WO2017135127A1 - Bioacoustic extraction device, bioacoustic analysis device, bioacoustic extraction program, and computer-readable storage medium and stored device - Google Patents

Abstract

Provided is a bioacoustic extraction device (100) for precisely extracting necessary bioacoustic data from raw acoustic data that includes the bioacoustic data. The bioacoustic extraction device (100) comprises: an input unit (10) for acquiring raw acoustic data that include bioacoustic data; a sound section estimation unit (20) that estimates a sound section, from the raw acoustic data input from the input unit (10); an auditory image generation unit (30) that generates an auditory image in accordance with an auditory image model, on the basis of the sound section estimated by the sound section estimation unit (20); an acoustic feature amount extraction unit (40) that extracts an acoustic feature amount in respect to the auditory image generated by the auditory image generation unit (30); a classification unit (50) that classifies, into a prescribed type, the acoustic feature amount extracted by the acoustic feature amount extraction unit (40); and a determination unit (60) that determines, on the basis of a prescribed threshold value, whether the acoustic feature amount classified by the classification unit (50) is bioacoustic data.

Description

Bioacoustic extraction apparatus, bioacoustic analysis apparatus, bioacoustic extraction program, computer-readable recording medium, and recorded apparatus

The present invention relates to a bioacoustic extraction device, a bioacoustic analysis device, a bioacoustic extraction program, a computer-readable recording medium, and a recorded device.

Bioacoustic analysis is performed to analyze bioacoustics, which are sounds generated by humans, and to determine and analyze cases of diseases and diseases. In performing such bioacoustic analysis, the analysis target acoustic data includes only bioacoustic data. In other words, acoustics other than bioacoustics such as noise are excluded, and necessary acoustic data is obtained. Extraction work is required. If noise is included, it will affect the accuracy of case analysis, judgment, and diagnosis. However, even if the original acoustic data is removed together with the noise, it will also affect the reliability of judgment results and so on. Therefore, in the bioacoustic analysis, it is required to accurately select only bioacoustic data. Conventionally, such work has been performed manually by hand, which is a heavy burden, and there is a need for a system that can automatically and accurately extract only necessary bioacoustic data from acoustic data. However, a method capable of automatic extraction with practical accuracy has not yet been established.

Snoring sounds are taken up as an example of bioacoustics. In recent years, Sleep Apnea Syndrome (SAS) has attracted attention, which is accompanied by constant apnea or hypopnea during sleep, excessive daytime sleepiness, feeling of suffocation and panting during sleep, repetitiveness It is a disease that causes symptoms such as awakening. Obstructive 無 Sleep Apnea 特 に Syndrome (OSAS), in particular, has been pointed out to be associated with cardiovascular diseases such as hypertension, stroke, angina pectoris, and myocardial infarction. (For example, Patent Document 1, Non-Patent Documents 1 to 4).

Currently, examination using a polysomnography (PSG) is performed overnight for diagnosis of SAS. This is a large-scale examination requiring hospitalization of the patient, which is expensive and requires the electrodes to be attached to the body all night, which places a burden on the patient's body. Specifically, since it is necessary to record the patient's nasal airflow, tracheal sound, oxygen saturation, etc., it was necessary to attach a large number of measuring devices and sensors to the patient. In addition, since the SAS simple inspection is performed during sleep, the mounting state of these measurement sensors greatly affects the measurement results. For example, problems such as the measurement sensor coming off due to the patient turning over, the wearing position being displaced, and the clothes being rubbed are picked up. In addition, the situation where a measurement sensor or the like is attached to a patient during sleep is not desirable due to physical and mental burdens and pain. For this reason, realization of a simpler inspection method is expected. As one approach, attention has recently been focused on methods based on acoustic analysis of snoring sounds.

However, in the past research on snoring sounds, the work of extracting the snoring episodes of interest from the recorded data during sleep (Sleep Related ： Sound: SRS, hereinafter referred to as “sleep-related sounds”) was performed manually. Therefore, the burden at the time of data collection is large, and as a result, only relatively small snoring sound data can be analyzed.

In order to pray for snoring sounds as a diagnostic technique, it is necessary to pay attention to snoring sounds recorded over a long time while sleeping. To that end, it is essential to automate the work of extracting snoring episodes that have been performed manually.

On the other hand, it is conceivable to adopt a non-contact type microphone instead of a contact type myophone as a method for collecting the snoring sound of the patient without burden. However, since the distance between the patient and the microphone is relatively larger than that of the contact-type microphone due to the non-contact, the volume of the snoring sound (the amplitude value of the acoustic spectrum) is reduced, and sleep, cough, The signal-to-noise ratio (SNR) is relatively high due to the increase in noise components such as breathing and other sounds generated by patients other than snoring sounds and noises caused by non-patients such as bed squeaks and metal sounds. There is a concern that it will get worse. Therefore, when actually performing the bioacoustic analysis, it is necessary to remove the noise component as a pre-processing. However, it is not possible to accurately extract the necessary acoustic data such as the snoring sound from the acoustic data whose SNR has deteriorated. It was extremely difficult and a practical method was required.

JP 2014-166568 A U.S. Pat. No. 8,880,207 US Patent Application Publication No. 2015-0039110

The present invention has been made to solve such conventional problems. The main objects of the present invention are a bioacoustic extraction apparatus, a bioacoustic analysis apparatus, a bioacoustic extraction program, a computer-readable recording medium, and a computer-readable recording medium that can accurately extract necessary bioacoustic data from acoustic data including bioacoustics. To provide recorded equipment.

Means for Solving the Problems and Effects of the Invention

To achieve the above object, the bioacoustic extraction apparatus according to the first aspect of the present invention is a bioacoustic extraction apparatus for extracting necessary bioacoustic data from original acoustic data including bioacoustic data. An input unit for acquiring original acoustic data including bioacoustic data, a voiced section estimating unit for estimating a voiced section from the original acoustic data input from the input unit, and the voiced section estimating unit An auditory image generation unit that generates an auditory image according to an auditory image model based on the sounded section estimated in step S1, and an acoustic feature amount that extracts an acoustic feature amount from the auditory image generated by the auditory image generation unit An extraction unit, a classification unit that classifies the acoustic feature amount extracted by the acoustic feature amount extraction unit into a predetermined type, and a biological body based on a predetermined threshold with respect to the acoustic feature amount classified by the classification unit Determine whether it is acoustic data It may include a that discrimination unit. With the above configuration, noise and necessary bioacoustic data can be accurately discriminated by converting the original acoustic data into an auditory image using an auditory image model and then classifying it based on the acoustic feature amount.

