CN104581516A - Dual-microphone noise reduction method and device for medical acoustic signals - Google Patents

Abstract

The invention discloses a dual-microphone denoising method and device applied to medical acoustic signals such as cardiopulmonary sounds and fetal heart sounds. The dual-microphone denoising method for medical acoustic signals includes: acquiring noisy medical acoustic signal data and environmental noise data, respectively processing the medical acoustic signal data and the environmental noise data to obtain time-frequency units; calculating eigenvalues for the time-frequency units; generating masking values according to the relationship between the eigenvalues and the masking threshold, and marking In the time-frequency unit, the medical acoustic signal-based part; according to the generated masking value, the time-frequency unit of the noisy medical acoustic signal data is masked, the medical acoustic signal-based part is retained, and the masked data Perform reconstruction to obtain denoised results. The noise elimination method proposed by the invention adopts dual microphones to separately collect noisy medical sound signals and environmental noises, and effectively separates the medical sound signals from the environmental noises.

Description

A dual-microphone denoising method and device for medical acoustic signals

technical field

The invention relates to the field of digital signal processing of medical equipment, in particular to a dual-microphone noise elimination method and device for medical sound signals.

Background technique

Cardiovascular diseases and respiratory diseases are common and frequently-occurring diseases that endanger human health. With the enhancement of people's awareness of health care, there is a need for cardiopulmonary health monitoring.

Common diagnostic methods for heart disease include electrocardiography, echocardiography, and auscultation. The electrocardiogram reflects the electrical activity characteristics of the heart and can be used to identify arrhythmias and atrial and ventricular functional defects. However, the electrocardiogram is not accurately related to the contraction and relaxation of the ventricle, and it cannot be completely concluded that the heart is functioning normally from a normal electrocardiogram. Echocardiography produces high-frequency sound pulses and uses the echoes to localize and study the motion and planes of various heart structures. Ultrasonic waves have certain thermal and acoustic effects on human tissues, which may pose potential hazards to human health. Examination methods for respiratory diseases include chest imaging, bronchoscopy, and auscultation. Chest imaging examination can well show the condition of lung lesions, and the bronchoscope inserted into the bronchi can directly peek into the trachea, both of which have important applications in clinical treatment. As a simple method, auscultation is non-invasive and has good comfort, and it has universal applicability in clinical medicine and health care.

Heart and lung sound detection is a passive monitoring method based on auscultation, which records heart and lung sounds, and is easy to operate and non-invasive. More preferably, the heart sound detection can be used to diagnose the murmur in the heart sound in the early stage of cardiovascular disease and before the electrocardiogram shows abnormalities, and take corresponding treatment measures as soon as possible; the lung sound monitoring can record the apnea phenomenon during sleep, the bronchial Spasm, and adjuvant therapy.

Heart disease is easily induced by emotional stress, physical activity, smoking, drinking, etc., and the heart sound signal is a non-stationary random process, which only has stable statistical properties in a short period of time, which makes it impossible to ensure accurate diagnosis during hospital examinations disease. Through the portable monitoring system, the heart sound signal can be collected anytime and anywhere, which is convenient for data recording and reproduction, and is of great benefit to health care and assistance in medical diagnosis.

The lung sound signal also shows a non-stationary random process. In the hospital, lung sound auscultation is used to monitor the respiratory status of patients in the intensive care unit, monitor the changes in the lungs of patients after anesthesia in the operating room, etc.; in daily applications, it can be used to record sleep time. apnea, bronchospasm in asthmatic patients. Therefore, using a portable monitoring system to collect lung sound signals anytime and anywhere is of great help to medical treatment and health care.

