JP2024074099A - Biological sound sensor system - Google Patents

Abstract

To provide a biological sound sensor that can be mounted on the head of a subject easily capable of simultaneously measuring a blood vessel sound and a respiratory sound without giving a feeling of discomfort to a subject and enhancing noise reduction capability.SOLUTION: A biological sound sensor system includes: a biological sound sensor 1 consisting of a main input sensor 5 installed in the vicinity of a lower tip of an extension part 4 extending downward from a holding part 3 that can be mounted on the head of a subject, which is brought into contact with the vicinity of the mastoid, and a reference input sensor 6 installed in the vicinity of an upper part of the extension part 4, and disposed in the vicinity of an auricle: and biological sound signal processing means 2. The biological sound signal processing means 2 executes reduction of noise from the main input using a main input amplitude spectrogram |S(t,ω)| and a reference input amplitude spectrogram |N(t,ω)| in which the main input and the reference input acquired by the biological sound sensor 1 are subjected to STFT, and a multichannel Wiener filter H(t,ω) designed on the basis of |S(t,ω)|2-|N(t,ω)|2 and |N(t,ω)|/|S(t,ω)|.SELECTED DRAWING: Figure 1

Description

This invention relates to a contact-type biological sound sensor that can measure blood vessel sounds and respiratory sounds using a microphone attached to the skin, including narrow spaces such as the inside of the ear, even in noisy environments.

The proportion of elderly people in Japan's total population was 10% in 1985, exceeded 20% in 2005, and is increasing year by year to 28.7% in 2020. This proportion is expected to continue to increase in the future and reach 35.3% in 2040. National medical expenses were 16.159 trillion yen in 1985 and 43.3949 trillion yen in 2018, increasing year by year in proportion to the elderly population. Self-medication in daily life (taking responsibility for one's own health and treating minor physical ailments oneself) is necessary to reduce medical expenses.
To realize self-medication, it is desirable to develop wearable devices that can constantly measure biological information, but there are various types of biological information such as pulse, respiration, blood pressure, and brain waves, and there are also a wide variety of measurement methods and corresponding devices. If it becomes possible to simultaneously measure pulse and respiration, early detection of circulatory and respiratory diseases (heart disease, pneumonia, etc.) can be expected, but the devices that can measure pulse and respiration installed in medical institutions are difficult to use for people without specialized knowledge other than medical professionals, and are not easy to use.

Therefore, the present inventors developed a contact-type biological sound sensor that uses a microphone attached relatively deep in the ear canal to detect vascular sounds and respiratory sounds conducted through bones, etc., separates the vascular sounds from the respiratory sounds, removes pulse noise from the separated respiratory sound waveform, and accurately detects abnormal lung conditions and respiratory sounds, as described in Patent Document 1 (JP 2020-121120 A) and Patent Document 2 (JP 2021-74464 A).
In addition, the inventors have discovered that, as described in non-patent document 1 (36th Life Support Society Conference (LIFE2020-2021), September 16-18, 2021, pp. 168-171), the signal-to-noise ratio can be improved by using a multi-channel Wiener filter that uses an adaptive margin m(t,ω) determined by the squared difference (|S(t,ω)| ² -|N(t,ω)| ² ) between the amplitude spectrograms of the main input and the reference input, which are composed of biological sounds and noise measured by two sensor units (condenser microphones) attached to the subject's head.
Furthermore, Patent Document 3 (Japanese Patent No. 6226301) describes the following: a first microphone (11) that outputs a main signal x(t) and a second microphone (12) that outputs a reference signal r1(t); the main signal x(t) and the reference signal r1(t) are input, and the signal is converted from the time domain to the frequency domain to output a main signal spectrum X(ω) and a reference signal spectrum R1(ω); the main signal spectrum X(ω) and the reference signal spectrum R1(ω) are input, and a main signal power spectrum Px(ω) and a second reference signal power spectrum Pr2(ω) are output; a noise suppression coefficient calculation unit (108B) outputs a noise suppression coefficient H(ω); and a weighting coefficient α(ω) corresponding to a time average of the spectral ratio Px(ω)/Pr1'(ω) is calculated using Px(ω) and the power spectrum Pr1'(ω) (see, in particular, paragraphs 0042 to 0054, 0173 to 0179, and FIGS. 1, 2, and 12).