Further, according to the bioacoustic extraction apparatus according to the second aspect of the present invention, the auditory image generation unit is configured to generate a stabilized auditory image using an auditory image model, and the acoustic feature amount extraction The unit can extract the acoustic feature amount based on the stabilized auditory image generated by the auditory image generator.

Furthermore, according to the bioacoustic extraction device according to the third aspect of the present invention, the auditory image generation unit is configured to further generate a generalized stabilized auditory image and an auditory spectrum from the stabilized auditory image. The acoustic feature amount extraction unit can extract an acoustic feature amount based on the overall stabilized auditory image generated by the auditory image generation unit and the auditory spectrum.

Furthermore, according to the bioacoustic extraction device according to the fourth aspect of the present invention, the acoustic feature quantity extraction unit includes the kurtosis, skewness, spectrum centroid, spectrum band of the auditory spectrum and / or the overall stabilized auditory image. At least one of width, spectrum w flatness, spectrum roll-off, spectrum entropy, and octave-based spectrum contrast can be extracted as an acoustic feature amount.

Furthermore, according to the bioacoustic extraction apparatus according to the fifth aspect of the present invention, the auditory image generation unit is configured to generate a neural activity pattern using an auditory image model, and the acoustic feature amount extraction The unit can extract the acoustic feature amount based on the neural activity pattern generated by the auditory image generation unit. With the above configuration, it is possible to reduce the processing load and improve the processing speed as compared with the case where a stabilized auditory image is used. The acoustic feature quantity extraction unit can also extract at least one of the total number of peaks, the appearance position, the amplitude, the center of gravity, the inclination, the increase, and the decrease as the acoustic feature quantity obtained from the acoustic spectrum.

Furthermore, the bioacoustic extraction device according to the sixth aspect of the present invention is a bioacoustic extraction device for extracting necessary bioacoustic data from original acoustic data including bioacoustic data. An input unit for acquiring original sound data including data, a sound interval estimation unit for estimating a sound interval from the original sound data input from the input unit, and the sound interval estimation unit An auditory image generator that generates an auditory image according to an auditory image model based on a voiced section, an auditory spectrum generator that generates an auditory spectrum for the auditory image generated by the auditory image generator, and an auditory image On the other hand, a generalized stabilized auditory image generating unit that generates a generalized stabilized auditory image, an auditory spectrum generated by the auditory spectrum generating unit, and a generalized stable auditory image generated by the generalized stabilized auditory image generating unit An acoustic feature amount extraction unit that extracts an acoustic feature amount from an auditory image, a classification unit that classifies the acoustic feature amount extracted by the acoustic feature amount extraction unit into a predetermined type, and an acoustic classified by the classification unit The feature amount can include a determination unit that determines whether the feature amount is bioacoustic data based on a predetermined threshold.

Furthermore, according to the bioacoustic extraction apparatus according to the seventh aspect of the present invention, it is possible to extract a section having a period from the original acoustic data.

Furthermore, according to the bioacoustic extraction apparatus according to the eighth aspect of the present invention, the sounded section estimation unit performs preprocessing by differentiating or subtracting original sound data, and the preprocessing. A squarer for squaring the preprocessed data pre-processed by the detector, a downsampler for downsampling the squared data squared by the squarer, and the downsampled data downsampled by the downsampler And a median filter for obtaining a median value from.

Furthermore, according to the bioacoustic extraction device according to the ninth aspect of the present invention, the input unit can be a non-contact microphone that is installed in a non-contact manner with the patient to be examined.

Furthermore, according to the bioacoustic extraction device according to the tenth aspect of the present invention, the discrimination of the bioacoustic data by the discrimination unit can be performed as non-language processing. As described above, based on the auditory image model in processing of cases and diseases for bioacoustic data such as snoring sounds and intestinal sounds, instead of conventional processing on speech signals, for example, language processing such as speaker identification and speech recognition. Can be applied widely, regardless of language.

Furthermore, according to the bioacoustic extraction device according to the eleventh aspect of the present invention, the original acoustic data is a bioacoustic acquired when the patient sleeps, and is necessary from the bioacoustic data acquired under sleep. Bioacoustic data can be extracted.

Furthermore, according to the bioacoustic extraction device according to the twelfth aspect of the present invention, the original acoustic data is sleep-related sounds collected during sleep of the patient, and the bioacoustic data is snoring sound data. The predetermined type can be classified into a snoring sound and a non-snoring sound.

Furthermore, according to the bioacoustic analyzer according to the thirteenth aspect of the present invention, there is provided a bioacoustic analyzer for extracting and analyzing necessary bioacoustic data from original acoustic data including bioacoustic data. An input unit for acquiring original acoustic data including bioacoustic data, a voiced section estimating unit for estimating a voiced section from the original acoustic data input from the input unit, and the voiced section estimating unit An auditory image generation unit that generates an auditory image according to an auditory image model based on the estimated voiced section, and an acoustic feature amount extraction that extracts an acoustic feature amount from the auditory image generated by the auditory image generation unit And a classification unit that classifies the acoustic feature amount extracted by the acoustic feature amount extraction unit into a predetermined type, and the acoustic feature amount classified by the classification unit based on a predetermined threshold Determine if it is data A determination section, the true value data determined with the biological sound data in the determination unit may include a screening unit for performing screening. With the above configuration, the original acoustic data is converted into an auditory image using an auditory image model and then classified based on the acoustic feature amount, whereby noise and necessary bioacoustic data can be accurately distinguished.

Furthermore, according to the bioacoustic analyzer according to the fourteenth aspect of the present invention, the bioacoustic analyzer for extracting and analyzing necessary bioacoustic data from the original acoustic data including the bioacoustic data is provided. An input unit for acquiring original acoustic data including bioacoustic data, a voiced section estimating unit for estimating a voiced section from the original acoustic data input from the input unit, and the voiced section estimating unit An auditory image generator that generates an auditory image according to an auditory image model based on the estimated voiced section; an auditory spectrum generator that generates an auditory spectrum for the auditory image generated by the auditory image generator; A generalized stabilized auditory image generating unit that generates a generalized stabilized auditory image for the auditory image, an auditory spectrum generated by the auditory spectrum generating unit, and a generalized auditory image generating unit. An acoustic feature amount extraction unit that extracts an acoustic feature amount from the generalized stabilized auditory image, a classification unit that classifies the acoustic feature amount extracted by the acoustic feature amount extraction unit into a predetermined type, and a classification by the classification unit A discriminating unit that discriminates whether or not the acoustic feature quantity is bioacoustic data based on a predetermined threshold value, and screening that performs screening on true value data discriminated as bioacoustic data by the discriminating unit A portion.

