminal without accessing the external network. After recognizing high-risk voices, it only pushes graded silent warning signals to the internal network receiving terminal.

[0009] Preferably, the thin card terminal adopts a three-layer integrated sealed shell, which consists of a flexible solar power supply layer, a main control and computing layer, and a micro energy storage layer from top to bottom;

[0010] The power supply unit consists of a dual power supply structure of a flexible amorphous silicon thin film and an ultra-thin soft-pack lithium battery, and is equipped with a dedicated charge and discharge management chip. The overall dimensions of the casing are 75mm×60mm×25mm, and the weight of the whole machine does not exceed 88g.

[0011] The device's static standby power consumption is ≤10μA, and the continuous working time of the lithium battery in a dark, sealed environment is ≥48h.

[0012] The shell is sealed with IP67 polyurethane vacuum potting and has sewing holes on the side for direct sewing and fixing to the prison uniform lining. The shell is equipped with a copper foil electromagnetic shielding layer.

[0013] Preferably, the local offline speech processing unit adopts a dual collaborative architecture of LD3320 offline recognition chip and STM32 low-power microcontroller, and locally burns the prison's exclusive high-risk sensitive word library, without the need for cloud training;

[0014] The voice acquisition unit is equipped with a Φ4mm waterproof and noise-reducing mini microphone, and is paired with a multi-level passive hardware filtering circuit to filter out composite noise in the monitoring area;

[0015] Keyword recognition response time ≤ 0.5s, and recognition accuracy in noisy prison environments ≥ 95%;

[0016] After a risky word is detected, the terminal's built-in micro-vibration motor vibrates in stages. The local storage unit uses an SPI flash memory chip to encrypt and retain the warning log, and the original voice is not transmitted to the outside.

[0017] Preferably, the radio frequency early warning transmission unit is a 433M low-power radio frequency transmitter board, which only starts transmission briefly when high-risk words are detected, and is powered off and put into sleep mode at other times; all radio frequency transmission messages are encrypted using the AES-256 encryption algorithm, and the signal is only transmitted within the prison's local area network, without an external network communication channel.

[0018] Preferably, the sewn hole of the housing penetrates through the front and rear walls of the housing. When the housing sealant is broken by human intervention and the housing is separated, the computing unit automatically generates an anti-tampering abnormality log and stores it in the local flash memory. The log cannot be manually deleted.

[0019] Preferably, the casing has a hidden micro debugging interface, which is only enabled when the sensitive word library is burned at the factory and the logs are exported locally. During daily use, the interface is sealed with sealant.

[0020] A narrow slit sound guide hole is made on the housing corresponding to the microphone position. The width of the guide hole is less than 2mm to prevent the sound pickup channel from being blocked by human intervention.

[0021] Preferably, the arithmetic unit has a built-in hardware reset switch, and the reset contact is enclosed inside the potting housing.

[0022] Preferably, the charge and discharge management chip integrates overcharge, over-discharge, and short-circuit protection circuits. When the lithium battery voltage is lower than a set threshold, it automatically writes a low battery log and pushes a low battery reminder message to the receiving terminal.

[0023] Preferably, the shell material is ABS+PC composite engineering plastic. After the filling is completed, all the edges of the shell are rounded and polished. After sewing and fixing, there are no hard protrusions, so as not to affect the daily work, living and washing of clothes of the detainees.

[0024] Preferably, the intranet early warning receiving terminal is deployed in the duty room of the prison area, and the receiving terminal displays early warning information in pop-up windows according to low, medium and high risk levels;

[0025] The system is equipped with a dedicated local log export device. Only authorized personnel can access the terminal's hidden interface to read encrypted and stored logs, and all export operations are automatically recorded for audit purposes.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] 1. The overall dimensions of the matching thin-card terminal in this invention are 75mm×60mm×25mm, and the weight of the whole machine is controlled within 88g. The side of the shell has reserved sewing and fixing holes, which can be directly sewn into the chest and shoulder lining of the prison uniform with cotton thread. The device is completely hidden between two layers of fabric, with no exposed hard protrusions, and will not interfere with the daily activities of detainees such as living, working, and washing clothes. The terminal moves synchronously with the detainees, and can achieve full-time audio monitoring in areas that traditional fixed monitoring equipment cannot cover, such as blind spots in the cell and private work positions, eliminating audio blind spots in supervision. The shell adopts IP67 polyurethane vacuum potting process, combined with ABS+PC wear-resistant composite plastic, which is resistant to sweat corrosion and water washing and soaking. It can withstand thousands of bending without circuit breakage or sealing failure. The internal ultra-thin copper foil shielding layer isolates the electromagnetic interference of the cell block intercom and the monitoring host. The sound pickup and recognition stability is significantly better than traditional external fixed audio pickup equipment, which is suitable for the 24 / 7 uninterrupted operation requirements of prisons.

