TWI442907B - A whey sound recognition device based on biorthogonal wavelet filter - Google Patents

Description

Wheezing sound recognition device based on double orthogonal wavelet filter group

The invention relates to a wheezing sound recognition device, in particular to a wheezing sound recognition device based on a biorthogonal wavelet filter group.

In the past, the most common way to detect abnormal lung sounds (eg, wheezing) in patients is to have a chest diagnosis by a doctor. However, due to the particularity of the wheezing sound (the corresponding frequency range is above and below 400 Hz, and lasts for at least 250 ms), the doctor may use the auscultation to determine whether the wheezing sound may be objectionable. In addition, the traditional auscultation method is not immediate enough for those who are prone to wheezing.

In order to solve the above problem, a conventional method is to continuously analyze the patient's voice signal in the frequency range by using a wheezing sound detecting device to monitor the condition of the patient at any time.

In general, the algorithm used to analyze the sound signal is Short Time Fourier Transform (STFT). However, the shortcoming of the above algorithm is that the windowing used is fixed, that is, it is not easy to clearly distinguish the high frequency signal and the low frequency signal in the sound signal, so the above algorithm is less suitable for processing the wheezing sound. Sound signal.

Accordingly, it is an object of the present invention to provide a wheezing sound recognition apparatus based on a biorthogonal wavelet filter set.

Therefore, the present invention is based on a wobble sound recognition device of a biorthogonal wavelet filter group, and is suitable for recognizing a wheezing sound in a lung sound of a patient, the device comprising a lung sound signal processing unit, a characteristic signal obtaining unit, and a Wheezing recognition unit.

The lung sound signal processing unit is configured to receive the lung sound of the patient and generate a corresponding sound signal.

The characteristic signal obtaining unit includes a plurality of bi-orthogonal wavelet filtering groups, and is configured to pass the audio signal through the bi-orthogonal wavelet filtering groups to obtain a characteristic signal of the audio signal. The feature signal corresponds to a frequency range and includes a plurality of sound frames.

The wheezing sound recognition unit is configured to calculate a relative wavelet energy of each sound frame of the characteristic signal. When at least one of the relative wavelet energies exceeds a threshold value and the corresponding time lasts for an observation time (250 ms) or more, the wheezing sound recognition unit determines that the lung sound contains a wheezing sound.

The function of the present invention is to obtain the characteristic signal corresponding to the high frequency wheezing sound by the bi-orthogonal wavelet filtering group, and to identify the patient by determining whether the relative wavelet energy of the characteristic signal exceeds a critical value. Whether the lung sound contains a wheezing sound.

The above and other technical contents, features and advantages of the present invention will be apparent from the following detailed description of the preferred embodiments.

Referring to FIG. 1 and FIG. 2, the Wheezing identification device 1 of the present invention is based on a Bi-orthogonal Wavelet Filter Bank, which is suitable for identifying a wheezing sound in a patient's lung sound, and The present invention includes a body 11 , a lung sound signal processing unit 12 disposed in the body 11 , a characteristic signal obtaining unit 13 disposed in the body 11 , a wheezing sound identifying unit 14 disposed in the body 11 , and An alert unit 15 disposed in the body 11. The lung sound signal processing unit 12 includes a microphone 121 and an analog to digital convertor 122. The feature signal obtaining unit 13 includes a plurality of bi-orthogonal wavelet filter groups. In the preferred embodiment, the characteristic signal obtaining unit 13 and the wheezing sound identifying unit 14 are implemented in a hardware manner, and the implementation is a logic circuit integrated in a wheezing sound recognition chip. It is worth mentioning that the preferred embodiment is a portable device that can be worn by a patient.

The following is a description of the operation between the above units for the wheezing sound recognition method based on the biorthogonal wavelet filter group and an application example.

As shown in step S1, after the microphone 121 of the lung sound signal processing unit 12 is configured to receive the lung sound of the patient, the analog to digital converter 122 of the lung sound signal processing unit 12 receives the lung sound according to the microphone 121. , generating a corresponding digital sound signal. In the preferred embodiment, the sampling frequency employed is 8 kHz.