Another application background is fetal heart sound detection. During pregnancy, especially in the middle and late stages of pregnancy, various indicators of the fetus can be monitored to understand the health status of the fetus in the womb and detect fetal abnormalities early. The main purpose of monitoring the fetal heart rate is to obtain the real-time heart rate of the fetus. Whether the fetal heart rate is normal is an important indicator for judging whether the fetus is hypoxic in the mother's body. Therefore, fetal heart rate monitoring is an important item in pregnancy inspection. Doppler ultrasound monitoring is commonly used for fetal heart sound monitoring. A probe that can emit ultrasound is used to transmit ultrasound to the fetal heart. heart rate. Doppler ultrasonic monitoring can actively emit ultrasonic waves for detection, which can effectively detect the weak heartbeat of the fetus, and the analysis accuracy is high. But ultrasound may be potentially harmful to fetal health, and ultrasound dose needs to be strictly tested and monitored.

The passive monitoring method based on auscultation is not harmful to the safety of the fetus. A simple and non-invasive monitoring system that uses acoustic sensors to collect sound signals from the surface of the mother's abdomen. Pregnant women and their family members can use the portable monitoring system to collect fetal heart sound data at any time and record fetal heart sound for reproduction and medical assistance.

In a typical medical acoustic signal monitoring system, while the sound acquisition device monitors the medical acoustic signal, it also collects the surrounding environmental noise. Environmental noise seriously affects the quality of the collected signals, which is not conducive to the later physiological and pathological diagnosis.

In clinical practice, auscultation requires long-term training and accumulation of clinical experience for medical staff to distinguish noise from useful medical acoustic signals. The frequency of medical sound signals is relatively low, such as: the frequency range of heart sound is below 1kHz, the frequency range of lung sound is 60-1000Hz, the frequency of fetal heart sound is mainly within 170Hz, and the best hearing range of people is 1k-2kHz. It makes the low-frequency medical sound signal difficult to hear. At the same time, medical acoustic signals are weak and easily covered by noise. These all lead to auscultation is easily affected by the noise of the surrounding environment, and a quiet auscultation environment is required. In the medical acoustic signal monitoring system, after the medical acoustic signal is amplified, collected, and stored, the digital signal processing method can be used to eliminate the noise of the collected signal, greatly reducing the requirements for the auscultation environment, and it is convenient to detect the health status anytime and anywhere .

In the denoising methods of medical acoustic signals, common methods include directly using low-pass filter to filter out-of-band noise, spectral subtraction and adaptive filter. A low-pass filter is the simplest, but cannot eliminate noise in the passband. The spectral subtraction uses the short-time Fourier transform, and the algorithm is more complicated than the low-pass filter. Using the spectral subtraction will produce strong musical noise. Adaptive filter is the most widely used, but there is a compromise between convergence time and accuracy.

Contents of the invention

(1) Technical problems to be solved

The technical problem to be solved by the present invention is: aiming at the interference of environmental noise in the medical acoustic signal monitoring system, how to realize a noise elimination method and device for effectively eliminating environmental noise.

(2) Technical solution

In order to solve the above problems, the present invention provides a dual-microphone denoising method for medical acoustic signals, comprising: acquiring noisy medical acoustic signal data and environmental noise data, and performing Processing to obtain a time-frequency unit; calculating the eigenvalue for the time-frequency unit; generating a masking value according to the relationship between the eigenvalue and the masking threshold, marking the part where the medical acoustic signal is the main part of the time-frequency unit; according to the generated The masking value masks the time-frequency unit of the noise-bearing acoustic signal data, retains the medical acoustic signal-based part, reconstructs the masked data, and obtains a denoising result.

Preferably, the acoustic signal data with noise is obtained by an acoustic sensor placed on the surface of the human body.

Preferably, the acoustic signal data and the environmental noise data are divided into subbands, and time frames are divided into time frames to obtain the time-frequency units.

Preferably, it also includes: according to the auditory characteristics of the human ear, using a filter bank that divides subbands unevenly, divides the frequency band into exponentially distributed subbands, and divides the low frequency carefully while the high frequency is relatively rough, which can help distinguish low frequency sound;

Preferably, the method further includes: according to the auditory characteristic of the human ear and the need of auscultation, filtering the obtained dual-microphone data, and enhancing the data within a predetermined frequency range.