Furthermore, according to the bioacoustic analyzer of the fifteenth aspect of the present invention, the screening unit can be configured to perform disease screening on bioacoustic data extracted from the original acoustic data.

Furthermore, according to the bioacoustic analyzer of the sixteenth aspect of the present invention, the screening unit performs screening for obstructive sleep apnea syndrome on the bioacoustic data extracted from the original acoustic data. Can be configured.

Furthermore, according to the bioacoustic extraction method according to the seventeenth aspect of the present invention, there is provided a bioacoustic extraction method for extracting necessary bioacoustic data from original acoustic data including bioacoustic data. A step of acquiring original sound data including data, a step of estimating a sound section from the acquired original sound data, and generating an auditory image according to an auditory image model based on the estimated sound section A step of extracting an acoustic feature amount from the generated auditory image, a step of classifying the extracted acoustic feature amount into a predetermined type, and the classified acoustic feature amount. And determining whether or not the data is bioacoustic data based on a predetermined threshold value. Thereby, after converting the original acoustic data into an auditory image using an auditory image model and then classifying based on the acoustic feature amount, it is possible to accurately determine noise and necessary bioacoustic data.

Furthermore, according to the bioacoustic extraction method according to the eighteenth aspect of the present invention, there is provided a bioacoustic extraction method for extracting necessary bioacoustic data from original acoustic data including bioacoustic data. A step of acquiring original sound data including data, a step of estimating a sound section from the acquired original sound data, and a stabilized auditory image according to an auditory image model based on the estimated sound section. A step of generating, a step of generating an overall stabilized auditory image from the stabilized auditory image, a step of extracting a predetermined acoustic feature obtained from the generated overall stabilized auditory image, and the extracted A step of determining whether or not the acoustic feature value is bioacoustic data based on a predetermined threshold value.

Furthermore, according to the bioacoustic extraction method of the nineteenth aspect of the present invention, in the step of generating an auditory spectrum from the stabilized auditory image and extracting the predetermined acoustic feature amount, the overall stabilization In addition to the auditory image, a predetermined acoustic feature amount obtained from the generated auditory spectrum can be extracted.

Furthermore, according to the bioacoustic extraction method according to the twentieth aspect of the present invention, the acoustic features that contribute to the identification from the extracted acoustic feature amounts prior to the step of extracting the predetermined acoustic feature amount. A step of selecting an amount can be included.

Furthermore, according to the bioacoustic extraction method according to the twenty-first aspect of the present invention, the step of determining whether or not the bioacoustic data is a classification of a snoring sound or a non-snoring sound using a multinomial distribution logistic regression analysis it can.

Furthermore, the bioacoustic analysis method according to the twenty-second aspect of the present invention is a bioacoustic analysis method for extracting and analyzing necessary bioacoustic data from original acoustic data including bioacoustic data. A step of acquiring original sound data including bioacoustic data; a step of estimating a sound section from the acquired original sound data; and stabilization according to an auditory image model based on the estimated sound section. A step of generating an auditory image, a step of generating an auditory spectrum and an overall stabilized auditory image from the stabilized auditory image, and a predetermined acoustic feature obtained from the generated auditory spectrum and the overall stabilized auditory image. A step of extracting, a step of determining whether or not the extracted acoustic feature quantity is bioacoustic data based on a predetermined threshold value, and a step of determining the bioacoustic data in the determining step. And the true value data may include the step of screening.

Furthermore, according to the bioacoustic analysis method according to the twenty-third aspect of the present invention, the step of performing the screening comprises an obstructive sleep apnea syndrome or a non-obstructive sleep apnea using a multinomial logistic regression analysis. Can be screened for syndrome.

The bioacoustic extraction program according to the twenty-fourth aspect of the present invention is a bioacoustic extraction program for extracting necessary bioacoustic data from original acoustic data including bioacoustic data, An input function for acquiring original sound data including data, a sound section estimation function for estimating a sound section from the original sound data input by the input function, and the sound section estimation function An auditory image generation function for generating an auditory image according to an auditory image model based on a voiced section; an acoustic feature amount extraction function for extracting an acoustic feature amount from the auditory image generated by the auditory image generation function; A classification function for classifying the acoustic feature quantity extracted by the acoustic feature quantity extraction function into a predetermined type, and a biological sound based on a predetermined threshold with respect to the acoustic feature quantity classified by the classification function A discrimination function of discriminating whether the data or not can be realized on the computer. With the above configuration, the original acoustic data is converted into an auditory image using an auditory image and then classified based on the acoustic feature amount, whereby noise and necessary bioacoustic data can be accurately distinguished.

Furthermore, according to the bioacoustic extraction program according to the twenty-fifth aspect of the present invention, there is provided a bioacoustic extraction program for extracting necessary bioacoustic data from original acoustic data including bioacoustic data, An input function for acquiring original sound data including data, a sound section estimation function for estimating a sound section from the original sound data input by the input function, and the sound section estimation function A stabilized auditory image generation function that generates a stabilized auditory image according to an auditory image model based on a voiced section, a function that generates a general stabilized auditory image from the stabilized auditory image, and the generated general stable image Function for extracting a predetermined acoustic feature amount from the auditory auditory image, and a classification function for classifying the predetermined acoustic feature amount extracted by the acoustic feature amount extraction function into a predetermined type , To the acoustic feature quantity classified in the classifier, and a determination function of determining whether the biometric acoustic data can be implemented in a computer based on a predetermined threshold value.

The bioacoustic analysis program according to the twenty-sixth aspect of the present invention is a bioacoustic analysis program for extracting and analyzing necessary bioacoustic data from original acoustic data including bioacoustic data. An input function for acquiring original sound data including bioacoustic data, a sound section estimation function for estimating a sound section from the original sound data input by the input function, and the sound section estimation function A stabilized auditory image generation function that generates a stabilized auditory image according to an auditory image model based on the estimated voiced section, a function that generates a generalized stabilized auditory image from the stabilized auditory image, and the generated The acoustic feature extraction function for extracting a predetermined acoustic feature from the overall stabilized auditory image, and the predetermined acoustic feature extracted by the acoustic feature extraction function are classified into predetermined types. A discrimination function for discriminating whether or not it is bioacoustic data based on a predetermined threshold with respect to the acoustic features classified by the classification function, and a true value discriminated as bioacoustic data by the discrimination function The function of screening data can be realized by a computer.