The waveform of the sound signal is shown in Figure 6. It can be observed from FIG. 6 that the waveform of the sound signal can be initially divided into a first portion 61 and a second portion 62. Compared to the first portion 61, the amplitude of the waveform of the second portion 62 may be relatively densely oscillated over a certain number of time intervals.

As shown in step S2, the characteristic signal obtaining unit 13 is configured to pass the audio signal through the bi-orthogonal wavelet filtering group to obtain a characteristic signal of the audio signal. The characteristic signal corresponds to a frequency range of 250 Hz to 500 Hz, and includes a plurality of sound frames. Each of the sound boxes includes 512 sampling points.

In the preferred embodiment, the bi-orthogonal wavelet filtering group is a Dobecy 9/7 biorthogonal wavelet filtering group (also known as a Cohen-Daubechies-Feauveau 9/7 tap bi-orthogonal wavelet filtering group), It is implemented in a polyphase structure with a distributed arithmetic algorithm (Distributed Arithmetic). Among them, the advantage of adopting the above embodiment is that the calculation is more efficient.

The implementation details of these bi-orthogonal wavelet filter groups are further described below.

Discrete wavelet transform:

The bi-orthogonal wavelet filter group is one of the categories of Discrete Wavelet Transform (DWT). It includes a high pass wavelet filter set (High Pass Filter) and a low pass wavelet filter set (Low Pass Filter). In general, the discrete wavelet converted signal is obtained by the following formula:

Where x[n] represents the voice signal in the preferred embodiment, h[2k-n] represents a high-pass wavelet filter group, d[k] represents a high communication number, and g[2k-n] represents a low-pass wavelet filter group. , a[k] represents a low communication number.

In the preferred embodiment, the feature signal obtaining unit 13 passes the audio signal through the four-level bi-orthogonal wavelet filtering group (see FIG. 3) to obtain the characteristic signal d4[k] (ie, through the fourth The characteristic signal generated after the layer's biorthogonal wavelet filter group). That is, the feature signal obtaining unit 13 obtains a1[k], a2[k], and a3[k] sequentially, and then obtains d4[k] according to a3[k]. Since the frequency range (250 Hz to 500 Hz) corresponding to the characteristic signal d4[k] covers the frequency range corresponding to the occurrence of the wheezing sound, the characteristic signal d4[k] is suitable for use as a basis for judging whether there is a wheezing sound. The details of the audio signal passing through the bi-orthogonal wavelet filtering groups are easily understood by those skilled in the art, and therefore will not be described herein.

Multi-phase structure:

In the preferred embodiment, the bi-orthogonal wavelet filter sets are further implemented in a multi-phase structure. As shown in FIG. 4, the filter coefficients include a high pass filter group and a low pass filter group. First, the high-pass filter group and the low-pass filter group are divided into odd components (h _odd and g _odd ) and even components (h _even and g _even ). Then, the sound signals generated by the lung sound signal processing unit 12 are multiplexed into odd samples (Odd Samples, X _odd ) and even samples (Even Samples, X _even ). Finally, the odd sample and the even sample are convolved with the odd component and the even component, respectively. among them, 2 stands for downsampling.

The following formula is used to represent the relationship between the sound signal and the bi-orthogonal wavelet filter groups:

Where d[n] represents a high communication number and a[n] represents a low communication number.

In addition, the odd and even components in the filter coefficients are represented by binary 16-bit floating point numbers, as shown in Table 1 and Table 2, respectively:

Decentralized arithmetic algorithm: In order to reduce the multiplication operation in the operation, it is usually possible to add the parts multiplied by the same coefficient first and then multiply the corresponding coefficients. In the preferred embodiment, the feature signal can be represented by the following formula by a decentralized arithmetic algorithm:

Where x[n] represents an audio signal in the format of M bits. In the preferred embodiment, the format of the audio signal is 8 bits.