[0028] 2. In this invention, all voice acquisition, keyword matching, and risk assessment logic are completed locally on the thin-card terminal. The entire process does not connect to the internet or transmit raw human audio to external servers. The terminal's built-in W25Q series flash memory chip provides local encrypted storage of warning, fault, and anti-tamper logs. All stored and transmitted messages are encrypted using the AES-256 encryption algorithm. Only encrypted short warning messages are pushed to the receiving terminal in the duty room via the intranet. The raw dialogue audio is never output externally. Device debugging and log reading can only be performed locally through a sealed, hidden interface. All data reading operations are automatically recorded in audit logs. A tiered access control mechanism is implemented, strictly adhering to the principle of minimum necessary acquisition. Records are only retained when high-risk keywords are matched, avoiding indiscriminate storage of all dialogue audio. This balances supervisory security with the protection of detainees' personal information, mitigating compliance risks associated with excessive surveillance.

[0029] 3. This invention innovatively employs a dual-power supply architecture combining a flexible amorphous silicon solar film and an ultra-thin soft-pack lithium battery in the thin-card terminal. The flexible film adheres to the outer fabric of the prison uniform, allowing continuous charging of the lithium battery even in low-light environments (50 lux) such as indoor lighting or corridors, without the need for direct sunlight. The film can be bent 180 degrees without power loss after 500 bending cycles, making it suitable for scenarios involving repeated folding and washing of clothing. Equipped with a dedicated TP4057 charge / discharge management chip, integrating multiple circuit protections, the device's static standby power consumption is only 8μA. In a dark, sealed cell, the lithium battery can operate continuously and stably for up to 52 hours, far exceeding the basic design specification of 48 hours. Compared to traditional wired, high-capacity battery security equipment, this eliminates the need for on-site wiring, regular battery replacements, and individual battery checks. The device features a built-in low-battery automatic alert mechanism, with the duty room terminal centrally recording the low-battery device numbers, allowing staff to perform batch maintenance, significantly reducing the manpower required for equipment maintenance in the prison area.

[0030] 4. This invention features hardware and software optimization for multi-source noise environments in prison areas. A Φ4mm miniature waterproof microphone, paired with a multi-level passive hardware filter circuit, filters out power frequency noise and low-frequency mechanical impact noise. In noisy environments, the accuracy rate for identifying high-risk keywords remains stable at over 95%, with a false alarm rate of less than 3%, and a response time controlled within 0.5 seconds. The terminal is equipped with a miniature flat vibration motor with three vibration levels to differentiate risk levels. Only the prison guard's indoor network terminal simultaneously receives tiered warnings. The warning process is silent, avoiding disturbance to inmates and facilitating discreet risk management. The terminal can locally program differentiated sensitive word libraries adapted to prisons, detention centers, and drug rehabilitation centers. It also features multi-dimensional anomaly monitoring functions, including equipment disassembly, circuit failure, and low battery monitoring. In case of human-caused damage or circuit failure, warning information is immediately pushed, allowing management to investigate and handle the situation promptly. This comprehensive approach covers the entire process from pre-event risk identification and in-event tiered warnings to post-event log traceability, effectively reducing the probability of security incidents such as fights, collusion, and prison break plots. Attached Figure Description

[0031] Figure 1 This is a functional block diagram of a wearable thin-card terminal for an embedded intelligent voice risk early warning system for prisons, as proposed in this invention.

[0032] Figure 2 This is a block diagram of a three-layer hardware stacked structure of a thin-card terminal for an embedded intelligent voice risk early warning system for prisons, as proposed in this invention.

[0033] Figure 3 This is a block diagram of the power supply unit of an embedded intelligent voice risk early warning system for prisons proposed in this invention.

[0034] Figure 4 This is a network architecture block diagram of an embedded intelligent voice risk early warning system for prisons proposed in this invention. Detailed Implementation

[0035] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0036] In the description of this invention, it should be understood that the terms "upper", "lower", "front", "rear", "left", "right", "top", "bottom", "inner", "outer", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this 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 this invention.