Furthermore, a computer-readable recording medium or recorded device according to the twenty-seventh aspect of the present invention stores the above program. CD-ROM, CD-R, CD-RW, flexible disk, magnetic tape, MO, DVD-ROM, DVD-RAM, DVD-R, DVD + R, DVD-RW, DVD + RW, Blu-ray (registered) (Trademark) and other magnetic disks, optical disks, magneto-optical disks, semiconductor memories, and other media that can store programs. The program includes a program distributed in a download manner through a network line such as the Internet, in addition to a program stored and distributed in the recording medium. Further, the recording medium includes a device capable of recording the program, for example, a general purpose or dedicated device in which the program is implemented in a state where the program can be executed in the form of software, firmware, or the like. Furthermore, each process and function included in the program may be executed by computer-executable program software, or each part of the process or hardware may be executed by hardware such as a predetermined gate array (FPGA, ASIC), or program software. And a partial hardware module that realizes a part of hardware elements may be mixed.

It is a block diagram which shows the bioacoustic extraction apparatus which concerns on one embodiment of this invention. It is a block diagram which shows an example of a sound area estimation part. It is a flowchart which shows the bioacoustic extraction method using AIM which concerns on this Embodiment. It is a block diagram which shows the process of AIM. It is an image figure which shows an auditory image. 6A and 6B are graphs showing the relationship between the feature vector index and accuracy. It is a graph which shows the ROC analysis result at the time of using AIMF _opt . It is a flowchart which shows an example of the estimation method of a sound area. 9A is the original acoustic data, FIG. 9B is the preprocessed data, FIG. 9C is the square data, FIG. 9D is the down-sampling data, and FIG. 9E is a graph showing the median waveform. FIG. 10A is the original acoustic data, FIG. 10B is the ZCR processing result according to Comparative Example 1, FIG. 10C is the STE processing result according to Comparative Example 2, and FIG. 10D is the waveform of the voiced section estimation processing result according to Example 1. It is a graph which shows.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the embodiments described below exemplify a bioacoustic extraction device, a bioacoustic analysis device, a bioacoustic extraction program, a computer-readable recording medium, and a recorded device for embodying the technical idea of the present invention. In the present invention, the bioacoustic extraction device, the bioacoustic analysis device, the bioacoustic extraction program, the computer-readable recording medium, and the recorded device are not specified as follows. Further, the present specification by no means specifies the members shown in the claims to the members of the embodiments. In particular, the dimensions, materials, shapes, relative arrangements, and the like of the components described in the embodiments are not intended to limit the scope of the present invention only unless otherwise specified, and are merely illustrative examples. Only. Note that the size, positional relationship, and the like of the members shown in each drawing may be exaggerated for clarity of explanation. Furthermore, in the following description, the same name and symbol indicate the same or the same members, and detailed description thereof will be omitted as appropriate. Furthermore, each element constituting the present invention may be configured such that a plurality of elements are constituted by the same member and the plurality of elements are shared by one member, and conversely, the function of one member is constituted by a plurality of members. It can also be realized by sharing.
(Bioacoustic extraction apparatus 100)

Hereinafter, as an example of a bioacoustic extraction apparatus, a bioacoustic extraction apparatus that automatically extracts snoring sound as ecological acoustic data to be extracted from sleep-related sound as original acoustic data will be described. A bioacoustic extraction apparatus according to an embodiment of the present invention is shown in the block diagram of FIG. The bioacoustic extraction apparatus 100 shown in this figure includes an input unit 10, a sound section estimation unit 20, an auditory image generation unit 30, an acoustic feature amount extraction unit 40, a classification unit 50, and a determination unit 60.

The input unit 10 is a member for acquiring original acoustic data including bioacoustic data. The input unit 10 includes a microphone unit and a preamplifier unit, and inputs the collected original sound data to a computer constituting the bioacoustic extraction device 100. A non-contact microphone that is preferably installed in a non-contact manner with the patient to be examined can be used for the microphone section.

The voiced section estimation unit 20 is a member for estimating a voiced section from the original acoustic data input from the input unit 10. As shown in the block diagram of FIG. 2, the voiced section estimation unit 20 performs pre-processing by differentiating or subtracting the original sound data and pre-processing data pre-processed by the pre-processing unit 21. To obtain a median value from the downsampled data downsampled by the downsampler 23, a downsampler 23 for downsampling the squared data squared by the squarer 22, Median filter 24.

The auditory image generation unit 30 is a member for generating an auditory image according to the established auditory image model (AIM) based on the voiced section estimated by the voiced section estimation unit 20.

The acoustic feature amount extraction unit 40 is a member for extracting feature amounts from the auditory image generated by the auditory image generation unit 30. The acoustic feature amount extraction unit 40 is generated by synchronously adding an auditory spectrum (AS) generated by synchronously adding a stabilized auditory image (Stabilized auditory image: SAI) in the horizontal axis direction and SAI in the vertical axis direction. The feature amount can be extracted based on the generalized stabilized auditory image (SSAI). Specifically, the acoustic feature quantity extraction unit 40 extracts at least one of kurtosis, distortion, spectrum centroid, spectrum bandwidth, spectrum flatness, spectrum roll-off, spectrum entropy, and OSC of the auditory spectrum as a feature quantity. . The acoustic feature quantity extraction unit 40 can also extract at least one of the total number of peaks, the appearance position, the amplitude, the center of gravity, the inclination, the increase, and the decrease as the feature quantity obtained from the acoustic spectrum.

The classification unit 50 is a member for classifying the feature amount extracted by the acoustic feature amount extraction unit 40 into a predetermined type.

The discriminating unit 60 is a member for discriminating whether or not the feature quantity classified by the classifying unit 50 is bioacoustic data based on a predetermined threshold value.

In this way, it is possible to automatically extract a snoring sound with high accuracy by constructing the bioacoustic extraction device 100 that simulates from the human auditory pathway to the learning mechanism.
(Bioacoustic analyzer 110)

Furthermore, a bioacoustic analysis device for analyzing bioacoustic data extracted by the bioacoustic extraction device can also be configured. The bioacoustic analysis apparatus 110 further includes a screening unit 70 that performs screening on true value data determined by the determination unit 60 as bioacoustic data.

