A wearable cardiopulmonary sound monitoring device with voice privacy protection function and a data processing method thereof

Abstract

This invention discloses a wearable cardiopulmonary sound monitoring device with voice privacy protection function and its data processing method, belonging to the field of wearable medical monitoring and edge privacy computing technology. The device includes a cardiopulmonary sound sensor, an environmental microphone, an MCU, and a data cache unit. The core adopts a cascaded voice detection architecture prioritizing the environmental microphone, performing voice activity detection on the environmental microphone signal first. If a detection is successful, voice contamination is directly determined and the cardiopulmonary sound channel detection is skipped; if not, the cardiopulmonary sound sensor channel is activated for supplementary detection. When voice activity is detected, the MCU discards the entire data from the current acquisition cycle while it is in its local RAM cache, ensuring the data does not leave the device. This invention achieves voice privacy protection from the source of acquisition, completely eliminating the data transmission leakage window. The cascaded strategy balances full-scene voice detection coverage with low power consumption requirements, while supporting time-frequency domain complementary discrimination to reduce false positive rates and a dual-mode privacy strategy to adapt to multiple scenarios.

Description

Technical Field

[0001] This invention relates to the field of wearable medical monitoring equipment technology, and in particular to a wearable cardiopulmonary sound monitoring device with voice privacy protection function and its data processing method. Background Technology

[0002] Heart sounds and lung sounds are the core physiological acoustic evidence for the clinical diagnosis of cardiopulmonary diseases. Traditional mechanical auscultation relies on on-site operation by medical staff and can only achieve short-term, intermittent detection, which cannot meet the needs of long-term dynamic monitoring for patients with chronic cardiopulmonary diseases. With the development of MEMS sensing and flexible electronics technology, patch-type wearable heart and lung sound monitoring devices are gradually being implemented, allowing users to continuously collect heart and lung sound data at home for extended periods, providing an important technical means for chronic disease management and postoperative rehabilitation.

[0003] However, existing wearable cardiopulmonary sound monitoring devices generally pose a security risk of voice privacy leakage: during the long-term wear of the device, it is inevitable that the device will collect the wearer's own voice and the voice conversations of others in the environment. Such voice data is sensitive personal information, and once leaked, it will seriously infringe on the user's privacy.

[0004] To address the aforementioned issues, existing technologies primarily employ a "collect first, process later" privacy protection approach: raw cardiopulmonary sound data is transmitted to a mobile app via Bluetooth, where voice detection and filtering are then performed. This architecture suffers from the following inherent flaws: 1. Privacy protection has a leakage window: Voice detection and filtering occur after the data leaves the device. The original audio is no longer under the device's local control during the Bluetooth transmission stage, and there is a risk of it being intercepted and retained. This is a post-event remedial protection and cannot prevent privacy leakage from the root.

[0005] 2. Single detection path, coverage and power consumption cannot be balanced: Most solutions are based on single-channel detection of the main acquisition sensor, which is difficult to cover both air-conducted environmental speech and bone-conducted self-speech speech at the same time, resulting in a high false negative rate; if dual-channel parallel detection is used, the MCU computing power will be doubled, significantly increasing the power consumption of the device and shortening the battery life, which does not meet the low power consumption requirements of wearable devices.

[0006] 3. High misjudgment rate and significant loss of effective data: Traditional VADs rely solely on time-domain energy and zero-crossing rate characteristics. Non-speech signals such as heart sound peaks and equipment friction and impacts can easily trigger misjudgments, resulting in the erroneous discarding of effective heart and lung sound data.

[0007] 4. The discarding strategy is complex and the hardware implementation cost is high: Some solutions adopt the processing method of accurately extracting voice segments, which requires signal boundary recognition and segment reconstruction. The algorithm is complex and difficult to implement on low-power MCUs.

[0008] 5. Poor scenario adaptability: It only supports a single filtering strategy, which cannot simultaneously meet the strong privacy requirements of personal home scenarios and the clinical data continuity requirements of ward monitoring scenarios; moreover, there is no traceability basis after data is discarded, and users and medical staff cannot assess the integrity of the data.