JP 2020-121120 A JP 2021-074464 A Japanese Patent Publication No. 6226301 (International Publication No. 2014/097637)

Katsumasa Fujimoto, Mikiya Tanaka, Sotaro Shimada, Seiji Nishito, Shota Nakajima, "Noise Reduction Method for Biological Sound Using Multi-Channel Wiener Filter with Adaptive Margin," 36th Annual Conference of the Society for Life Support (LIFE2020-2021), September 16-18, 2021, pp.168-171

The biological sound sensors described in Patent Documents 1 and 2 are both attached to the ear canal to simultaneously measure blood vessel sounds and respiratory sounds, and separate the blood vessel sounds from the respiratory sounds. However, they do not take into consideration the elimination of noise, and because they are attached to the ear canal, they are likely to cause discomfort to the subject.
Furthermore, Non-Patent Document 1 describes a noise reduction method based on a main input and a reference input measured by sensors at two locations. However, as described in the last line of the right column on p. 169, the adaptive margin m(t,ω) used in the multi-channel Wiener filter is only 3 or 100, which is determined by (|S(t,ω)| ² -|N(t,ω)| ² ), and therefore the noise reduction ability may be inferior.
Furthermore, since the noise reduction method described in Patent Document 3 calculates a weighting coefficient α(ω) equivalent to the time average of the spectral ratio Px(ω)/Pr1'(ω), if there is a difference between the noise level corrected by calculation and the noise level mixed into the input, it may not be possible to determine an optimal weighting coefficient α(ω) for extracting the target voice, and a sufficient noise reduction effect may not be obtained.
The present invention is intended to solve these problems, and has as its first object to simultaneously measure vascular sounds and respiratory sounds without causing discomfort to the subject, and to improve noise reduction capabilities.
A second object of the present invention is to provide a biological sound sensor which can be easily attached to the head of a subject and which has sensors for mainly measuring vascular sounds and respiratory sounds and a sensor for measuring noise.

The biological sound sensor system according to claim 1 comprises:
a biological sound sensor including a main input sensor for mainly measuring vascular sounds and respiratory sounds to obtain a main input, and a reference input sensor for mainly measuring noise to obtain a reference input;
a Wiener filter designed from a primary input acquired by the primary input sensor and a reference input acquired by the reference input sensor, and performing noise reduction from the primary input;
the main input sensor is brought into contact with the subject near the mastoid process,
the reference input sensor is disposed above the main input sensor and in the vicinity of the subject's auricle;
the Wiener filter is a multi-channel Wiener filter that uses a main input amplitude spectrogram obtained by performing a short-time Fourier transform on the main input, a reference input amplitude spectrogram obtained by performing a short-time Fourier transform on the reference input, and an adjustment margin;
The adjustment margin is determined based on a value obtained by subtracting the square of the reference input amplitude spectrogram from the square of the main input amplitude spectrogram and a ratio of the main input amplitude spectrogram to the reference input amplitude spectrogram.

The invention according to claim 2 provides the biological sound sensor system according to claim 1,
the biological sound sensor includes a holding portion that can be detachably attached to the head of the subject and an extension portion that extends downward from the holding portion so that the primary input sensor is brought into contact with the vicinity of the mastoid process of the subject and the reference input sensor is disposed above the primary input sensor and in the vicinity of the auricle of the subject;
The primary input sensor is disposed near a lower end of the extension portion,
The reference input sensor is characterized in that it is installed near the upper portion of the extension portion.

