JP4913666B2 - Cough detection device and cough detection program - Google Patents

Description

  The present invention relates to a cough detection device and a cough detection program capable of distinguishing speech and cough and detecting cough with high accuracy.

  In general, cough is often a parameter for respiratory diseases, and how to detect cough easily and with high accuracy is a problem. In particular, the frequency of how often the patient or subject coughs during sleep may not be recognized by the patient or subject, or may be recognized at an incorrect frequency. There are many cases.

  In view of such circumstances, for example, in Patent Document 1, a sound pressure curve is created, and a region in which a sound pressure higher than or equal to a predetermined value in the sound pressure curve continues for a predetermined time or more is defined as a cough sound candidate. An apparatus for detecting whether a coughing sound is detected based on rising slope, kurtosis, and the like is shown.

According to this apparatus, although it is said that a good cough sound is detected, the detection accuracy is about 80 to 85% of cough, which is not necessarily sufficient.
JP 2003-38460 A

  The present invention has been made in view of the current state of the conventional cough detection apparatus as described above, and an object thereof is to provide a cough detection apparatus and a cough detection program capable of performing cough detection with higher accuracy. It is.

  The cough detection device according to the present invention includes a voice capturing unit that captures a subject's voice, an A / D converter that digitizes the voice captured by the voice capturing unit, and a signal converted by the A / D converter. Pulse detection means for detecting a pulse by obtaining an envelope using a digital value, Fourier transform means for performing Fourier transform on the detected pulse, and cepstrum generation means for generating a cepstrum using the Fourier transform result And a determination unit that compares the cepstrum generated by the cepstrum generation unit with a predetermined threshold value and determines that the cepstrum is cough when the cepstrum is smaller than the predetermined threshold value.

  In the cough detection apparatus according to the present invention, the pulse detection means includes a pulse selection means for selecting a pulse having an amplitude and a pulse width that are greater than or equal to a predetermined value.

  In the cough detection device according to the present invention, the cepstrum generation means includes logarithm calculation means for performing a logarithmic operation on the Fourier transform result, and an inverse Fourier transform means for performing an inverse Fourier transform on the output of the logarithmic operation means. To do.

  The cough detection device according to the present invention is characterized by comprising a predetermined threshold generation means for obtaining a predetermined threshold used by the determination means.

  The cough detection program according to the present invention captures a subject's voice, processes a voice data obtained by digitizing the captured voice, and obtains an envelope by obtaining an envelope using an A / D converted digital value. A pulse detection means for detecting the current, a Fourier transform means for performing a Fourier transform on the detected pulse, a cepstrum generation means for generating a cepstrum using the Fourier transform result, and a cepstrum generated by the cepstrum generation means. It is compared with a threshold value, and when it is smaller than the predetermined threshold value, it functions as a determination means for determining cough.

  In the cough detection program according to the present invention, the pulse detection means includes pulse selection means for selecting a pulse whose amplitude and pulse width are greater than or equal to a predetermined value.

  In the cough detection program according to the present invention, the cepstrum generation means includes logarithm calculation means for performing a logarithmic operation on the Fourier transform result and an inverse Fourier transform means for performing an inverse Fourier transform on the output of the logarithmic operation means. And

  The cough detection program according to the present invention is characterized by comprising a predetermined threshold generation means for obtaining a predetermined threshold used by the determination means.

  According to the present invention, using the Fourier transform result, a cepstrum that demonstrates the distinctive power of voiced sound and unvoiced sound is generated, and the generated cepstrum is compared with a predetermined threshold value. Since it is determined, the cough is an unvoiced sound and can be determined appropriately, and the detection accuracy of the cough can be improved.

  In the present invention, since a pulse having an amplitude and a pulse width exceeding a predetermined value is selected, noise and unnecessary wave components are appropriately removed to ensure highly accurate cough detection.