[0009] In summary, existing technologies lack a wearable cardiopulmonary sound privacy protection solution that can perform voice detection and active discarding at the data collection source (device hardware) while also taking into account detection coverage, low power consumption, and scenario adaptability. Summary of the Invention

[0010] This application is made in view of the above and other ideas.

[0011] The purpose of this invention is to address the shortcomings of existing technologies by providing a wearable cardiopulmonary sound monitoring device with voice privacy protection and its data processing method, solving the following technical problems: 1. Moving voice privacy protection forward to the data collection source, completing detection and discarding locally on the device, eliminating the risk of leakage during data transmission; 2. Achieving full-scene voice detection coverage, while reducing average computing power consumption through a cascaded detection architecture, balancing detection reliability and low device power consumption requirements; 3. Reducing the false judgment rate of voice detection and minimizing the erroneous discarding of valid cardiopulmonary sound data; 4. Providing auditable discard records and switchable scenario-based privacy policies to adapt to the needs of different usage scenarios.

[0012] To achieve the above-mentioned objectives, this invention provides a wearable cardiopulmonary sound monitoring device with voice privacy protection function, comprising a cardiopulmonary sound signal acquisition unit, an environmental sound acquisition unit, a signal processing and control unit, and a data buffer unit. The cardiopulmonary sound signal acquisition unit includes at least one cardiopulmonary sound sensor for picking up heart sounds transmitted through the wearer's chest cavity and / or... Or lung sound vibration signal; the environmental sound acquisition unit includes at least one environmental microphone, facing away from the skin, for picking up airborne environmental sound signals; the signal processing and control unit is a microcontroller (MCU) with a built-in lightweight speech activity detection (VAD) module, configured to execute cascaded speech detection logic: with a preset duration as one acquisition cycle, two signals are simultaneously acquired and temporarily stored in the data buffer unit within the acquisition cycle; speech activity detection is first performed on the output signal of the environmental microphone, and if speech activity is detected, the current acquisition cycle is directly determined to be in a speech pollution state, and speech activity detection on the cardiopulmonary sound sensor signal is skipped; only when the environmental microphone does not detect speech activity will speech activity detection be further performed on the output signal of the cardiopulmonary sound sensor as a supplementary detection; when speech activity is detected in either of the two signals, the microcontroller clears all sensor data temporarily stored in the data buffer unit within the current acquisition cycle from the buffer, does not write it to the persistent storage medium, and does not transmit it wirelessly.

[0013] In this embodiment, the cardiopulmonary sound sensor is a MEMS microphone or a piezoelectric ceramic sensor; the environmental microphone is an electret condenser microphone; the device as a whole is a patch-type flexible form that can be attached to the heart sound auscultation position and / or the lung sound auscultation position.

[0014] In this embodiment, the speech activity detection module implements time-domain speech discrimination based on a dual-threshold decision mechanism of short-time energy and zero-crossing rate. The speech activity detection module also includes a second-order IIR digital bandpass filter bank, including a cardiopulmonary sound segment filter and a speech segment filter, used to extract the frame energy of low-frequency cardiopulmonary sound components and high-frequency speech components respectively, and calculate the ratio of high-frequency to low-frequency energy as a frequency domain discrimination feature. The time-domain discrimination result and the frequency-domain discrimination result are fused together using a configurable fusion strategy to output the final speech judgment result.

[0015] In this embodiment, the fusion strategy includes: a logical OR fusion mode: when the time domain is determined to be speech or the frequency domain is determined to be speech, the final determination is a speech frame, which is suitable for scenarios with high recall requirements; and a logical AND fusion mode: when the time domain and frequency domain are both determined to be speech, the final determination is a speech frame, which is suitable for scenarios with high precision requirements.