The invention according to claim 3 provides the biological sound sensor system according to claim 1,
the biological sound sensor includes a left ear hook portion and a right ear hook portion that can be detachably attached to the left and right auricles of the subject, and a connecting portion that connects rear ends of the left ear hook portion and the right ear hook portion to each other, so that the main input sensor is brought into contact with the vicinity of the mastoid process of the subject and the reference input sensor is disposed above the main input sensor and in the vicinity of the auricle of the subject;
the main input sensor is disposed near a rear end of the left ear hook portion or the right ear hook portion,
The reference input sensor is characterized in that it is installed near the upper part of the left ear hook portion or the right ear hook portion.

According to the biosound sensor system of the invention of claim 1, the main input sensor is brought into contact with the subject's mastoid process, so that vascular sounds and respiratory sounds can be measured without causing discomfort to the subject. In addition, the Wiener filter that reduces noise from the main input is a multi-channel Wiener filter that uses a main input amplitude spectrogram, a reference input amplitude spectrogram, and an adjustment margin, and the adjustment margin is determined based on the value obtained by subtracting the square of the reference input amplitude spectrogram from the square of the main input amplitude spectrogram and the ratio between the main input amplitude spectrogram and the reference input amplitude spectrogram, so that the adjustment margin, which is a parameter that adjusts the amount of noise reduction, can be dynamically changed, and a better signal-to-noise ratio (hereinafter referred to as "SNR") can be obtained compared to conventional methods, thereby improving noise reduction capability.

The biological sound sensor system of the invention according to claim 2 has the effect of the invention according to claim 1, and further includes a holding part that can be detachably attached to the subject's head and an extension part that extends downward from the holding part, and the main input sensor is installed near the lower tip of the extension part, and the reference input sensor is installed above the extension part, so that the main input sensor and the reference input sensor can be easily attached to the subject's head.

According to the biosound sensor system of the invention of claim 3, in addition to the effects of the invention of claim 1, the system is equipped with a left ear hook part and a right ear hook part that can be attached and detached freely to the left and right auricles of the subject, and a connection part that connects the rear ends of the left and right ear hook parts, and the main input sensor is installed near the rear end of the left ear hook part or the right ear hook part, and the reference input sensor is installed near the upper part of the left ear hook part or the right ear hook part, so that the main input sensor and the reference input sensor can be easily attached to the subject's head.

FIG. 1 is a conceptual diagram of a biological sound sensor system according to a first embodiment. FIG. 2 is a diagram showing a state in which the biological sound sensor 1 of the first embodiment is attached to the head of a subject. 3 is a cross-sectional view showing the structures of the main input sensor 5 and the reference input sensor 6. FIG. Graph of adjustment margin α(t,ω) and adaptation margin m(t,ω) of Non-Patent Document 1. 11 is a table showing the SNR after noise reduction processing using each margin. FIG. 13 is a diagram showing a state in which the biological sound sensor 12 of the second embodiment is attached to the head of a subject. FIG. 13 is a diagram showing a state in which the biological sound sensor 17 of the third embodiment is attached to the head of a subject.

The following describes the embodiment of the present invention using examples.

FIG. 1 is a conceptual diagram of a biological sound sensor system according to a first embodiment, and FIG. 2 is a diagram showing a state in which the biological sound sensor 1 of the first embodiment is attached to the head of a subject.
As shown in FIG. 1, the body sound sensor system of Example 1 includes a body sound sensor 1 that is attached to the head of a subject, and a body sound signal processing means 2 that processes the body sounds acquired by the body sound sensor 1, reduces noise, and outputs the subject's vascular sound waveform and respiratory sound waveform.
The biological sound sensor 1 has a U-shaped elastic holding portion 3 that can be attached and detached freely to the subject's head, two elastic extension portions 4 extending downward from the holding portion 3, a main input sensor 5 installed near the lower tip of the extension portion 4, and a reference input sensor 6 installed facing outward near the upper part of the extension portion 4 (in Figure 1, in front of the upper end of the extension portion 4).
As shown in Figure 2, the holding portion 3 of the biological sound sensor 1 can be fixed to the back and sides of the subject's head by elastic force, and the main input sensor 5 installed near the lower tip of the extension portion 4 is positioned so that when the holding portion 3 is fixed to the subject's head, it comes into contact with the area around the mastoid process of the subject and is pressed slightly inward by elastic force.
The mastoid process is a cone-shaped protuberance located at the lower back of the temporal bone. By contacting the main input sensor 5 with the area around the mastoid process, vascular sounds and respiratory sounds can be measured simultaneously, without causing discomfort to most subjects.
In addition, since the subjects' heads vary in size and the positions of the mastoid processes, several different sized holding parts 3 and extension parts 4 are prepared, and 7 to 10 holes are provided on the front side of the holding part 3 so that the upper end of the extension part 4 can be screwed into any of these holes.