  In the present invention, in the cepstrum generation, the logarithmic operation is performed on the result of the Fourier transform, and the output of the logarithmic operation is inversely Fourier transformed. These components can be separated and cough detection can be performed with high accuracy.

  In the present invention, since the predetermined threshold value generation means for obtaining the predetermined threshold value used by the determination means is provided, the predetermined threshold value can be generated to enable appropriate cough detection, which can lead to highly accurate cough detection. it can.

  Embodiments of a cough detection device and a cough detection program according to the present invention will be described below with reference to the accompanying drawings. In each figure, the same components are denoted by the same reference numerals, and redundant description is omitted. There are voiced sounds and unvoiced sounds. As shown in FIG. 1, the voiced sound is a signal produced by vibration of the vocal cord S that is output as an audio signal from the mouth M through a filter called a vocal tract P.

  On the other hand, as shown in FIG. 2, the unvoiced sound is a breath noise N generated in the back of the throat or the like, which is output as an audio signal from the mouth M through a filter called a vocal tract P. In general, in the case of Japanese, a voice used by a person in conversation or the like is a voiced sound or a voiced sound followed by a voiced sound. In contrast, a cough is composed only of unvoiced sounds.

  In view of the above, the voice used for conversation or the like includes a voiced sound and thus has a strong spectrum, whereas the cough is composed only of an unvoiced sound and does not have a spectrum. By detecting this difference, cough detection is performed.

  In FIG. 3, the block diagram which concerns on the Example of a cough detection apparatus is shown. The cough detection apparatus according to this embodiment includes a microphone 11, an amplifier 12, an A / D converter 13, and a computer 14 that constitute a voice capturing unit. The microphone 11 takes the subject's voice, converts it into a voice signal, and outputs it. The amplifier 12 amplifies the output of the microphone 11 to a level at which signal processing is possible and sends it to the A / D converter 13.

  The A / D converter 13 digitizes the audio signal at, for example, 8 KHz, and the computer 14 captures the digital value every 0.5 seconds, for example, and processes it. For example, as shown in FIG. 4, the computer 14 is configured around a CPU 20, and further includes a display unit 21, an input unit 22 such as a keyboard and a mouse, a storage unit 23, and a port connected to the A / D converter 13. 24.

  The CPU 20 performs the cough detection process by executing the cough detection program and realizing the means shown in FIG. That is, a pulse detection unit 30, a Fourier transform unit 40, a cepstrum generation unit 50, a determination unit 60, and a predetermined threshold generation unit 70 are provided.

  The pulse detection means 30 detects a pulse by obtaining an envelope using the digital value converted by the A / D converter 13 described above. Specifically, the pulse detection unit 30 includes an envelope creation unit 31, an audio pulse detection unit 32, and a pulse selection unit 33. The envelope creation means 31 creates an envelope so that line segments corresponding to digital values subjected to A / D conversion are arranged in time series and the vertices are connected. The sound pulse detecting means 32 detects, as a pulse, a part where the envelope created by the envelope creating means 31 is convex upward. The pulse selection means 33 selects and outputs a pulse whose amplitude and pulse width are greater than or equal to a predetermined value for the pulse detected by the audio pulse detection means 32.

  The Fourier transform means 40 performs Fourier transform on the pulses detected by the pulse detection means 30. As a result of the Fourier transform by the Fourier transform means 40, the spectrum of the pulses detected by the pulse detection means 30 is obtained.

  The cepstrum generation unit 50 generates a cepstrum using the Fourier transform result, and includes a logarithmic operation unit 51 that performs logarithmic operation on the Fourier transform result, an inverse Fourier transform unit 52 that performs inverse Fourier transform on the output of the logarithmic operation unit 51, and It has. Here, if the sound component is A (ω) and the transfer function by the airway is H (ω), the output sound (including voiced and unvoiced sound) is A (ω) H (ω). The Fourier-transformed spectrum is also in the form of multiplication, but by logarithmic calculation it becomes a summation form like logA (ω) + logH (ω), and this is inverse Fourier transform to obtain the cepstrum, thereby separating the components It is easy to remove, and low frequency H (ω) is removed.