[0016] In this embodiment, the microcontroller is further configured to: when data is determined to be discarded, record a discard event metadata entry in local persistent storage; the metadata entry includes a timestamp, trigger channel identifier, discard reason classification and cycle number, but does not contain any audio waveform data; the metadata entry can be reported to the receiving end via Bluetooth and / or wired interface to generate a data integrity statistical report.

[0017] In this embodiment, the microcontroller has a built-in configurable privacy policy register that supports switching between two privacy processing modes: Complete Discard Mode: After detecting voice activity, all data in the cache is cleared, no data is written to any storage medium, and no wireless transmission is performed; Degradation Marking + Transmission Blocking Mode: After detecting voice activity, a voice pollution mark is added to the data and written to local persistent storage, while the wireless transmission of the data segment is blocked, and the marked data can only be exported through a wired interface.

[0018] This embodiment also includes a data privacy protection method for wearable cardiopulmonary sound monitoring, comprising the following steps: S1 Segmented Acquisition: Acquires cardiopulmonary sound sensor signals and environmental microphone signals simultaneously with a preset acquisition period, and temporarily stores the real-time acquired data in the device's local cache. S2 Cascaded Speech Detection: Speech activity detection is performed in priority order within the acquisition period. S21 First-level detection: Perform speech activity detection on the environmental microphone signal; if speech activity is detected, directly determine that there is speech pollution in the current acquisition period and jump to step S3; wherein, the speech activity detection adopts a time-frequency domain complementary discrimination method: extract the short-time energy and zero-crossing rate time-domain features of the signal, and at the same time extract the frame energy of the cardiopulmonary audio segment and the speech audio segment through the IIR digital bandpass filter bank respectively, and calculate the high-frequency to low-frequency energy ratio as the frequency domain feature; fuse the time domain features and the frequency domain features according to the preset fusion strategy, and output the single-frame speech judgment result.

[0019] S22 Second-level detection: If no speech activity is detected in the first-level detection, speech activity detection is further performed on the cardiopulmonary sound sensor signal as a supplementary detection. S3 Data Decision and Processing: At the end of the acquisition cycle, if speech contamination is detected, all sensor data cached for the current cycle is discarded without persistent storage or external transmission. If no speech activity is detected at either of the two detection levels, the data for the current cycle is persistently stored and / or transmitted externally. After determining that data should be discarded, the following steps are also taken: Recording discard event metadata in local storage. The metadata contains only structured information and does not include audio content, for data integrity auditing.

[0020] This embodiment also includes a low-power wearable cardiopulmonary sound monitoring voice privacy protection method, comprising the following steps: Two independent voice activity detection channels are configured on the device-side microcontroller, corresponding to the environmental microphone and the cardiopulmonary sound sensor, respectively. Within each acquisition cycle, the voice activity detection operation of the environmental microphone channel is initiated first. If the environmental microphone channel detects voice activity, it will directly determine that the current acquisition cycle is in a state of voice pollution and terminate the voice activity detection operation of the cardiopulmonary sound sensor channel to reduce the computing power consumption of the microcontroller and the power consumption of the system. The voice activity detection operation of the cardiopulmonary sound sensor channel is only initiated when no voice activity is detected in the ambient microphone channel, as a supplementary detection to cover the bone conduction voice scenario; If cascaded detection determines the presence of voice pollution, all sensor data from the current data collection period will be discarded.

[0021] (1) Privacy and security are fully guaranteed: voice detection and data discarding are completed by the MCU on the device side. The voice data is judged and cleared before leaving the RAM, with zero leakage, meeting the requirements of privacy regulations, and fundamentally overcoming the delayed protection defect of existing technology that "data is processed after leaving the device".

[0022] (2) Comprehensive detection coverage and optimized power consumption: The cascaded detection strategy of dual-channel sensors (environmental microphone priority + heart and lung sound sensor supplement) covers all scenarios of self-speaking and environmental speaking, maintains a low false negative rate, and skips subsequent VAD operations in most cases by prioritizing the detection of the environmental microphone channel with higher hit rate, which significantly reduces the average computing power consumption of MCU and system power consumption, and overcomes the shortcomings of insufficient single-channel detection coverage and high power consumption of parallel full detection in existing technologies.