[0037] Reference Figure 1-4This invention relates to an embedded intelligent voice risk early warning system for prisons. The entire system consists of three main parts: multiple wearable thin-card terminals, an intranet early warning receiving terminal in the duty room, and a dedicated local log export device. Each inmate is equipped with an independent thin-card terminal. All thin-card terminals connect to the same prison intranet using a 433M radio frequency signal. All data interaction is completed only within the closed intranet of the prison, without any external network communication links. An intranet early warning receiving terminal is deployed in the duty room to uniformly collect the graded early warning messages pushed by all terminals. The log export device is kept only by the prison area management supervisor. With exclusive access, the data cable is connected to the hidden debugging interface of the thin-card terminal to read the encrypted stored logs. All export operations are automatically written to the terminal's flash memory for audit records. The thin-card terminal has a rounded rectangular integrated thin-card structure. The shell is made of ABS+PC composite engineering plastic injection molding. The external dimensions are strictly controlled to be 75mm long, 60mm wide, and 25mm thick. The actual measured net weight of a single unit is 86g, which is lower than the design limit of 88g. The internal structure of the casing adopts a three-layer stacked structure, consisting of a flexible solar power supply layer, a main control and computing layer, and a micro-energy storage battery layer from top to bottom. The three modules are tightly bonded together with high-temperature resistant insulating adhesive. A 0.03mm ultra-thin copper foil electromagnetic shielding layer is glued to the entire casing, completely covering all circuit boards and isolating electromagnetic noise generated by metal doors and windows, walkie-talkies, and monitoring hosts in the prison area, thus avoiding interference with the accuracy of the voice recognition chip. Four symmetrical oval sewing holes with a diameter of 3mm are made on the side of the casing, penetrating the front and rear walls of the casing. During on-site installation, ordinary cotton thread is used to sew the terminals together in a cross pattern, embedding and fixing them between the two layers of fabric lining the prison uniform's chest. After sewing, the outline of the hard equipment is not visible from the outside of the prison uniform, and there are no protruding edges, so detainees do not experience any discomfort from foreign objects while walking, working, or resting. The housing has only one narrow 1.8mm slit for sound transmission, corresponding to the internal miniature microphone installation location. The slit is 12mm long, ensuring omnidirectional voice pickup within a 3-meter range while preventing detainees from blocking the sound pickup channel with cloth or tape. Even with a small amount of cloth covering the slit, the microphone can still capture close-range conversations normally. After the housing is assembled, it is placed in a vacuum potting machine and injected with IP67-rated polyurethane waterproof adhesive. The air inside the housing is purged in a vacuum environment, and the adhesive completely fills the gaps between the modules. After curing, it forms a one-piece sealed structure with no water ingress channels, meeting the requirements for washing, sweatproofing, and dustproofing.

[0038] Part 1: Implementation Details of the Thin Card Terminal Power Supply Unit

[0039] The power supply unit consists of three parts: a flexible solar thin-film unit, a micro soft-pack lithium battery unit, and a TP4057 charge / discharge management chip unit, all integrated into the upper and lower layers of the casing. The flexible solar thin film is a 20mm×40mm amorphous silicon flexible film with a thickness of only 0.1mm. A breathable insulating fabric is attached to the outer surface of the film, which is directly attached to the outer fabric of the prison uniform. The film is connected to the charge / discharge management chip inside the casing via an ultra-thin FPC flexible cable. This film can stably output a 1.5V charging voltage in a 50lux indoor low-light environment, continuously replenishing the lithium battery without direct sunlight. The film can be bent 180 degrees at will, and can be repeatedly bent 500 times without breakage or power loss, making it suitable for repeated folding and washing of prison uniforms. The micro energy storage lithium battery uses a 3.7V, 80mAh ultra-thin soft-pack polymer lithium battery, only 1.0mm thick, without a rigid metal casing, and is completely housed in the energy storage layer under the casing. The two terminals of the battery are soldered to the charge and discharge management chip via surface-mount wires. The battery is wrapped with insulating buffer foam to prevent short circuits between the battery and the main control circuit board. It is equipped with a 1N5819 Schottky anti-reverse current diode to prevent reverse current from flowing back into the lithium battery and damaging the solar film in the absence of light. The charge and discharge management unit uses an SOT-23-6 surface-mount TP4057 chip, which integrates overcharge, over-discharge, and short-circuit triple hardware protection logic. When the lithium battery voltage reaches 4.2V, it automatically cuts off the solar charging circuit; when the voltage drops below 2.8V, it cuts off the power supply to the entire device and sends a low battery signal to the main control microcontroller. After receiving the signal, the microcontroller writes a low battery log to its local flash memory and pushes a low battery reminder message to the receiving terminal in the duty room via the radio frequency module. The static standby power consumption of the entire power supply unit is only 8μA, which is lower than the design threshold of 10μA. The power consumption of the device increases to 25mA only when the microphone picks up a valid human voice and the voice chip starts the recognition operation. When there is no human voice, the device automatically enters the low power consumption sleep mode. In a dark, closed cell environment, the lithium battery can run continuously and stably for 52 hours when fully charged, which exceeds the basic design standard of 48 hours.