FIG. 3 is a cross-sectional view showing the structures of the main input sensor 5 and the reference input sensor 6. As shown in FIG.
Both the main input sensor 5 and the reference input sensor 6 are acoustic sensors that use an electret condenser microphone (hereinafter referred to as "ECM") 7 shown in Fig. 3(B), and as shown in Fig. 3(A), are composed of the ECM 7, an acoustic conductive part 8 made of polyurethane elastomer, and a case 9 made of photocurable resin. The ECM 7 is housed in a housing 10, and has a wave receiving surface 11 which is the surface of a vibration plate (diaphragm) that detects sound, etc.
The housing 10 is housed inside a case 9 made of photocurable resin, and the surface of the wave-receiving surface 11 and the periphery of the housing 10 are covered with an acoustic conductive portion 8 made of polyurethane elastomer.
The reason why the surface of the wave receiving surface 11 is covered with the acoustic conductive part 8 is that if the ECM 7 is brought directly close to the surface of the human body to acquire body sound, the body sound will propagate through the air, which has acoustic properties different from those of the human body, and will be reflected and attenuated at the boundary surface. In order to propagate the body sound to the ECM 7 without attenuating it, it is advisable to fill the space between the surface of the human body and the surface of the wave receiving surface 11 with a medium that has the same acoustic properties as the human body.
Therefore, in the first embodiment, the surface of the wave receiving surface 11 is covered with polyurethane elastomer having acoustic properties equivalent to those of human skin. This makes it possible to transmit the body sounds acquired from the mastoid process area, particularly in the main input sensor 5, to the ECM 7 with high accuracy.
2, the main input sensor 5 can mainly measure vascular sounds and respiratory sounds, and the reference input sensor 6 can mainly measure noise. Note that since the reference input sensor 6 is used without contacting the subject, only the ECM 7 may be used.