The bioacoustic extraction device and the bioacoustic analysis device described above can be implemented as software by a program in addition to being configured by dedicated hardware. For example, by installing a bioacoustic extraction program or a bioacoustic analysis program on a general-purpose or dedicated computer, loading or downloading and executing it, a virtual bioacoustic extraction device or bioacoustic analysis device is realized. You can also.
(Acoustic analysis of conventional snoring sound)

In recent years, acoustic analysis of snoring sound using a non-contact microphone has been performed. In those studies, since it is necessary to extract only the snoring sound from the sound recorded during sleep (sleep-related sound), various automatic snoring sound extraction methods have been proposed. For example,
(I) a method using a network in which Mel-frequency cepstral coefficients (MFCC) and a hidden Markov model (HMM) are interconnected;
(Ii) Subband spectral energy, a method using robust linear regression (RLR) or principal component analysis (PCA),
(Iii) A method using subband energy distribution, PCA, unsupervised Fuzzy C-Means (FCM) clustering,
(Iv) 34 feature amounts combining a plurality of acoustic analysis methods and a method using Ada Boost have been proposed.

According to the report example regarding these methods, it is shown that the snoring sound and the non-snoring sound can be automatically classified with high accuracy. However, in these methods, it is necessary to extract feature amounts using several signal processing techniques. In general, the performance evaluation of sound classification methods is based on manual classification, which is considered a gold standard technique, that is, classification of manual work by human ears. Therefore, the present inventors have thought that a high-performance sound classifier can be constructed by imitating human hearing ability, and have achieved the present invention. Specifically, a bioacoustic extraction device that can automatically classify snoring sounds / non-snoring sounds using an auditory image model (AIM) has been achieved. AIM is an auditory image model that imitates an “auditory image” that is considered to be an expression in the brain that humans use to perceive sound. Specifically, AIM is a model of an auditory image that simulates the function from the peripheral system of the human auditory system including the cochlear basement membrane to the central system. Such AIM has been established mainly in research on hearing and spoken language perception, and is used in the field of speaker recognition and speech recognition, but is used to discriminate bioacoustics such as snoring sounds and intestinal sounds. There are no reported examples as far as the present inventors know.
(Example)

In order to confirm the effectiveness of the present invention, confirmation was performed using a large-scale database of sleep-related sounds obtained from 40 subjects. FIG. 3 shows a flowchart of the bioacoustic extraction method using the AIM according to the present embodiment.
(Sound section estimation)

First, sleep-related sounds are collected in step S301, and then a sound section is estimated in step S302. Here, with the cooperation of Anan Kyoei Hospital (Hanaura Nakanosho-no-ho-36, Ano-city, Anan-city, Tokushima Prefecture), during an overnight polysomnography (PSG) examination, using the input unit 10 for 6 hours from the patient Recorded sleep-related sounds. The input unit 10 includes a non-contact type microphone that is a form of a microphone unit and a preamplifier unit, and collects the obtained audio data by a computer. Here, the non-contact type microphone was installed at a position about 50 cm away from the patient's mouth. The microphone used for recording was Model NT3 manufactured by Australia RODE, the preamplifier was Mobile-Pre USB manufactured by M-AUDIO, USA, the sampling frequency during recording was 44.1 kHz, and the digital resolution was 16 bits / sample.

From the sleep-related sound recorded in this way, the sound section (Audio events: AE) is detected in step S302 using the sound section estimation unit 20. The voiced section estimation unit 20 uses a short-term energy method (STE) and a median filter 24. The STE method is a method of detecting signal energy equal to or higher than a certain threshold value as a sound section. Here, the k-th short-term energy Ek of the sleep-related sound s (n) can be expressed by the following equation.

In the above formula, n is the sample number and N is the segment length. In the example, the sleep-related sound s (n) was divided into segments with N = 4096 and a shift width of 1024, and the signal energy in the kth segment was calculated. A 10th-order median filter was used to smooth Ek.

Further, in the embodiment, the AE is extracted by detecting a sound having an SNR of 5 dB or more in the segment. Here, in calculating the SNR, the background noise is used as an average value of all frames of short-term energy obtained by performing the STE method from a signal of only background noise for one second.

According to Non-Patent Document 2, it is reported that in a listening experiment in which the relationship between singing voice and voice identification and sound duration is investigated, the signal length is 200 ms or more and the identification rate exceeds 70%. Accordingly, in this embodiment, a detection sound having a signal length of 200 ms or more is defined as AE.
(Generation of auditory image model)

Next, in step S303, an auditory image is generated using an auditory image model (Auditory Image Model: AIM). Here, the auditory image generation unit 30 analyzes a voiced section (AE) using AIM. As shown in Non-Patent Document 3, an AIM simulator is provided by the Patterson group. Although the simulator can operate in a C language environment, in this embodiment, AIM 2006 <http://www.pdn.cam.ac.uk/groups/cnbh/aim2006 which can be used for MATLAB. /> (Module: gm2002, dcgc, hl, sf2003, ti2003) was used. The five main stages available are Pre-cochlea processing (PCP), Basilar membrane motion (BMM), Neural activity pattern (NAP), Strobe identification (STROBES) ), Stabilized auditory image (SAI). Through these processes, the input sound can be output as an auditory image.

An example of AIM processing is shown in the block diagram of FIG. First, in the stage of pre-cochlea processing (PCP), in order to express the response characteristics of the inner ear up to the vestibule window, filter processing using a band pass filter is performed.

At the basement membrane motion (BMM) stage, filters are arranged at regular intervals, like an equivalent rectangular bandwidth (ERB), to represent the spectral analysis performed in the basement membrane of the cochlea. Auditory filter banks (gamma-chirp filter bank, gamma tone filter bank) are used. At the BMM stage, the output from each filter in the filter bank can be obtained. In the present embodiment, a gamma chirp filter bank is used in which 50 filters having different center frequencies and bandwidths are arranged for each location between 100 Hz and 6000 Hz. Note that the number of filters to be used may be adjusted as appropriate.

In the neural activity pattern (NAP) stage, the output of each filter of the BMM is low-pass filtered and half-wave rectified to represent the neural signal conversion process performed by the inner hair cells.