[0040] Part Two: Implementation Structure of Voice Acquisition Unit and Local Offline Voice Processing Unit

[0041] The voice acquisition unit uses a 4015-specification Φ4mm miniature waterproof electret microphone. A waterproof and sound-permeable film with a 0.02mm pore size is attached to the front of the microphone to prevent sweat and water vapor from entering the microphone and causing short circuits. The microphone is equipped with a 5cm long flexible lead wire, which is connected to a two-stage passive filter circuit built with 0402 packaged surface mount resistors and ceramic capacitors. The filter circuit filters out 50Hz power frequency AC noise and low-frequency noise from mechanical impacts in the monitoring area, improving the signal-to-noise ratio of human voice signals by 22dB and significantly reducing recognition interference in noisy environments. The local offline voice processing unit uses an LD3320A offline voice recognition chip paired with an STM32L051 low-power microcontroller in a dual-chip collaborative architecture. Both chips are soldered onto a 12mm×10mm ultra-thin PCB main control board using SMT surface mount technology. The main control board is housed in the main control processing layer in the middle of the casing. Before leaving the factory, the LD3320A chip uses a hidden micro-USB debugging interface within its casing to locally program a high-risk, sensitive word library specific to prisons. It can store up to 60 custom keywords at a time, covering high-risk terms related to prison control such as fighting, provocation, collusion, escape plans, and violent threats. It only supports Mandarin recognition and requires no cloud-based voice sample training. After programming, the debugging interface is sealed with waterproof sealant, ensuring the device operates entirely offline with no data transmission. The chip analyzes the microphone input voice signal in real time, matching it against the built-in keyword library. Once a high-risk word is detected, a trigger signal is immediately sent to the STM32L051 microcontroller. The STM32L051 microcontroller has a main frequency of 32MHz and built-in 32KB Flash and 8KB RAM. Upon receiving a trigger signal, it simultaneously executes three instructions: First, it drives the built-in Φ8mm miniature flat vibration motor to output graded vibrations: a short 0.3-second vibration for low-risk words, three consecutive 0.5-second vibrations for medium-risk words, and continuous strong vibrations for high-risk words; second, it wakes up the 433MHz RF transmitter module to send an encrypted warning short message; third, it controls the W25Q64 flash memory chip to write an encrypted warning log. The log only records four pieces of information: trigger time, device number, matching keywords, and real-time device battery level, without storing any original human audio. The main control PCB is equipped with a surface-mount miniature tactile reset switch. The reset contacts are completely hidden inside the potted housing, with no external operation buttons. Manual reset can only be performed after disassembling the housing, preventing detainees from maliciously restarting the device and erasing the locally stored log. If the housing sealant is broken or the housing is separated, the internal anti-tamper detection circuit is broken, and the microcontroller immediately writes a permanent, undeletable device damage anomaly log for evidence collection.

[0042] Part Three: Implementation Methods of Local Storage Unit and Radio Frequency Early Warning Transmission Unit