Referring to FIG. 5, according to the above manner, the feature signal obtaining unit 13 obtains the high communication number after the audio signal is passed through the bi-orthogonal wavelet filtering group implemented by the multi-phase structure with the distributed arithmetic algorithm. d[n] and the low communication number a[n] can be expressed by the following formula:

The waveform of the characteristic signal d4[k] is as shown in FIG. 7. The characteristic signal can be divided into a first portion 71 and a second portion 72, and corresponds to the first portion 61 and the second portion 62 of FIG. 6, respectively. . It can be observed from Fig. 7 that after the double orthogonal wavelet filter group, the amplitude of the waveform of the second portion 72 is significantly larger than the amplitude of the waveform of the first portion 71 compared to the first portion 71.

As shown in step S3, the wheezing sound recognition unit 14 calculates the relative wavelet energy of each sound frame of the characteristic signal according to a pre-obtained average energy and wavelet energy of each sound frame of the characteristic signal. .

The wheezing sound recognition unit 14 calculates the wavelet energy and the relative wavelet energy by the following formula.

Where k is the sampling point of a sound box, F is the total number of sampling points of the sound box, j is the frequency range corresponding to the characteristic signal, d _j [ k ] ² is the wavelet energy of the sampling point k, E _j The wavelet energy of the frame, E _normal is the average energy, and the RWE is the relative wavelet energy of the frame.

It is worth mentioning that the pre-obtained average energy patient is in general (ie, the lung sound does not contain wheezing sound), and the relative of each sound box corresponding to the normal lung sound is calculated in the above manner. The average of the wavelet energy.

As shown in step S4, the wheezing sound recognition unit 14 determines whether at least one of the relative wavelet energies exceeds a critical value, and the corresponding time lasts for more than one observation time. If so, the wheezing recognition unit 14 determines that the lung sound contains a wheezing sound, and proceeds to step S5. In the preferred embodiment, the observation time is 250 ms and the threshold is 4. Of course, the threshold may be adjusted to other values as the test environment differs from the patient, and is not limited to the preferred embodiment.

The relationship between the relative wavelet energy of the characteristic signal and the critical value is as shown in FIG. The first portion 81 of FIG. 8 represents the relative wavelet energy corresponding to the first portion 71 of FIG. 7, and the relative wavelet energies are both below the threshold. The second portion 82 of Figure 8 represents the relative wavelet energy corresponding to the second portion 72 of Figure 7, with a portion of the relative wavelet energy being above the threshold and lasting longer than 250 ms.

As shown in step S5, when at least one of the relative wavelet energies exceeds the threshold and the corresponding time continues above the observation time, the wheezing recognition unit 14 generates an alert signal. That is, the wheezing sound recognition unit 14 generates the warning signal if it is determined that the lung sound contains a wheezing sound.

As shown in step S6, the alert unit 15 is configured to receive the alert signal and generate a warning effect. The warning effect is at least one of a visual warning effect and an audible warning effect. That is, when the patient's lung sound contains a wheezing sound, the warning effect generated by the warning unit 15 can be used by the surrounding medical staff or relatives to immediately assist the patient to avoid the golden time of missing the rescue. In the preferred embodiment, the alert unit 15 can be a speaker that produces the audible alert effect, or a Light Emitting Diode (LED) that produces the visual alert effect.

In summary, by obtaining the characteristic signal of the sound signal, and calculating the relative wavelet energy of the characteristic signal, it is determined whether the relative wavelet energy exceeds the critical value; when at least one of the relative wavelet energies exceeds When the critical value and the corresponding time continue above the observation time, the warning effect is generated, and the object of the present invention can be achieved.

The above is only the preferred embodiment of the present invention, and the scope of the invention is not limited thereto, that is, the simple equivalent changes and modifications made by the scope of the invention and the description of the invention are All remain within the scope of the invention patent.