  About the cepstrum obtained as mentioned above, the determination means 60 compares with a predetermined threshold value, and when it is smaller than the said predetermined threshold value, it determines with a cough. The predetermined threshold value is created and given by the predetermined threshold value generation means 70.

  The operation of the cough detection apparatus configured as described above will be described based on the flowchart shown in FIG. When the processing of the cough detection program is started, the CPU 20 takes in a digital value related to the A / D conversion result, creates an envelope having an amplitude corresponding to the digital value, and generates a pulse by this envelope shape. A pulse having a predetermined size (amplitude and pulse width) or more is selected (S11).

  Next, a Fourier transform is performed on the pulse selected in step S11 to obtain a spectrum (S12). For example, an analog waveform shown in FIG. 7A is obtained by the occurrence of “A”, and the spectrum corresponding to this is shown in FIG. 8B. On the other hand, when a cough is captured with a macrophone, the analog waveform shown in FIG. 8A is obtained, and the spectrum corresponding to this is as shown in FIG. 8B.

  When the spell spectrum is obtained, a logarithmic operation is performed on the spell spectrum, and a cepstrum is obtained by performing inverse Fourier transform as a summation format (S13). FIG. 7C shows a cepstrum corresponding to “A”, and FIG. 8C shows a cepstrum corresponding to cough. As is clear from the comparison between FIG. 7C and FIG. 8C, a spike having a large amplitude appears in the cepstrum corresponding to “A”, but no spike appears in the cepstrum corresponding to cough.

  The cepstrum as described above is compared with a predetermined threshold value. If the cepstrum is smaller than the predetermined threshold value, it is determined as cough. For example, the cough frequency count value is increased by 1 (S14), and it is detected whether an end instruction is issued. If the end instruction is not issued, the process returns to step S11 to continue the process.

  An observation experiment was performed on nine subjects using the cough detection apparatus and the cough detection program that perform the above processing. FIG. 9 shows the experimental results. The division in the horizontal axis direction is the division of each person, and no. 1 to no. Nine results up to 9 are shown. The vertical axis direction represents the amplitude value of the pitch component of the cepstrum as a relative value. As shown in the figure, the maximum amplitude value of the pitch component of the cepstrum is plotted for cough, vowel, plosive and friction sound. For these nine people, it was found that by performing the determination in step S14 with the predetermined threshold value being “0.05”, it is possible to accurately discriminate between cough and voice except for the “no.

  Similarly, FIG. 10 shows the results of observation experiments conducted on 130 subjects (7 males and 2 females) using a total of 130 data acquisitions using the cough detection device and the cough detection program. According to the results of this experiment, the number of times that cough was mistakenly determined as a voice was 1 (no. 3), the number of times that voice was incorrectly determined as cough was 0, and a total of 130 times was determined for 130 times. Judgment of correct times was made, and for seats, 44 out of 45 times were correct.

  The overall discrimination rate, which is a ratio of “the number of correctly discriminated data” divided by “the total number of data”, is 99.2%, and “the number of detected coughs” is set to “true cough”. The detection rate, which is the ratio divided by the “number of”, is 97.8%, which is a significant improvement in accuracy compared to the conventional example.

  Next, the predetermined threshold generation means 70 will be described. The predetermined threshold generation means 70 obtains the output of the cepstrum generation means 50. For example, as shown in FIG. 9, the predetermined threshold value generation unit 70 obtains a plurality of coughs and a plurality of voice cepstrum and obtains the maximum amplitude of the pitch component for each subject or a plurality of subjects, and copes with coughing. And the average value of the maximum amplitude values corresponding to speech. The predetermined threshold generation means 70 gives an intermediate value of these average values to the determination means 60 as a predetermined threshold. Alternatively, the predetermined threshold generation means 70 obtains the largest value among the maximum amplitude values corresponding to cough and the smallest value among the maximum amplitude values corresponding to speech. An intermediate value of these average values (or the value of the point divided into 3: 2) is detected by the predetermined threshold value generation means 70 and given to the determination means 60 as the predetermined threshold value.