[0023] (3) Time-frequency domain complementarity reduces misjudgment: Based on the time-domain VAD, the frequency-domain energy ratio discrimination of the IIR filter bank is integrated. The spectral difference between heart and lung sounds (mainly low frequency) and speech (significant high frequency) is further used to further distinguish them. Without increasing the computational burden significantly, the error rate caused by non-speech noise such as bumps and friction is reduced, overcoming the defect of high error rate of pure time-domain discrimination in the existing technology.

[0024] (4) Auditable discard events: The metadata recording mechanism allows users and doctors to understand the data integrity status and provides "auditable privacy protection" - knowing which time periods the data was discarded and why it was discarded, but not knowing what was said, overcoming the shortcomings of existing technologies that have no auditable records.

[0025] The embodiments of this application can achieve other advantageous technical effects not listed one by one. These other technical effects may be partially described below and can be expected and understood by those skilled in the art after reading this application. Attached Figure Description

[0026] The above-described features and advantages, as well as other features and advantages, and the ways in which they are implemented, of these embodiments will become more apparent and the embodiments of this application will be better understood by referring to the following description in conjunction with the accompanying drawings, in which: Figure 1 is a hardware architecture block diagram of the device of the present invention; Figure 2 is a flowchart of the cascaded speech detection and data processing of the present invention; Figure 3 is a schematic diagram of the speech detection path in four typical scenarios of the present invention; Figure 4 is a schematic diagram of the processing rules for speech boundaries across acquisition cycles in this invention. Detailed Implementation

[0027] The details of one or more embodiments of this application will be set forth in the following description of the accompanying drawings and specific embodiments. Other features, objects, and advantages of this application will become clear from these descriptions, drawings, and claims.

[0028] It should be understood that the illustrated and described embodiments are not limited in application to the details of the construction and arrangement of the components set forth in the following description or illustrated in the accompanying drawings. The illustrated embodiments may be other embodiments and can be implemented or performed in various ways. The examples are provided by way of explanation rather than limitation of the disclosed embodiments. In fact, it will be apparent to those skilled in the art that various modifications and variations can be made to the embodiments of this application without departing from the scope or spirit of this disclosure. For example, features illustrated or described as part of one embodiment may be used with another embodiment to still produce another embodiment. Therefore, this disclosure covers such modifications and variations that fall within the scope of the appended claims and their equivalents.

[0029] Similarly, it is understood that the phrases and terms used in this document are for descriptive purposes and should not be considered restrictive. The use of “including,” “contains,” or “has,” and their variations, in this document is intended to include, in an open-ended manner, the items listed thereafter, their equivalents, and any additional items.

[0030] The present application will now be described in more detail with reference to various embodiments and examples of several aspects thereof.

[0031] One of the objectives of the embodiments described below is to address the aforementioned deficiencies, as well as other problems.

[0032] Example 1: MEMS Microphone-Type Cardiopulmonary Sound Monitoring Device 1.1 Hardware Configuration This example is a patch-type wearable cardiopulmonary sound monitoring device, and the hardware configuration is as follows: Cardiopulmonary sound sensor: A MEMS microphone (such as Knowles SPH0641LU4H-1 or similar device) is attached to a flexible PCB with the microphone aperture facing the skin side. It contacts the skin through a medical-grade silicone coupling layer to pick up the acoustic signals of heart and lung sounds transmitted through the chest wall.

[0033] Ambient sound acquisition unit: It adopts an electret condenser microphone, which is installed on the outer side of the patch device (the side away from the skin) to pick up sound signals in the environment.

[0034] Microcontroller (MCU): Uses low-power ARM Cortex-M series chips (such as STM32L4 series), with built-in ADC and DSP instruction sets, main frequency not less than 48MHz, and supports real-time running of voice activity detection (VAD) algorithm.