Next, a method for processing body sounds in the body sound signal processing means 2 will be described.
As shown in Fig. 1, first, a primary input signal acquired by a primary input sensor 5 and a reference input signal acquired by a reference input sensor 6 are transmitted to a biological sound signal processing means 2 via a wired or wireless transmission means. Then, the biological sound signal processing means 2 processes the primary input signal and the reference input signal in the following procedure, performs noise reduction from the primary input signal, and then performs signal separation by the conventional technology to obtain a highly accurate vascular sound waveform and respiratory sound waveform.
(Step 1) A short-time Fourier transform (hereinafter referred to as "STFT") is performed on the received main input signal and reference input signal to obtain the main input amplitude spectrogram (|S(t,ω)|) and the reference input amplitude spectrogram (|N(t,ω)|).
(Step 2) Noise is reduced from the main input using a noise reduction method with a multi-channel Wiener filter that uses a margin. Specifically, in addition to |S(t,ω)| and |N(t,ω)| obtained in step 1 above, a multi-channel Wiener filter H(t,ω) designed from an adjustment margin α(t,ω) for gain adjustment is used. The designed multi-channel Wiener filter H(t,ω) is shown in the following equation (1). Note that (t,ω) indicates the components at each time t and angular frequency ω.
H(t,ω)=|S(t,ω)| ² / (|S(t,ω)| ² + α(t,ω)|N(t,ω)| ² ) ... Equation (1)
In the first embodiment, the adjustment margin α(t,ω) is defined by the following equations (2-1) and (2-2) according to the value obtained by subtracting the square of the reference input amplitude spectrogram from the square of the main input amplitude spectrogram (the value of |S(t,ω)| ² - |N(t,ω)| ² ).
|S(t,ω)| ² -|N(t,ω)| ² ＞0: α(t,ω)=10(|N(t,ω)|/|S(t,ω)|)・・・Equation (2-1)
|S(t,ω)| ² -|N(t,ω)| ² ≦0: α(t,ω)=30(|N(t,ω)|/|S(t,ω)|)・・・Equation (2-2)
Then, H(t,ω) is designed using α(t,ω) defined in equations (2-1) and (2-2), and noise is reduced from the main input by calculating H(t,ω)S(t,ω), and a restored body sound X'(t,ω) that is close to the original body sound X(t,ω) can be obtained.
That is, when there is not much noise and |S(t,ω)| ² -|N(t,ω)| ² > 0, the adjustment margin α(t,ω) changes in proportion to the value of |N(t,ω)|/|S(t,ω)| within the range of 0≦α(t,ω)<10 (see the solid line on the left side of Figure 4), but since α(t,ω) is relatively small, noise is reduced while suppressing degradation of the main input. Also, when noise is mixed in and |S(t,ω)| ² -|N(t,ω)| ² ≦0, the adjustment margin α(t,ω) changes in proportion to the value of |N(t,ω)|/|S(t,ω)| within the range of 30≦α(t,ω) (see the solid line on the right side of Figure 4), but since α(t,ω) is relatively large, noise is significantly reduced.
In contrast, the adaptive margin m(t,ω) in Non-Patent Document 1 is fixed to 3 when |S(t,ω)| ² -|N(t,ω)| ² > 0, as shown by the dotted line in Fig. 4, so that noise reduction becomes insufficient when the value of |N(t,ω)|/|S(t,ω)| is close to 1. Conversely, when |S(t,ω)| ² -|N(t,ω)| ² ≦ 0, the adaptive margin is fixed to 100, so that when the value of |N(t,ω)|/|S(t,ω)| is close to 1, excessive noise reduction is performed, resulting in degradation of the primary input.

(Step 3) Since the body sound X(t,ω) contains both vascular sounds and respiratory sounds, it is necessary to separate them based on the frequency difference between the two. Specifically, since the frequency of the vascular sounds is about 75 to 200 Hz and the frequency of the respiratory sounds is about 200 to 2000 Hz, the restored body sound X'(t,ω) obtained in step 2 above is processed using a low-pass band-pass filter to pass only signals with frequencies of 20 to 100 Hz, and using a high-pass band-pass filter to pass only signals with frequencies of 200 to 2000 Hz, thereby separating the vascular sounds and respiratory sounds.
(Step 4) The vascular sound separated in step 3 above is subjected to an inverse short-time Fourier transform (hereinafter referred to as "ISTFT") to obtain a vascular sound waveform.
(Step 5) Since slight vascular sounds remain in the respiratory sounds separated in step 3 above due to differences in sound pressure, the residual vascular sounds are reduced by harmonic percussive sound separation (HPSS) for the separated respiratory sounds, and then ISTFT is performed to obtain the respiratory sound waveform.
(Step 6) The acquired vascular sound waveform and respiratory sound waveform are sent to the waveform display means depending on the purpose of subsequent use, for example, when a doctor or the like checks the waveforms, and are displayed as vascular sound waveforms and respiratory sound waveforms, or the fluctuations in vascular sounds, the fluctuations in respiratory sounds, or the stress index (SI) are calculated and displayed. In addition, when a condition is judged based on the calculated vascular sounds, respiratory sounds, etc. and the result of the judgment is to be displayed, the waveforms are sent to the vascular sound/breath sound, etc. calculation and display means.