  In any case, the cepstrum of a plurality of coughs and a plurality of voices is obtained, the maximum amplitude of the pitch components is obtained and distributed as shown in FIG. 9, and the distribution of the maximum amplitudes of the pitch components of the cough and the voices is separated. The predetermined threshold value generation means 70 may obtain a possible value (a value in which both values do not exist).

The figure which shows the principle of voiced sound generation. The figure which shows the principle of unvoiced sound generation. The schematic block diagram of the cough detection apparatus of the Example of this invention. The principal part block diagram of the cough detection apparatus of the Example of this invention. The functional block diagram of the cough detection apparatus of the Example of this invention. The flowchart which shows operation | movement of the cough detection apparatus of the Example of this invention. The figure which shows the process processed by the cough detection apparatus of the Example of this invention about wired sound "A". The figure which shows the process processed by the cough detection apparatus of the Example of this invention about cough. The figure which shows the experimental result regarding the determination of a cough and an audio | voice by the cough detection apparatus of the Example of this invention. The figure which shows the experimental result regarding the determination of a cough and an audio | voice by the cough detection apparatus of the Example of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 11 Microphone 12 Amplifier 13 Converter 14 Computer 21 Display part 22 Input part 23 Memory | storage part 24 Port 30 Pulse detection means 31 Envelope creation means 32 Audio | voice pulse detection means 33 Pulse selection means 40 Fourier transform means 50 Cepstrum generation means 51 Logarithm calculation means 52 Inverse Fourier transform means 60 Determination means 70 Predetermined threshold generation means

Claims (8)

Voice capturing means for capturing the subject's voice;
An A / D converter for digitizing the voice captured by the voice capturing means;
Pulse detection means for detecting a pulse by obtaining an envelope using a digital value converted by the A / D converter;
Fourier transform means for performing Fourier transform on the detected pulse;
Cepstrum generation means for generating a cepstrum using the Fourier transform result;
A cough detection apparatus comprising: a determination unit that compares the cepstrum generated by the cepstrum generation unit with a predetermined threshold value and determines cough when the cepstrum is smaller than the predetermined threshold value. 2. The cough detection device according to claim 1, wherein the pulse detection means includes pulse selection means for selecting a pulse having an amplitude and a pulse width that are greater than or equal to a predetermined value. The cepstrum generation means includes logarithm calculation means for performing logarithmic operation on the Fourier transform result, and inverse Fourier transform means for performing inverse Fourier transform on the output of the logarithmic calculation means. Cough detection device. The cough detection device according to claim 1, further comprising a predetermined threshold generation unit that calculates a predetermined threshold used by the determination unit. A computer that captures the subject's voice and processes the voice data obtained by digitizing the captured voice.
Pulse detection means for detecting a pulse by obtaining an envelope using a digital value subjected to A / D conversion;
Fourier transform means for performing Fourier transform on the detected pulse;
Cepstrum generation means for generating a cepstrum using the Fourier transform result;
A cough detection program for comparing a cepstrum generated by a cepstrum generation unit with a predetermined threshold value and functioning as a determination unit for determining cough when the cepstrum is smaller than the predetermined threshold value. 6. The cough detection program according to claim 5, wherein the pulse detection means includes pulse selection means for selecting a pulse having an amplitude and a pulse width that are greater than or equal to a predetermined value. 7. The cepstrum generating means comprises logarithmic arithmetic means for performing logarithmic computation on the Fourier transform result, and inverse Fourier transform means for performing inverse Fourier transform on the output of the logarithmic arithmetic means. Cough detection program. The cough detection program according to any one of claims 5 to 7, further comprising a predetermined threshold value generation unit that calculates a predetermined threshold value used by the determination unit.

Images