[0043] The local storage unit uses W25Q64SPI flash memory chips with a storage capacity of 8MB. These chips are soldered to the back of the main control PCB board. The chips have no moving mechanical parts, are shock-resistant, and can withstand high and low temperatures. They can stably store no less than 1200 warning, fault, low power, and anti-tampering abnormality logs. All data written to the flash memory is encrypted by the microcontroller using the AES-256 encryption algorithm. Logs can only be exported locally through a hidden debugging interface. The export process automatically records the access time and the number of the operating device, forming a complete audit traceability record. The flash memory retains data for no less than 10 years after power failure, meeting the needs of post-event traceability and evidence collection in prisons and detention centers. The radio frequency (RF) early warning transmission unit consists of a 20mm×15mm 433M low-power RF transmitter board. The transmitter board only briefly wakes up when the microcontroller receives a high-risk word recognition signal, with each message transmission lasting no more than 0.2 seconds. During the remaining time, the RF module is completely powered off and in sleep mode, resulting in no unnecessary power consumption. The RF transmission power is controlled within 10mW, ensuring complete signal coverage for each floor of a single prison building. The signal is transmitted only within the closed local area network of the prison area and is not connected to the external network. Early warning messages are encrypted using the AES-256-GCM encryption algorithm to prevent interception and tampering during transmission. The duty room's intranet early warning receiving terminal is a standard intranet industrial control device equipped with dedicated early warning software. The terminal is not connected to the internet and only receives encrypted messages pushed by all thin-card terminals in the prison area. The terminal displays early warning information in color-coded pop-up windows according to low, medium, and high risk levels, simultaneously displaying the corresponding prison uniform number and risk trigger time. Supervisory personnel can retrieve the device's historical encrypted early warning logs with one click for rapid on-site response.

[0044] Part Four: Terminal Assembly Process and Full-Process Testing Standards

[0045] After the three-layer module of the thin-card terminal is assembled, it is sent to a vacuum potting equipment to complete the IP67 sealing. After potting and curing, the sewing holes and edges of the shell are rounded to avoid cutting the prison uniform fabric during sewing. On-site embedded installation process: Select the position of the double-layer fabric lining inside the chest of the prison uniform, attach the thin-card terminal to the inside of the fabric, and use cotton thread to sew it in a cross shape through the four sewing holes of the shell. After sewing, the device is completely hidden in the fabric interlayer. The sewing process does not damage the original cut structure of the prison uniform. After soaking and washing in 60℃ laundry detergent 30 times, the device's waterproof, voice recognition, and power supply functions have not been weakened. The equipment undergoes four mandatory full inspection processes before leaving the factory: First, voice recognition function testing, simulating a combined noise environment of broadcasting, mechanical noise, and multi-person conversation, completely traversing the entire sensitive word database to verify recognition accuracy and response time; Second, power supply and battery life testing, running continuously for 72 hours under two conditions: low light and no light, recording voltage and total power consumption every two hours; Third, environmental reliability testing, including 48 hours of high and low temperature cycling from -20℃ to 60℃, salt water spray waterproofing test, 500-cycle fabric bending fatigue test, and electromagnetic interference shielding test; Fourth, long-term stability aging test, running continuously 24 / 7 for 30 days, with the system automatically recording recognition accuracy, equipment power consumption, and fault logs daily. Only after all four tests are passed can the equipment leave the factory and be put into storage.

[0046] Part 5: System Deployment and Expansion / Adaptation Implementation Instructions

[0047] When a complete intelligent voice risk early warning system for prisons is deployed, each inmate's uniform in the cells and workshops is equipped with an independent thin-card terminal. All terminals' radio frequency signals are connected to the same intranet. The duty room only needs to be equipped with one intranet receiving terminal to collect early warning information from the entire area. The dedicated log export device is only equipped to the prison area management person in charge. It requires entering a special permission password and connecting a dedicated data cable to the hidden debugging interface of the terminal to read the encrypted and stored logs. The exported files have an electronic operation watermark, and the entire process can be audited and traced. The entire system does not require setting up a cloud server. All monitoring, early warning, and storage processes are completed in a closed loop locally in the prison. This embodiment is the first-generation basic mass-produced terminal, which can be expanded and adapted according to the needs of different regulatory sites: for outdoor labor scenarios in drug rehabilitation centers, the area of ​​flexible solar film can be increased to improve the efficiency of low-light power generation; for short-term detention scenarios in detention centers, the flash memory storage capacity can be reduced to lower the cost of mass production of hardware; for detention areas for ethnic minority criminals, the LD3320 chip that supports bilingual recognition can be replaced, and dialects and bilingual sensitive word libraries can be locally burned; at the hardware structure level, the shell thickness can be adjusted to 20mm to further reduce the size, and the charging and discharging chip can be optimized to reduce the power consumption of the whole machine. All expansion and modification schemes are based on the three-layer stacked embedded structure and local offline computing core architecture as defined in the claims of this invention, and all fall within the protection scope of this invention.