Preferably, the calculated eigenvalue is an energy ratio of the time-frequency unit of the dual-microphone data.

Preferably, the method further includes: smoothing the masking value, so that the denoising result maintains continuity in the time domain.

Preferably, the masked data is time-superimposed on each time frame, and each sub-band is superimposed to obtain a denoising result.

The present invention also provides a dual-microphone denoising device for acoustic signals, including: a peripheral analysis module for acquiring noisy medical acoustic signal data and environmental noise data, and analyzing the medical acoustic signal data and the environmental noise data respectively Perform processing to obtain a time-frequency unit; a feature extraction module is used to calculate a feature value for the time-frequency unit; a masking separation module is used to generate a masking value according to the relationship between the feature value and a masking threshold, and mark the time-frequency The part where the acoustic signal is the main part in the unit; the signal reconstruction module is used to mask the time-frequency unit of the noisy medical acoustic signal data according to the generated masking value, retain the part where the medical acoustic signal is the main part, and perform masking The data is reconstructed to obtain denoising results.

Preferably, the peripheral analysis module further includes a signal enhancement module, configured to filter the obtained dual-microphone data and enhance data within a predetermined frequency range according to the auditory characteristics of the human ear and auscultation needs.

Preferably, the masking separation module further includes a smoothing module for smoothing the masking value so that the denoising result maintains continuity in the time domain.

(3) Beneficial effects

The noise elimination method proposed by the invention adopts dual microphones to separately collect noisy medical sound signals and environmental noises, and effectively separates the medical sound signals from the environmental noises. Using this method, the subject can monitor adult cardiopulmonary sound and fetal heart sound anytime and anywhere, and denoise the obtained data to provide reference for medical care.

Description of drawings

FIG. 1 is a schematic flow chart of a dual-microphone noise cancellation method according to an embodiment of the present invention;

Fig. 2 is a schematic structural diagram of a dual microphone according to an embodiment of the present invention;

Fig. 3 is a schematic flow chart of an optimized dual-microphone denoising method according to an embodiment of the present invention;

4 is a schematic diagram of sub-band data after sub-band division according to an embodiment of the present invention;

Fig. 5 is a schematic diagram of a calculation result of an eigenvalue according to an embodiment of the present invention;

Fig. 6 is a schematic diagram of the effects before and after smoothing of masking values according to an embodiment of the present invention.

Detailed ways

The specific implementation manners of the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

As shown in FIG. 1 , the dual-microphone denoising method provided by the present invention includes four parts: peripheral analysis, feature extraction, masking separation and signal reconstruction.

Wherein, the peripheral analysis obtains noisy medical acoustic signal data and environmental noise data, and processes the two sets of data respectively, and the obtained result is a time-frequency unit, which is used for feature extraction and signal reconstruction;

The feature extraction is to calculate a feature value for the time-frequency unit for masking separation;

In the masking separation, a masking value is generated according to the relationship between the feature value and the masking threshold, and a part of the time-frequency unit that is dominated by medical acoustic signals is marked for signal reconstruction;

The signal reconstruction includes masking the time-frequency unit of the noisy medical acoustic signal data according to the generated masking value, retaining the main part of the medical acoustic signal, and reconstructing the masked data to obtain a denoising result .

In order to ensure accurate acquisition of the noisy medical sound data and the environmental noise data, it is required that the noisy medical sound signal data be obtained by an acoustic sensor 1 placed close to the body surface of the human body: when placed close to the chest wall of an adult, the acquired Noisy adult cardiopulmonary sound data; when placed close to the abdomen of a pregnant woman, noisy fetal heart sound data is obtained; environmental noise data is obtained by another acoustic sensor 2 that is not close to the human body. In this embodiment, both the acoustic sensor 1 that is placed close to the body surface of the human body and the acoustic sensor 2 that is not close to the human body can be microphones.