[0035] Wireless transmission unit: Uses a Bluetooth module with BLE 5.0 or higher, supporting low-power wireless data transmission.

[0036] 1.2 VAD Algorithm Implementation In this embodiment, the VAD algorithm adopts a dual-threshold decision method based on short-time energy and zero-crossing rate. The specific steps are as follows: Frame segmentation: The audio signal is segmented into 30ms frames.

[0037] Feature extraction: Calculate the short-time energy E and zero-crossing rate ZCR for each frame of signal.

[0038] Dual threshold decision: A high energy threshold E_high, a low energy threshold E_low, and a zero-crossing rate threshold ZCR_th are preset; when E > E_high, or E > E_low and ZCR > ZCR_th, the current frame is determined to be a speech frame.

[0039] Temporal smoothing: A hangover mechanism is adopted, which means that a speech is determined to end only after N consecutive non-speech frames, avoiding misjudgment caused by short pauses in speech.

[0040] The two signals (MEMS heart and lung sound channel and environmental microphone channel) are detected in a cascaded manner rather than in parallel: the MCU prioritizes performing VAD calculation on the environmental microphone channel; if the environmental microphone channel detects speech, it directly determines that the current acquisition period "contains speech" and marks it as "contaminated", skipping the subsequent heart and lung sound channel VAD calculation; only when the environmental microphone channel does not detect speech will it continue to perform VAD on the heart and lung sound channel, and the results of the two detections are logically ORed as the final speech determination result for the current period. 1.3 Workflow The complete workflow of the device in this embodiment is as follows: Users attach the patch to the chest auscultation position for heart sounds or lung sounds.

[0041] After the device is started, it enters the cyclic acquisition mode, with each acquisition cycle lasting 30 seconds.

[0042] During a single acquisition cycle, the MEMS heart and lung sound signals and the ambient microphone signal are synchronously converted by ADC at a sampling rate of 8kHz, and the acquired data is temporarily stored in RAM cache.

[0043] While data is being collected, the MCU runs the VAD algorithm according to the cascaded detection process: first, it detects the environmental microphone channel; if no speech is detected, it then detects the cardiopulmonary sound channel.

[0044] At the end of the acquisition cycle, the MCU performs the corresponding processing based on the VAD result: If the VAD result is "no voice": write the heart and lung sound data for that period in the RAM cache to the Flash storage, and send it to the mobile APP via BLE.

[0045] If the VAD result is "Contains Voice": Clear all data (including cardiopulmonary sound data and ambient microphone data) in the RAM cache for this period, do not write to Flash storage, and do not transmit via Bluetooth; optionally, record a "Voice Drop Event" marker in the data log, which only records the timestamp and event type, without any audio content.

[0046] Proceed to the next data collection cycle and repeat steps 3 through 5.

[0047] Example 2: Piezoelectric Sensor Type Cardiopulmonary Sound Monitoring Device 2.1 Hardware Configuration This example is a piezoelectric sensor type wearable cardiopulmonary sound monitoring device, with the following hardware configuration: Cardiopulmonary sound sensor: A PZT piezoelectric ceramic sensor is attached to a flexible substrate and closely adheres to the skin through a medical-grade adhesive layer to pick up the vibration signal of cardiopulmonary sounds from the chest wall; the output signal of the piezoelectric sensor is converted into a voltage signal by a charge amplifier and then sent to the ADC channel of the MCU.

[0048] The ambient microphone, MCU, Bluetooth module, and other hardware configurations are exactly the same as in Example 1.