Next, an experiment was carried out to verify the noise reduction effect of the multi-channel Wiener filter using the adjustment margin α(t,ω) of the first embodiment, and the experimental procedure and results will be described.
In this experiment, the SNR defined by the following equation (3) was used to evaluate the noise reduction effect.
SNR=20 log ₁₀ [max(|s(t)|)/max(|n(t)|)] (3)
Here, s(t) is a time signal in which only body sounds are observed, and n(t) is a time signal in which only noise is observed. The experimental method will be explained step by step below.
(Experimental Procedure 1) The subjects were 10 people (Subjects A to J) in their 20s to 80s. They were seated in a chair in a quiet environment. The biosound sensor 1 was attached to the head of each subject as shown in Figure 2, and the biosound was measured for 10 seconds by the main input sensor 5.
(Experimental Procedure 2) To verify the effect of noise reduction, two types of noise (a 600 Hz sine wave and a male voice) were added to the measured body sounds.
The noise intensity was adjusted to be equal to the maximum amplitude of the respiratory sounds.
(Experimental procedure 3) For the sound signal to which two types of noise were added, noise reduction processing was performed using a multi-channel Wiener filter with an adaptive margin m(t, ω) fixed at 3 as in Non-Patent Document 1, and a multi-channel Wiener filter using an adjustment margin α(t, ω).
(Experimental Procedure 4) The SNRs of the respiratory sound waveform obtained after noise reduction processing using an adaptive margin m(t, ω)=3 and the respiratory sound waveform obtained after noise reduction processing using an adjustment margin α(t, ω) were calculated.
Note that vascular sound waveforms were not verified because a high SNR was obtained without noise reduction processing.
(Experimental Procedure 5) For 20 sound signal patterns in which two types of noise were added to the biological sounds of subjects A to J, the SNR of the respiratory sound waveform obtained after noise reduction processing using an adaptive margin m(t, ω) = 3 was compared with the SNR of the respiratory sound waveform obtained after noise reduction processing using an adjustment margin α(t, ω).

(Experimental result)
FIG. 5(A) is a table showing the SNR of the respiratory sound waveform obtained after noise reduction processing with an adaptive margin m(t,ω)=3 and the SNR of the respiratory sound waveform obtained after noise reduction processing with an adjustment margin α(t,ω) for sound signals in which a sine wave has been added to the body sounds of subjects A to J. FIG. 5(B) is a table showing the SNR of the adaptive margin m(t,ω)=3 and the SNR of the adjustment margin α(t,ω) obtained by similar processing for sound signals in which a male voice has been added to the body sounds of subjects A to J.
Here, since a higher SNR indicates less noise influence, it can be said that a higher SNR indicates a higher noise reduction effect, and a lower SNR indicates a lower noise reduction effect.
5(A) and (B), the SNR with the adjustment margin α(t, ω) was higher than the SNR with the adaptation margin m(t, ω) = 3 for all subjects, and it can be seen that the noise reduction effect by the multi-channel Wiener filter using the adjustment margin α(t, ω) of Example 1 is higher than the noise reduction effect by the multi-channel Wiener filter in which the adaptation margin m(t, ω) of Non-Patent Document 1 is fixed at 3.
In other words, it can be said that using the adjustment margin α(t, ω) in the first embodiment is effective for noise reduction by a multi-channel Wiener filter regardless of the type of noise.

FIG. 6 is a diagram showing a state in which the biological sound sensor 12 of the second embodiment is attached to the head of a subject.
In the biological sound sensor 1 of Example 1, the holding portion 3 is fixed to the occipital and temporal parts of the subject by elastic force, whereas in the biological sound sensor 12 of Example 2, the holding portion 13 is fixed to the forehead and temporal parts of the subject by elastic force, and a main input sensor 15 is installed near the lower tip of an elastic extension portion 14 that extends downward from the holding portion 13, and a reference input sensor 16 is installed facing outward near the upper part of the extension portion 14.
In addition, the basic configurations of the holding unit 13, extension unit 14, main input sensor 15 and reference input sensor 16 are the same as those of the holding unit 3, extension unit 4, main input sensor 5 and reference input sensor 6 in Example 1, so detailed explanations are omitted, and the biological sound signal processing means 2 has exactly the same configuration as in Example 1, so explanations are omitted.
As described above, the holding portion 13 of Example 2 is fixed to the frontal and temporal regions of the subject, and therefore has the advantage of being easy to use even for subjects with long hair or who are in a supine position.