[0048] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. An embeddable intelligent voice risk early warning system for prisons, characterized in that, This includes several wearable thin-card terminals and intranet early warning receiving terminals; The thin-card terminal can be sewn into the lining of prison uniforms. The terminal integrates a power supply unit, a voice acquisition unit, a local offline voice processing unit, a local storage unit, and a radio frequency early warning transmission unit. The entire voice recognition and risk assessment process is completed offline on the local thin card terminal without accessing the external network. After recognizing high-risk voices, it only pushes graded silent warning signals to the internal network receiving terminal.

2. The embedded intelligent voice risk early warning system for prisons according to claim 1, characterized in that, The thin card terminal adopts a three-layer integrated sealed shell, which consists of a flexible solar power supply layer, a main control and computing layer, and a micro energy storage layer from top to bottom. The power supply unit consists of a dual power supply structure of a flexible amorphous silicon thin film and an ultra-thin soft-pack lithium battery, and is equipped with a dedicated charge and discharge management chip. The overall dimensions of the casing are 75mm×60mm×25mm, and the weight of the whole machine does not exceed 88g. The device's static standby power consumption is ≤10μA, and the continuous working time of the lithium battery in a dark, sealed environment is ≥48h. The shell is sealed with IP67 polyurethane vacuum potting and has sewing holes on the side for direct sewing and fixing to the prison uniform lining. The shell is equipped with a copper foil electromagnetic shielding layer.

3. The embedded intelligent voice risk early warning system for prisons according to claim 1, characterized in that, The local offline speech processing unit adopts a dual collaborative architecture of LD3320 offline recognition chip and STM32 low-power microcontroller, and locally burns the prison's exclusive high-risk and sensitive word library, without the need for cloud training. The voice acquisition unit is equipped with a Φ4mm waterproof and noise-reducing mini microphone, and is paired with a multi-level passive hardware filtering circuit to filter out composite noise in the monitoring area; Keyword recognition response time ≤ 0.5s, and recognition accuracy in noisy prison environments ≥ 95%; After a risky word is detected, the terminal's built-in micro-vibration motor vibrates in stages. The local storage unit uses an SPI flash memory chip to encrypt and retain the warning log, and the original voice is not transmitted to the outside.

4. The embedded intelligent voice risk early warning system for prisons according to claim 1, characterized in that, The radio frequency early warning transmission unit is a 433M low-power radio frequency transmitter board. It only starts transmission briefly when high-risk words are detected, and shuts down and goes into sleep mode at other times. All radio frequency transmission messages are encrypted using the AES-256 encryption algorithm. The signal is only transmitted within the prison's local area network and there is no external network communication channel.

5. The embeddable intelligent voice risk early warning system for prisons according to claim 2, characterized in that, The sewn hole in the housing penetrates the front and rear walls of the housing. When the housing sealant is broken by human intervention and the housing is separated, the computing unit automatically generates an anti-tampering abnormality log and stores it in the local flash memory. The log cannot be manually deleted.

6. The embeddable intelligent voice risk early warning system for prisons according to claim 3, characterized in that, The casing has a hidden micro-debugging interface, which is only enabled when the sensitive word library is burned at the factory and the logs are exported locally. During daily use, the interface is sealed with sealant. A narrow slit sound guide hole is made on the housing corresponding to the microphone position. The width of the guide hole is less than 2mm to prevent the sound pickup channel from being blocked by human intervention.

7. The embeddable intelligent voice risk early warning system for prisons according to claim 3, characterized in that, The arithmetic unit has a built-in hardware reset switch, and the reset contact is enclosed inside the potting housing.

8. The embedded intelligent voice risk early warning system for prisons according to claim 2, characterized in that, The charge and discharge management chip integrates triple protection circuits for overcharge, over-discharge, and short circuit. When the lithium battery voltage is lower than the set threshold, it automatically writes a low battery log and pushes a low battery reminder message to the receiving terminal.

9. The embeddable intelligent voice risk early warning system for prisons according to claim 2, characterized in that, The shell is made of ABS+PC composite engineering plastic. After the filling is completed, all the edges of the shell are rounded and polished. After sewing and fixing, there are no hard protrusions, so it does not affect the daily work, living and washing of clothes of the detainees.

10. The embeddable intelligent voice risk early warning system for prisons according to claim 1, characterized in that, The intranet early warning receiving terminal is deployed in the duty room of the prison area. The receiving terminal displays early warning information in pop-up windows according to low, medium and high risk levels. The system is equipped with a dedicated local log export device. Only authorized personnel can access the terminal's hidden interface to read encrypted and stored logs, and all export operations are automatically recorded for audit purposes.

Images