Wherein, during the peripheral analysis, the data is divided into subbands, and time frames are divided into time frames to obtain the time-frequency units.

Wherein, the peripheral analysis further includes a signal enhancement unit, which filters the obtained dual-microphone data and enhances the data in a specific frequency range according to the auditory characteristics of the human ear and the needs of auscultation.

Wherein, in the feature extraction, the calculated feature value is the energy ratio of the two-microphone data divided into time-frequency units.

Wherein, the masking separation further includes a smoothing unit, which is used to smooth the masking value to ensure the continuity of the denoising result in the time domain.

Wherein, in the signal reconstruction, the masked data is superimposed on each time frame according to time, and each subband is superimposed on frequency to obtain a denoising result.

The present invention can be applied to heart sound auscultation according to the dual-microphone structure shown in FIG. 2 to obtain noisy heart sound data and environmental noise data.

The present invention obtains the denoising result with the following denoising method, as shown in FIG. 3 , which is a flowchart of an optimized dual-microphone denoising method.

S1. According to the auditory characteristics of the human ear and the needs of auscultation, the noisy heart sound data and the environmental noise data are respectively enhanced;

Since the human ear is more sensitive to sounds from 1kHz to 2kHz, it is difficult to hear heart sounds below 1kHz. The equal loudness curve can be used to obtain the gain compensation value corresponding to a certain frequency. Using the gain compensation value to enhance the data can benefit hearing low-frequency sounds;

S2. Filter the enhanced signal with a filter bank respectively, and divide sub-bands;

The filter bank that divides sub-band can choose the mode of uniform division, also can select the mode of uneven division, adopt the mode of uneven division of sub-band in the embodiment;

Since the basilar membrane of the cochlea of the human ear has the characteristics similar to a spectrum analyzer, sounds of different frequencies cause vibrations at different positions of the basilar membrane, and the distribution of frequencies along the basilar membrane is uneven and exponentially distributed, making the human ear sensitive to Different frequencies are sensitive to different sounds. Using a filter bank that divides the subbands unevenly, divides the frequency band into exponentially distributed subbands, and divides the low frequency carefully while the high frequency is relatively rough, which can help distinguish low frequency sounds;

Fig. 4 is a schematic diagram of filtering results of one of the filters;

S3. Perform overlapping and framing on the data in each subband to obtain time-frequency units;

Overlapping framing makes the transition between frames smooth and maintains continuity in the time domain;

S4. Calculate the energy of each time-frequency unit, and calculate the energy ratio of the time-frequency unit corresponding to the nth subband t time frame of the noisy heart sound signal and the environmental noise one by one, to obtain the characteristic value;

Fig. 5 is a schematic diagram of the eigenvalue calculation results of one of the subbands;

S5. Determine the size of the masking value by comparing the eigenvalue with the masking threshold. The masking threshold is a segmental function of the subbands. Selecting multi-value masking can obtain a good denoising effect;

S6. Perform smoothing processing on the masked value. Since the masking value is discrete, adjacent time frames may correspond to masking values of different sizes. After being weighted by the masking value, the sudden masking value will make the data of adjacent time frames discontinuous, thus introducing high-frequency noise;

The smoothing process improves the mutation of the masking value and improves the continuity in time. Figure 6 is a schematic diagram of the effect before and after the smoothing of the masking value;

S7. The time-frequency unit corresponding to the noisy heart sound data is weighted by the masking value, and the time frames of each sub-band are frame-stacked respectively;

Due to the use of the overlapping frame division method, it is necessary to use a window function to process each frame of data during frame stacking, and then perform frame stacking on each sub-band separately to obtain frame-stacked data with a smooth transition;

S8. The frame-stacked data of each subband is filtered by a filter bank to obtain data of each subband that can be directly superimposed in the time domain;

Since the sub-bands are divided unevenly by filters, there is a time delay between the data of each sub-band, which is manifested as a phase difference in the frequency domain. The frame-stacked data of each sub-band needs to be filtered by the filter bank to eliminate the time delay between each sub-band, eliminate the phase difference, and obtain the data of each sub-band that can be directly superimposed in the time domain;

S9. Superimpose each sub-band data according to time to obtain a denoising result.