FIG. 7(A) is a diagram showing the appearance of the left ear hook part 18, the right ear hook part 19, and the connection part 20 of the biological sound sensor 17 of Example 3, and FIG. 7(B) is a diagram showing the biological sound sensor 17 of Example 3 attached to the head of a subject.
In the biological sound sensor 1 of Example 1 and the biological sound sensor 12 of Example 2, the holding parts 3, 13 are fixed to the head of the subject by elastic force, but in the biological sound sensor 17 of Example 3, the left ear hook part 18, the right ear hook part 19, and the connection part 20 connecting the rear ends of the left ear hook part 18 and the right ear hook part 19 can be fixed by hanging them on the left and right ears of the subject. Therefore, it takes less time to put on the sensor compared to the biological sound sensors 1, 12 of Examples 1 and 2, and the sensor can be positioned easily, so that stable biological sounds (blood vessel sounds and respiratory sounds) can be acquired.
The main input sensor is installed near the rear end of the left ear hook part 18 or the right ear hook part 19, and the reference input sensor is installed near the upper part of the left ear hook part 18 or the right ear hook part 19 facing outward.
In addition, in the body sound sensor 17 of Example 3, the main input sensor and the reference input sensor are installed on the right ear hook portion 19, so these sensors are not shown in FIG. 7(B).
Furthermore, the basic configurations of the main input sensor and the reference input sensor used in Example 3 are the same as those of the main input sensor 5 and the reference input sensor 6 in Example 1, so detailed explanations will be omitted, and the biological sound signal processing means 2 has exactly the same configuration as in Example 1, so explanations will be omitted.

Modifications of the biological sound sensor systems of the first to third embodiments will be listed below.
(1) In Examples 1 and 2, U-shaped elastic holding parts 3 and 13 that can be attached and detached to the subject's head were used, but a ring-shaped elastic holding part that can be attached and detached to the entire circumference of the subject's head or a hat-shaped holding part whose circumference can be adjusted may also be used.
(2) In the first and second embodiments, the body sound sensor 1 has two elastic extensions 4 that extend downward from the left and right sides of the holder 3. However, only one extension 4 may be provided.
(3) In the first and second embodiments, the main input sensors 5, 15 and the reference input sensors 6, 16 are disposed only on the left side of the subject, but they may be disposed only on the right side or on both sides.
In addition, in the third embodiment, the main input sensor and the reference input sensor are provided only on the right ear hook 19, but they may be provided only on the left ear hook 18 or on both ear hooks 18, 19.
(4) In Examples 1 and 2, the reference input sensor 6 is installed facing outward at the front of the upper end of the extension portion 4, but the installation position may be behind, above, or below the upper end of the extension portion 4, and the orientation may be inward, forward, backward, upward, or downward.
In addition, in the third embodiment, the reference input sensor is disposed facing outward near the upper portion of the right ear hook 19, but the orientation may be inward, forward, backward, upward, or downward.

(5) In the first to third embodiments, ECMs are used for the main input sensors 5, 15 and the reference input sensors 6, 16. However, instead of the ECMs, ordinary capacitor type, moving coil type, piezoelectric type, or other microphones may be used.
(6) In Examples 1 to 3, the surface of the wave receiving surface 11 and the periphery of the housing 10 are covered with an acoustic conductive section 8 made of polyurethane elastomer. However, the material is not limited to polyurethane elastomer and may be medical grade silicone. Any suitable elastic polymer material may be used as long as it is a hydrophobic resin that has acoustic impedance characteristics equivalent to those of human skin after hardening.
(7) In the first to third embodiments, the multi-channel Wiener filter H(t, ω) is designed using the adjustment margin α(t, ω) defined by equations (2-1) and ( ^2-2 ). However, it would be better to reduce the discontinuity of α(t, ω) or connect α(t, ω) seamlessly by, for example, dividing the value of |S(t, ω)| ² -|N(t, ω)| 2 into three or more stages.
For example, α(t,ω) can be defined by the following equations (4-1) to (4-3).
1-|N(t,ω)| ² /|S(t,ω)| ² ＞0.36: α(t,ω)=10(|N(t,ω)|/|S(t,ω)|)・・・・Equation (4-1)
0.36≧1-|N(t,ω)| ² /|S(t,ω)| ² ＞-1.25: α(t,ω)=20(|N(t,ω)|/|S(t,ω)|)…Equation (4-2)
-1.25 ≧ 1 - |N(t,ω)| ² /|S(t,ω)| ² : α(t,ω) = 30 (|N(t,ω)|/|S(t,ω)|) ... Equation (4-3)
(8) In the first to third embodiments, the vascular sounds and the respiratory sounds are separated using a low-pass band-pass filter and a high-pass band-pass filter. However, the vascular sounds and the respiratory sounds may be separated using a low-pass filter that passes all signals with frequencies lower than a predetermined frequency and a high-pass filter that passes all signals with frequencies higher than the predetermined frequency.