1. . . Wheezing sound recognition device based on double orthogonal wavelet filter group

11. . . Ontology

12. . . Lung sound signal processing unit

121. . . microphone

122. . . Analog digital converter

13. . . Feature signal acquisition unit

14. . . Wheezing recognition unit

15. . . Warning unit

61. . . Lung sound signal waveform

62. . . Lung sound signal waveform

71. . . Characteristic signal waveform

72. . . Characteristic signal waveform

81. . . Relative wavelet energy

82. . . Relative wavelet energy

S1~S6. . . step

1 is a system diagram illustrating a preferred embodiment of a wheezing sound recognition apparatus based on a biorthogonal wavelet filter set of the present invention;

2 is a flow chart illustrating a wheezing sound recognition method based on a bi-orthogonal wavelet filter group corresponding to the preferred embodiment;

3 is a schematic diagram showing a four-level bi-orthogonal wavelet filter group corresponding to the preferred embodiment;

Figure 4 is a schematic view showing a multi-phase structure corresponding to the preferred embodiment;

FIG. 5 is a schematic diagram showing a shift plus distributed computing architecture corresponding to the preferred embodiment; FIG.

Figure 6 is a schematic view showing the waveform of the lung sound signal corresponding to the preferred embodiment;

Figure 7 is a schematic view showing waveforms of characteristic signals corresponding to the preferred embodiment; and

Figure 8 is a schematic diagram showing the relative wavelet energy corresponding to the characteristic signals of the preferred embodiment.

1. . . Wheezing sound recognition device based on double orthogonal wavelet filter group

11. . . Ontology

12. . . Lung sound signal processing unit

121. . . microphone

122. . . Analog digital converter

13. . . Feature signal acquisition unit

14. . . Wheezing recognition unit

15. . . Warning unit

Claims (10)

A wheezing sound recognition device based on a biorthogonal wavelet filter set, which is suitable for recognizing a wheezing sound in a patient's lung sound, the device comprising: a lung sound signal processing unit for receiving a lung sound of a patient and generating a corresponding signal acquisition unit, comprising a plurality of bi-orthogonal wavelet filter groups, and configured to pass the audio signal through the bi-orthogonal wavelet filter groups to obtain a characteristic signal of the audio signal, wherein The characteristic signal corresponds to a frequency range and includes a plurality of sound frames; and a wheezing sound identifying unit configured to calculate a relative wavelet energy of each of the sound signals of the characteristic signal, wherein at least one of the relative wavelet energies When the threshold value is exceeded and the corresponding time continues for more than one observation time, the wheezing sound recognition unit determines that the lung sound contains a wheezing sound. The wobble sound recognition device based on the biorthogonal wavelet filter group according to the first aspect of the patent application, wherein the biorthogonal wavelet filter group is a Dobecy 9/7 biorthogonal wavelet filter group. The wheezing sound recognition device based on the biorthogonal wavelet filter group according to the second aspect of the patent application, wherein the Dobecy 9/7 biorthogonal wavelet filter group is implemented in a multi-phase structure. A wheezing sound recognition device based on a biorthogonal wavelet filter group according to claim 3, wherein the Dobercy 9/7 biorthogonal wavelet filter group is a multi-phase with a distributed arithmetic algorithm Structure to implement. The wheezing sound recognition device based on the biorthogonal wavelet filter group according to claim 1, wherein the characteristic signal corresponds to the frequency range of 250 Hz to 500 Hz. The wheezing sound recognition device based on the biorthogonal wavelet filter group according to claim 1, wherein the wheezing sound recognition unit is based on a pre-obtained average energy, and each of the sound signals of the characteristic signal Wavelet energy to calculate the relative wavelet energy of each frame of the feature signal. The wheezing sound recognition device based on the biorthogonal wavelet filter group according to claim 1, further comprising a warning unit, wherein when at least one of the relative wavelet energies exceeds the threshold and corresponds to a time The wheezing sound recognition unit further generates a warning signal for receiving the warning signal and generating a warning effect. The wheezing sound recognition device based on the biorthogonal wavelet filter group according to claim 7, wherein the warning effect is at least one of a visual warning effect and an audible warning effect. The wheezing sound recognition device based on the biorthogonal wavelet filter group according to claim 1, wherein the observation time is 250 ms. The wheezing sound recognition device based on the biorthogonal wavelet filter group according to claim 1 further includes a lung sound signal processing unit, the characteristic signal obtaining unit, and the wheezing sound identifying unit. An ontology.