The present invention also provides a dual-microphone denoising device, which includes four modules of peripheral analysis, feature extraction, masking separation and signal reconstruction.

Wherein, the peripheral analysis module acquires noisy medical acoustic signal data and environmental noise data, and processes the two sets of data respectively, and the obtained result is a time-frequency unit, which is used for feature extraction and signal reconstruction;

The feature extraction module calculates feature values for the time-frequency unit for masking separation;

The masking separation module generates a masking value according to the relationship between the feature value and the masking threshold, and marks the part of the time-frequency unit where the number of medical acoustic signals is the main number, and is used for signal reconstruction;

The signal reconstruction module masks the time-frequency unit of the noisy medical acoustic signal data according to the generated masking value, retains the main part of the medical acoustic signal, reconstructs the masked data, and obtains the denoising result.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, some improvements and replacements can also be made, these improvements and replacements It should also be regarded as the protection scope of the present invention.

Claims (10)

1. A dual-microphone denoising method for medical acoustic signals, comprising: acquiring noisy medical acoustic signal data and environmental noise data, and processing the medical acoustic signal data and the environmental noise data respectively to obtain a time-frequency unit; calculating eigenvalues for the time-frequency unit; Generate a masking value according to the relationship between the feature value and the masking threshold, and mark the part of the time-frequency unit where the medical acoustic signal is the main part; According to the generated masking value, the time-frequency unit of the noisy medical acoustic signal data is masked, the part where the medical acoustic signal is the main part is reserved, and the masked data is reconstructed to obtain a denoising result. 2. The method according to claim 1, wherein the noisy medical acoustic signal data is obtained by an acoustic sensor placed on the surface of a human body. 3. The method according to claim 1 or 2, wherein the medical acoustic signal data and the environmental noise data are divided into subbands, and time frames are divided by time to obtain the time-frequency unit. 4. The method of claim 3, further comprising: According to the auditory characteristics of the human ear and the needs of auscultation, the obtained dual-microphone data is filtered, and the data within a predetermined frequency range is enhanced. 5. The method according to claim 1 or 2, wherein the calculated eigenvalue is an energy ratio of the time-frequency unit of the dual-microphone data. 6. The method according to claim 1 or 2, further comprising: The masking value is smoothed so that the denoising result maintains continuity in the time domain. 7. The method according to claim 1 or 2, characterized in that the masked data is superimposed on each time frame according to time, and each subband is superimposed to obtain a denoising result. 8. A dual-microphone denoising device for medical acoustic signals, comprising: A peripheral analysis module, configured to acquire noisy medical acoustic signal data and environmental noise data, and separately process the medical acoustic signal data and the environmental noise data to obtain a time-frequency unit; A feature extraction module, configured to calculate feature values for the time-frequency unit; A masking separation module, configured to generate a masking value according to the relationship between the feature value and the masking threshold, and mark the part where the acoustic signal is dominant in the time-frequency unit; The signal reconstruction module is used to mask the time-frequency unit of the noisy medical acoustic signal data according to the generated masking value, retain the main part of the medical acoustic signal, and reconstruct the masked data to obtain denoising result. 9. The device according to claim 8, wherein the peripheral analysis module further includes a signal enhancement module, which is used to filter the obtained dual-microphone data according to the auditory characteristics of the human ear and the needs of auscultation, and to filter the predetermined frequency The data in the range is enhanced. 10. The device according to claim 8 or 9, wherein the masking separation module further comprises a smoothing module, configured to perform smoothing processing on the masking value, so that the denoising result maintains continuity in the time domain.