REFERENCE SIGNS LIST 1 Body sound sensor 2 Body sound signal processing means 3 Holding portion 4 Extension portion 5 Primary input sensor 6 Reference input sensor 7 ECM
8 Acoustic conduction section 9 Case 10 Housing 11 Receiving surface 12 Body sound sensor 13 Holding section 14 Extension section 15 Primary input sensor 16 Reference input sensor 17 Body sound sensor 18 Left ear hook section 19 Right ear hook section 20 Connection section ECM Electret condenser microphone HPSS Harmonic percussion sound separation ISTFT Inverse short-time Fourier transform SNR Signal-to-noise ratio SI Stress index STFT Short-time Fourier transform t Time ω Angular frequency α(t,ω) Adjustment margin H(t,ω) Multi-channel Wiener filter
m(t,ω) Adaptation margin |N(t,ω)| Reference input magnitude spectrogram
|S(t,ω)| Primary input amplitude spectrogram
X(t,ω) Body sound X'(t,ω) Reconstructed body sound

Claims (3)

a biological sound sensor including a main input sensor for mainly measuring vascular sounds and respiratory sounds to obtain a main input, and a reference input sensor for mainly measuring noise to obtain a reference input;
a Wiener filter designed from a primary input acquired by the primary input sensor and a reference input acquired by the reference input sensor, and performing noise reduction from the primary input;
the main input sensor is brought into contact with the subject near the mastoid process,
the reference input sensor is disposed above the main input sensor and in the vicinity of the subject's auricle;
the Wiener filter is a multi-channel Wiener filter that uses a main input amplitude spectrogram obtained by performing a short-time Fourier transform on the main input, a reference input amplitude spectrogram obtained by performing a short-time Fourier transform on the reference input, and an adjustment margin;
the adjustment margin is determined based on a value obtained by subtracting the square of the reference input amplitude spectrogram from the square of the main input amplitude spectrogram and a ratio of the main input amplitude spectrogram to the reference input amplitude spectrogram. the biological sound sensor includes a holding portion that can be detachably attached to the head of the subject and an extension portion that extends downward from the holding portion so that the primary input sensor is brought into contact with the vicinity of the mastoid process of the subject and the reference input sensor is disposed above the primary input sensor and in the vicinity of the auricle of the subject;
The primary input sensor is disposed near a lower end of the extension portion,
The biological sound sensor system according to claim 1 , wherein the reference input sensor is disposed in the vicinity of an upper portion of the extension portion. the biological sound sensor includes a left ear hook portion and a right ear hook portion that can be detachably attached to the left and right auricles of the subject, and a connecting portion that connects rear ends of the left ear hook portion and the right ear hook portion to each other, so that the main input sensor is brought into contact with the vicinity of the mastoid process of the subject and the reference input sensor is disposed above the main input sensor and in the vicinity of the auricle of the subject;
the main input sensor is disposed near a rear end of the left ear hook portion or the right ear hook portion,
2. The body sound sensor system according to claim 1, wherein the reference input sensor is disposed near an upper portion of the left ear hook part or the right ear hook part.