In the stage of strobe identification (STROBES), the maximum point in the output of each filter of NAP is detected by adaptive threshold processing.
(Stabilized auditory image: SAI)

Furthermore, at the stage of the stabilized auditory image (SAI), when a local maximum point is detected in each frequency channel, a 35 ms frame is created with the local maximum point as the origin, and a buffer storing past NAP expressions is stored. The auditory image is generated by converting the time axis into the time interval axis by integrating with the information from the time. This series of processing is called STI (Strobed temporal integration), and the auditory image can be output as SAI for each frame. STI can generate a stable auditory image by time-integrating the NAP expression over time. Therefore, in the present embodiment, the auditory spectrum (AS) and SSAI of the 10th and subsequent frames of the auditory image obtained from one episode of AE are analyzed.
(Hearing spectrum: AS)

An example of an auditory image is shown in FIG. In the auditory image shown in this figure, the vertical axis represents the center frequency axis of the auditory filter, and the horizontal axis represents the time interval axis. Here, a spectrum generated by synchronously adding auditory images in the horizontal axis direction is referred to as an auditory spectrum (AS). AS is an expression corresponding to an excitation pattern of the auditory nerve, and is a spectrum in the frequency domain where the maximum point of the formant can be confirmed. The number of AS dimensions corresponds to the number of auditory filters.
(General SAI: SSAI)

Furthermore, a spectrum generated by synchronous addition in the vertical axis direction is called a summary SAI (Summary SAI). SSAI is a time-domain spectrum that has vertices only at specific intervals because the output of each channel includes only a limited time interval when the signal is stationary and periodic. The number of dimensions of SSAI is determined by the frame size and the sampling rate of the input signal. In the present embodiment, AS and SSAI are normalized with the maximum amplitude 1 in order to minimize the influence of the signal amplitude envelope between frames.
(Acoustic features obtained from AIM)

Next, in step S304, the acoustic feature amount obtained from the AIM is extracted. Here, AS and SSAI in each SAI frame of AE can be calculated. Here, a method for extracting feature amounts from AS and SSAI will be described. Since AS and SSAI have a shape similar to a spectrum, features are extracted using the following eight types of feature amounts.

First, Kurtosis is a feature value that measures the tendency of protrusion of the spectrum per average value. The formula for kurtosis is shown below.

Next, the skewness is a characteristic amount for measuring the asymmetry of the spectrum per average value. The equation for skewness is shown below.

Furthermore, the spectral centroid is a feature quantity for calculating the spectral centroid. The equation of the spectrum centroid is shown below.

Spectral bandwidth is a feature quantity that quantifies the frequency bandwidth of a signal. The equation for the spectral bandwidth is shown below.

Spectral flatness is a feature value that quantifies sound quality. The equation for spectral flatness is shown below.

Spectral roll-off is a feature value for evaluating a frequency that occupies c × 100% of the entire band of the spectrum distribution. The equation for the spectrum roll-off is shown below.

Here, X> 0 and c = 0.95.

Spectral entropy is a feature that indicates the whiteness of the signal. The equation for spectral entropy is shown below.

Here, i is a spectrum sample point, N is a total number of spectrum sample points, k is a frame number, and X is a spectrum amplitude. However, X> 0 and c = 0.95.

Octave-based spectral contrast (OSC) is a feature quantity that represents spectral contrast. In this method, the spectrum is divided into subbands by an octave filter bank. In this embodiment, the number of subbands is set to 3 for AS and 5 for SSAI in consideration of the number of dimensions of the spectrum. The spectral peak Peak _k (b), spectral valley Valley _k (b), and spectral contrast OSC _k (b) of the b-th _subband are respectively expressed by the following equations.

Where X ′ is a feature vector rearranged in descending order within the subband, j is a sample point of the spectrum within the subband, N _b is the total number of sample points within the subband, and α is a stable peak and valley value. Represents a parameter for extracting. In this embodiment, α = 0.2. However, the spectral flatness was applied only to AS in this example because the value of SF _k approached zero as much as possible when integrating SSAI and could not be quantified.

Since the above-described feature values are extracted from all frames of AE, the average value and standard deviation of each feature value are defined as the feature values obtained from AE. That is, (i) a 20-dimensional AS feature vector, (ii) a 22-dimensional SSAI feature vector, and (iii) a 42-dimensional feature vector can be extracted from the AE. In addition to these, it is also possible to use feature quantities obtained from the spectrum, such as spectral asymmetry, band energy ratio, and the like.

In this embodiment, each feature vector is referred to as (i) ASF: Auditory spectrum features, (ii) SSAIF: Summary SAI features, and (iii) AIMF: AIM features.
(Classification of snoring / non-snoring using MLR)

Further, learning is performed based on the MLR model using the feature vector in step S306, the snoring sound / non-snoring sound classification using the MLR is performed in step S305, and the discrimination based on the threshold is performed in step S307. Here, in order to classify the acoustic feature quantity into a predetermined type by the classifying unit 50 and discriminate the snoring sound or the non-snoring sound by the discriminating unit 60, a multinomial distribution logistic regression using a feature vector extracted from the AE ( Multi-nomial logistic regression (MLR) analysis was used. The MLR analysis is an excellent statistical analysis technique as a discriminator for binary identification that classifies a plurality of measurement values into one of two categories using a logistic curve. Here, the equation of MLR is shown in the following equation.

Here, p indicates the probability that the sound to be classified is classified into the snoring sound category. Β _d = (d = 0, 1,..., D) is a parameter estimated by the maximum likelihood method. Further, f _d = (d = 0, 1,..., D) denotes a feature amount spectrum value that is regarded as an independent variable, and D denotes a dimension of the feature vector.

In the MLR analysis, a model of β _d estimated by learning based on the maximum likelihood method and a dependent variable Y is constructed. Here, Y approaches 1 (Y = 1) if there is a correlation with the snoring sound, and 0 (Y = 0) if there is a correlation with the non-snoring sound. It was confirmed that by executing the test with this model, it is possible to estimate the probability p of obtaining Y = 1 when an independent variable of the test set f _d is given. Based on the threshold value p _thre of p, each AE can be classified into one of two categories (snoring sound or non-snoring sound), and the classification of the snoring sound and the non-snoring sound can be performed by the classifier. This simulation was performed using Statistics Toolbox Version 9.0 of MATLAB (R2014a, The MathWorks, Inc., Natick, MA, USA). In the case of OSAS screening, it can be considered in the same way that it approaches 1 (Y = 1) if there is a correlation with OSAS and approaches 0 (Y = 0) if there is a correlation with non-OSAS.
(Performance evaluation of classifier using AIM)

Next, the performance of the classifier using the auditory image model (AIM) was evaluated. Here, sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) are used as indicators of classification performance. It was.

“Sensitivity” here is the ability of the discrimination result to detect snoring. Specificity is the proportion of non-snoring that results in a determination result equal to or less than a threshold value. Positive predictive value (PPV) represents the probability of actually snoring when the determination result is equal to or greater than a threshold value. Further, negative predictive value (NPV) represents the probability of non-snoring when the determination result is equal to or less than a threshold value. Based on these, the relationship between TP, FP, FN, and TN is defined as follows.

Here, the expressions indicating sensitivity, specificity, accuracy, positive appropriateness (PPV), and negative appropriateness (NPV) are expressed using TP, FP, FN, and TN, respectively. It is defined as the following formula.