[0049] 2.2 VAD Algorithm and Workflow In this embodiment, the VAD algorithm logic and cascaded detection sequence are completely consistent with Embodiment 1; the two detection signals are the environmental microphone channel (first-level detection) and the piezoelectric heart and lung sound channel (second-level detection), and the overall workflow is the same as in Embodiment 1. Embodiment 3: Implementation of Contextualized Privacy Strategy Based on the above device, this embodiment implements a configurable and switchable privacy processing strategy, including a complete discard mode and a downgraded marking + blocking transmission mode. 3.1 Mode Configuration Mechanism The MCU internally maintains a privacy policy configuration register, which can be stored in the configuration area of ​​Flash or EEPROM. Its value is defined as follows: MODE_FULL_DROP = 0x00: Fully discard mode, which is the factory default value; MODE_MARKED_LOCAL = 0x01: Degradation flag + Block transmission mode.

[0050] The mobile app can send configuration commands to the device via Bluetooth, for example, by writing custom GATT feature values. After receiving the configuration command, the MCU updates the value of the register, which takes effect at the start of the next acquisition cycle. Mode switching does not require restarting the device. 3.2 Complete Discard Mode (MODE_FULL_DROP) When the device is configured in complete discard mode, the MCU's processing logic is the same as in Embodiments 1 and 2: After the VAD detects voice, it clears the corresponding data in the RAM cache, does not write to Flash storage, does not perform Bluetooth transmission, and records discard event metadata.

[0051] This mode is suitable for personal home use scenarios, providing maximum privacy protection. 3.3 Degradation Marking + Blocking Transmission Mode (MODE_MARKED_LOCAL) When the device is configured in Degradation Marking + Blocking Transmission Mode, after the VAD detects voice, the MCU performs the following processing: Do not clear data from the RAM cache; Add a contamination marker field to the header of the data segment. This marker content includes information such as contamination identifier, trigger channel, trigger reason, and timestamp. Then, write the tagged cardiopulmonary sound data along with the marker field into the Flash storage. Block the Bluetooth transmission of data for this collection period, that is, skip the data for this period in the Bluetooth transmission queue and do not transmit it out of the device through the wireless channel; Record the metadata of the discard event, and categorize the reason for discarding as "degradation flag"; Data marked as contaminated can only be exported via a wired USB connection. When importing data from the USB device, the receiving app identifies the contaminated data and displays the corresponding data segments and clean data segments separately for medical staff to determine whether they need to refer to the data.

[0052] This mode is suitable for public monitoring scenarios such as ward monitoring and nursing home care: Patients may not be able to control when they speak (such as during ward rounds), and completely discarding this information can lead to the loss of key cardiopulmonary sound data. By blocking Bluetooth transmission, the basic requirements for privacy protection are met by ensuring that voice messages do not leak out through wireless channels. Healthcare workers can export the data via USB and analyze it offline, combining it with contamination markers to determine data usability; Simultaneously, metadata about the discard event is recorded, which can be used for data integrity auditing.

[0053] Example 4: Enhanced VAD with Fusion of Time-Frequency Domain Features This example optimizes the VAD algorithm based on the above examples, achieving complementary time-frequency domain speech discrimination on the MCU with extremely low computational cost. 4.1 IIR Bandpass Filter Bank Design Two second-order IIR digital bandpass filters are implemented on the MCU: Low-pass / band-pass filter (cardiopulmonary sound band): The cutoff frequency is set at about 500Hz to extract signal components in the 20-100Hz frequency band, which is the main distribution area of ​​heart sounds (S1 / S2 with main energy in 50-150Hz) and lung sounds (main energy in 100-1000Hz).

[0054] Bandpass filter (speech band): The passband is set at approximately 300-3400Hz to extract the main energy distribution area of ​​the speech signal.

[0055] The difference equation for a second-order IIR is: \(y[n] = b_0·x[n] + b_1·x[n-1] + b_2·x[n-2] - a_1·y[n-1] - a_2·y[n-2]\) Each sampling point requires only 5 multiplications and 4 additions; calculated at an 8kHz sampling rate, the two filters perform approximately 80,000 multiplication-addition operations per second, which is a very low percentage (<0.2%) for a 48MHz ARM Cortex-M4 MCU. 4.2 Frequency Domain Energy Ratio Determination For the signals output from the two filters, the short-time energy is calculated separately for each VAD frame (30ms, 240 sampling points): E_low: Frame energy of the cardiopulmonary audio segment channel; E_high: Frame energy of the audio segment channel.