(ROC curve)

The ROC (Receiver Operating Characteristic) curve is plotted with the false positive rate (1-specificity) on the horizontal axis and the true positive rate (sensitivity) on the vertical axis. In this example, the ROC curve can be constructed by P _thre . The optimum threshold value of the ROC curve, that is, the optimum P _thre can be obtained using the method of Youden's index. The ROC curve draws a large arc at the upper left in the case of an ideal classification unit. Due to this property, an area under the curve (AUC), which is the area of the lower region of the ROC curve, can be used as an index representing the performance of the classifier or the classification algorithm. The AUC value takes a value in the range of 0.5 to 1, and is a classification accuracy evaluation index having a characteristic approaching 1 when the classification accuracy is good.
(Learning dataset and test dataset)

Next, the performance evaluation of the bioacoustic extraction device according to the present embodiment was performed using AE extracted from 40 subjects. The results are shown in Table 1.

M: male; F: female; BMI: body mass index; AHI: apnea-hypopnea index

As shown in Table 1, AEs extracted from 40 people are divided into a learning data set 16141 (snoring sound 13406, non-snoring sound 2735) and a test data set 10651 (snoring sound 7346, non-snoring sound 3305). .
(labeling)

In this embodiment, the AE labeling work is performed based on the listening results. Three evaluators listened to the AE flowing from the headphones (SHURE SRH840) and labeled the AE by consensus. In this way, the snoring sound was not selected without everyone's consent.

Table 2 shows the AEs that were determined to be non-snore during such labeling work.

(Classification of snoring sound and non-snoring sound based on AIM)

Using the feature vectors ASF, SSAIF, and AIMF obtained as described above, the performance of this example was evaluated. The results are shown in Table 3. As can be seen from this table, any feature vector can be classified into snoring sounds and non-snoring sounds with high accuracy. Among them, AIMF showed the best performance. This seems to be because the human auditory sense analyzes sound using both frequency information and time information.

Generally, when the dimension number of the feature vector is high, the calculation amount becomes large. Therefore, we examined whether or not dimensional compression can be performed while maintaining high classification accuracy from the three feature vectors used. Here, in order to extract feature amounts that contribute to improvement of classification accuracy, the number of dimensions of ASF and SSAIF was increased, and the variation ratio of Accuracy was calculated.

When the number of dimensions is increased by 1, the feature amount that has increased Accuracy by 1% or more is extracted as the feature amount that contributes to the improvement of the classification accuracy. FIG. 6 shows the relationship between the feature vector index and Accuracy. From this figure, the feature quantity effective for the classification of the snoring sound / non-snoring sound can be understood.

Further, Table 4 shows feature vectors (ASF _{opt ′} , SSAIF _opt. ) Extracted from AS and SSAI and contributing to accuracy improvement of 1% or more. From this result, it was confirmed that the number of AS dimensions can be reduced to 4 dimensions, and the number of dimensions of SSAI can be significantly reduced to 5 dimensions.

Ave .: average

Further, in consideration of AIMF dimensional compression, 9-dimensional AIMF _opt. , Which is a combination of ASF _opt. And SSAIF _opt. , Is used as a feature vector. Table 5 shows the results of system performance evaluation using the above-described three feature vectors: ASF _opt. , SSAIF _opt. , And AIMF _opt . From the results of the table, it can be seen that it is possible to perform automatic classification and extraction of high-accuracy snoring sound with the same degree as compared with before dimension compression, using fewer feature amounts. In particular, when AIMF _opt. Was used, the highest system accuracy (Accuracy) was found to be 96.9% (sensitivity: 97.2%, specificity: 96.3%). FIG. 7 shows the ROC analysis result when AIMF _opt. Is used. Thus, the effectiveness of the bioacoustic extraction apparatus according to the present embodiment and the optimum feature amount for snoring sound extraction could be confirmed.

AS: auditory spectrum; SSAI: summary SAI; OT: optimum threshold; TP: true positive; FP: false positive; TN: true negative; FN: false negative; : Sensitivity; Spe. : Specificity; AUC: area under the curve; Acc. : Accuracy; PPV: positive predictive value; NPV: negative predictive value
(Contrast with conventional method)

As described above, the effectiveness of the classifier that classifies AIM-based snore / non-snoring sounds has been demonstrated. Next, a report example conventionally proposed as a snoring sound / non-snoring sound classification method is compared with the present embodiment to verify the superiority of the superiority. Here, as past reports, Duckitt et al. MFCCs, Cavusoglu et al. SED, Karunajeewa et al. Normalized AC's, LPCs, etc., Azarbarzin et al. SED (500 Hz), Dafna et al. Table 7 below summarizes the classification performance and the accuracy Acc obtained by classifying the data of 40 subjects by the above-described method of this example. As is clear from this table, the subject data and conditions for classification are different, but the method of this example achieves classification accuracy superior to any reported example.

Acc. : Accuracy; PPV: positive predictive value; MFCCs: mel-frequency cepstrum coefficients; OSAS: obstructive sleep apnea syndrome; SED: subband energy distribution; ACs: autocorrelation coefficients; LPCs: linear predictive coding coefficients; AIMF: AIM feature
(Duckitt et al.)

Duckitt et al. Use the Mel frequency cepstrum coefficient (MFCC) to learn non-snoring sounds by classifying them into breathing sounds, noises, silent intervals, and other noises (car sounds, dog barking sounds, etc.) In the future, it has been reported that it is necessary to expand the voice so that it is possible to distinguish and learn voices such as sleep (Non-Patent Document 5).
(Cavusoglu et al.)

Cavusoglu et al. Achieved 98.7% accuracy in snoring / non-snoring classification when trained using only a simple snoring data set using Subband Energy Distribution (SED). You are doing. However, it has been reported that the accuracy decreased to 90.2% when learning with a data set including a snore sound and a snore sound of an OSAS patient (Non-patent Document 7). In contrast, the classification method according to the embodiment of the present invention achieves an accuracy of 97.3% even in a data set including a simple snore sound and an OSAS snore sound.
(Karunajeewa)

Karunajeewa et al. Used only the breathing sound and silence period to classify the non-snoring sounds using normalized AC's, LPCs, etc., and the speech sounds and noises were 10 minutes or 20 minutes before the patient fell asleep. It is reported that it can be avoided by excluding it (Non-Patent Document 6). Therefore, in order to investigate the proportion of breathing sounds and other non-snoring sounds, a non-snoring sound in the database used in this example was subjected to an audibility evaluation test by three examples, breathing sounds, cough, voice The number of episodes was investigated by classifying into four classes (sleeping, moaning, speaking) and noise (bed squeak, metal, siren, etc.). As a result, despite the exclusion of the first hour from the start of recording, non-snoring sounds other than breathing sounds accounted for 24.4% of the total non-snoring sounds. It was found that can not be ignored. In such a data set, the classification method of the present example was able to show a high classification accuracy of 96.9%. Therefore, it is suggested that the classification method according to the example can cope with various sounds assumed to occur during sleep.
(Azarbarzin et al.)