[0056] Calculate the energy ratio \(R = E_{high} / (E_{low} + \varepsilon)\), where \(\varepsilon\) is a small constant to prevent division by zero.

[0057] When R > R_threshold (recommended value 1.5), the current frame is determined to match speech in the frequency domain features. 4.3 Time Domain and Frequency Domain Fusion Decision: The fusion strategy can be configured as needed to be one of the following two: Logical OR fusion (high recall priority): When the time domain VAD (energy + zero crossing rate) determines that it is a speech frame, or the frequency domain energy ratio determines that it is a speech frame, the current frame is marked as a speech frame; suitable for scenarios with extremely low requirements for false negative rate.

[0058] Logic and Fusion (High Precision Priority): The current frame is marked as a speech frame only when both the time-domain VAD and the frequency-domain energy ratio are determined to be speech frames; suitable for scenarios where it is desirable to reduce erroneous data loss and increase the proportion of effective data acquisition.

[0059] In the complete discard mode, it is recommended to use logical OR fusion, erring on the side of discarding more speech rather than missing any; in the degraded marking mode, logical AND fusion can be used to reduce unnecessary marking. It should be noted that, in this document, relational terms such as "first" and "second" are used only 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 process, method, article, or apparatus.

Claims

1. A wearable cardiopulmonary sound monitoring device with voice privacy protection function, comprising a cardiopulmonary sound signal acquisition unit, an environmental sound acquisition unit, a signal processing and control unit, and a data buffer unit, characterized in that: The cardiopulmonary sound signal acquisition unit includes at least one cardiopulmonary sound sensor for picking up the heart sound and / or lung sound vibration signals transmitted through the wearer's chest cavity; The ambient sound acquisition unit includes at least one ambient microphone, which is positioned facing away from the skin to pick up ambient sound signals conducted through the air. The signal processing and control unit is a microcontroller (MCU) with a built-in lightweight voice activity detection (VAD) module, configured to execute cascaded voice detection logic: A preset duration is used as one acquisition cycle. Within the acquisition cycle, two signals are acquired simultaneously and temporarily stored in the data buffer unit. Prioritize voice activity detection on the output signal of the environmental microphone. If voice activity is detected, directly determine that the current acquisition period is in a state of voice pollution and skip the voice activity detection on the cardiopulmonary sound sensor signal. Speech activity detection is performed on the output signal of the cardiopulmonary sound sensor only as a supplementary detection if no speech activity is detected by the ambient microphone. When either of the two signal paths detects voice activity, the microcontroller clears all sensor data temporarily stored in the data cache unit during the current acquisition cycle from the cache, without writing it to the persistent storage medium or transmitting it wirelessly.

2. The wearable cardiopulmonary sound monitoring device according to claim 1, characterized in that, The cardiopulmonary sound sensor is a MEMS microphone or a piezoelectric ceramic sensor; the environmental microphone is an electret condenser microphone; the device as a whole is a patch-type flexible form that can be attached to the heart sound auscultation position and / or the lung sound auscultation position.

3. The wearable cardiopulmonary sound monitoring device according to claim 1, characterized in that, The speech activity detection module achieves time-domain speech discrimination based on a dual threshold decision mechanism of short-time energy and zero-crossing rate; The speech activity detection module also includes a second-order IIR digital bandpass filter bank, including a cardiopulmonary audio segment filter and a speech audio segment filter, used to extract the frame energy of low-frequency cardiopulmonary sound components and high-frequency speech components respectively, and calculate the ratio of high-frequency to low-frequency energy as a frequency domain discrimination feature; The time-domain discrimination result and the frequency-domain discrimination result are combined using a configurable fusion strategy to output the final speech determination result.

4. The wearable cardiopulmonary sound monitoring device according to claim 3, characterized in that, The fusion strategy includes: Logical OR fusion mode: When the time domain is determined to be speech or the frequency domain is determined to be speech, the final determination is a speech frame, which is suitable for scenarios with high recall requirements; And: Logic and Fusion Mode: When both the time domain and frequency domain are determined to be speech, it is ultimately determined to be a speech frame, which is suitable for high-precision scenarios.

5. The wearable cardiopulmonary sound monitoring device according to claim 1, characterized in that, The microcontroller is also configured to: when it is determined that data has been discarded, record a discard event metadata entry in local persistent storage; The metadata entries include timestamps, trigger channel identifiers, drop reason classifications, and cycle numbers, but do not contain any audio waveform data. The metadata entries can be reported to the receiving end via Bluetooth and / or wired interface to generate a data integrity statistical report.

6. The wearable cardiopulmonary sound monitoring device according to claim 1, characterized in that, The microcontroller has a built-in configurable privacy policy register that supports switching between two privacy processing modes: Complete Discard Mode: Upon detecting voice activity, clear all data in the cache, do not write to any storage medium, and do not perform wireless transmission; Degradation tagging + transmission blocking mode: After detecting voice activity, add a voice pollution tag to the data and write it to local persistent storage, while blocking the wireless transmission of the data segment. The tagged data can only be exported through the wired interface.

7. A method for protecting data privacy in wearable cardiopulmonary sound monitoring, characterized in that, Includes the following steps: S1 Segmented Acquisition: Acquires cardiopulmonary sound sensor signals and environmental microphone signals simultaneously with a preset acquisition period, and temporarily stores the real-time acquired data in the device's local cache. S2 Cascaded Speech Detection: Speech activity detection is performed in priority order within the acquisition period. S21 First-level detection: Perform voice activity detection on the environmental microphone signal; if voice activity is detected, directly determine that there is voice pollution in the current acquisition period, and jump to step S3; S22 Second-level detection: If no speech activity is detected in the first-level detection, speech activity detection is further performed on the cardiopulmonary sound sensor signal as a supplementary detection. S3 Data Judgment and Processing: At the end of the acquisition cycle, if speech pollution is detected, all sensor data cached in the current cycle will be discarded and will not be persistently stored or transmitted. If no speech activity is detected in either of the two levels of detection, the data in the current cycle will be persistently stored and / or transmitted.

8. The data privacy protection method according to claim 7, characterized in that, In step S2, speech activity detection uses a time-frequency domain complementary discrimination method: Short-time energy and zero-crossing rate time-domain features of the signal are extracted. At the same time, the frame energy of the cardiopulmonary audio segment and the speech audio segment are extracted by IIR digital bandpass filter bank, and the ratio of high-frequency to low-frequency energy is calculated as the frequency domain feature. The time-domain features and frequency-domain features are fused according to a preset fusion strategy to output a single-frame speech determination result.

9. The data privacy protection method according to claim 7, characterized in that, After determining that data should be discarded in step S3, the process also includes: Record the metadata of the discard event in local storage. The metadata contains only structured information and does not contain audio content, and is used for data integrity auditing.

10. A low-power wearable method for protecting voice privacy in cardiopulmonary sound monitoring, characterized in that, Includes the following steps: Two independent voice activity detection channels are configured on the device-side microcontroller, corresponding to the environmental microphone and the cardiopulmonary sound sensor, respectively. Within each acquisition cycle, the voice activity detection operation of the environmental microphone channel is initiated first. If the environmental microphone channel detects voice activity, it will directly determine that the current acquisition cycle is in a state of voice pollution and terminate the voice activity detection operation of the cardiopulmonary sound sensor channel to reduce the computing power consumption of the microcontroller and the power consumption of the system. The voice activity detection operation of the cardiopulmonary sound sensor channel is only initiated when no voice activity is detected in the ambient microphone channel, as a supplementary detection to cover the bone conduction voice scenario; If cascaded detection determines that there is voice pollution, all sensor data for the current acquisition cycle will be discarded.

Images