Azarbarzin et al. Reported that the data set of simple snore sound and OSAS snore sound was classified using SED of 500 Hz, and the accuracy of 93.1% was obtained (Non-patent Document 9). However, the classification target data is extracted from only 15 minutes. In contrast, the present embodiment achieves 97.3% for data as long as 2 hours.
(Dafna et al.)

Dafna et al. Uses a 34-dimensional feature vector that combines a plurality of acoustic analysis methods such as MFCCs, LPCs, and SED (Non-Patent Document 10). However, this method needs to fully understand and use the theory about each feature amount, and the complexity of the system is high. In contrast, this embodiment has an advantage that a relatively low-dimensional feature vector can be used to construct a simple program using an existing AIM simulator.

In addition, when using these feature vectors, Tsuzaki et al. Calculated the total number / position / level of peaks, spectral centroid, spectral tilt, and quantity corresponding to spectral undulations for AS, and the peak time interval for SSAI. A pitch equivalent value is calculated by the logarithm of the reciprocal number, and feature extraction is performed as an index of the clarity of the pitch by the peak height (Non-Patent Document 11). However, there is little information on SSAI being used, and it is necessary to mention means for more effectively using it. In addition, robustness with respect to detection accuracy is an issue to realize automatic peak detection. On the other hand, in the present embodiment, as a result of verifying the effect of the snoring sound and non-snoring sound classification method by the AIM feature extraction method that does not use the peak detection system, the result when both AS and SSAI are used is as follows. The highest accuracy is shown. Further, even when only SSAI is used, the accuracy of 94% is obtained, and SSAI information is also effectively used.

In addition, according to the voiced classification method using AS or SSAI according to the present embodiment, the feature amount obtained from the acoustic spectrum, for example, the feature amount corresponding to the total number, position, and amplitude of the peak, the center of gravity, inclination, and increase of the spectrum. The feature amount corresponding to the decrease can be extracted from the AS or SSAI.

In addition, screening can be performed only for a section having a pitch (period) in the extracted snoring sounds. Alternatively, only a section having a pitch may be extracted when the snoring sound is extracted in advance. In the above example, a sound having an SNR of 5 dB or more in the segment is used as the AE, but a sound with SNR <5 dB can also be used by using the sounded section detection method.

In the AIM processing, it is not necessary to perform all the stages shown in the block diagram of FIG. 4. For example, feature speeds obtained up to the NAP (neural activity pattern) stage are used to speed up the processing. be able to.
(Example)

Using the AIM-based snoring sound and non-snoring sound classification methods using the non-contact microphone technology described above, the accuracy of the bioacoustic extraction method was evaluated using 40 subjects. These results will be described below as examples.
(System noise immunity)

In this example, a non-contact microphone was used to record sleep-related sounds. This approach is often discussed in sleep-related sound classification studies compared to contact microphones. The non-contact microphone has an advantage that recording can be performed without imposing a load on the subject, while the magnitude of the SNR during recording is a problem. In past reports using a non-contact microphone, noise reduction processing such as a spectral subtraction method is used for preprocessing as an approach to improve the SNR of a signal. However, the spectrum subtraction process generates synthesized speech called musical noise, and it becomes difficult to estimate the fundamental frequency at a low SNR. On the other hand, the gamma chirp filter bank used in the BMM of this embodiment can effectively extract voice from a noisy environment even with a low SNR such as -2 dB without causing musical noise. This is presumably because AIM has the characteristic of preserving the fine structure of periodic sounds rather than noise. AIM-based feature vectors are also reported to have higher noise suppression than MFCC. For the above reasons, it can be said that AIM has excellent noise resistance against recording in a real environment.
(Sound section estimation)

Here, an example of a method in which the sound segment estimation unit 20 estimates the sound segment will be described based on the flowchart in FIG. 8 and the graphs in FIGS. 9A to 9E. Here, as an example of the waveform of the original acoustic data (time change in signal intensity), it is considered that a snoring episode is extracted as a voiced segment from sleep-related sound data as shown in FIG. 9A.

First, in step S801, sleep related sounds are collected. Here, the sleep related sound used as original sound data (FIG. 9A) is recorded from the patient during sleep using a non-contact type microphone.

Next, in step S802, the original sound data is differentiated or differentiated. This process is performed by the pre-processor 21 shown in FIG. Here, the original acoustic data of FIG. 9A is differentiated by a differentiator which is the pre-processor 21, and the signal waveform of the pre-processed data obtained as a result is shown in FIG. 9B. The difference can be performed by a first-order FIR (Finite Impulse Response) filter which is one of digital filters.

Further, in step S803, the preprocess data is squared. This process is performed by the squarer 22 shown in FIG. FIG. 9C shows a signal waveform of the square data obtained as a result of squaring the preprocessed data in FIG. 9B by the squarer 22.

In step S804, the square data is down-sampled. This processing is performed by the downsampling device 23 shown in FIG. FIG. 9D shows a signal waveform of the down-sampling data obtained as a result of down-sampling the square data of FIG. 9C by the down-sampler 23. Instead of downsampling, the square data may be divided into segments with N = 400 and a shift width of 200, for example, and the signal energy in the kth segment may be obtained.

In step S805, the median value is acquired from the downsampling data. This processing is performed by the median filter 24 shown in FIG. FIG. 9E shows a signal waveform obtained as a result of obtaining the median value by the median filter 24 from the down-sampling data of FIG. 9D. In this way, it becomes possible to extract only necessary bioacoustic data as shown in FIG. 9E from the original acoustic data buried in the background noise as shown in FIG. 9A, and it is a sleep related sound buried in the background noise. In addition, it is possible to accurately extract a sound section (breathing sound, snoring sound, etc.).

In addition, the detection of the voiced section may be realized using a learning machine such as a neural network, other time series analysis techniques, and signal analysis / modeling techniques in addition to using the difference, the square, and the like as described above.
(Comparative Examples 1 and 2)

Here, for comparison, a conventionally known method as a method for detecting a speech section is applied as a comparative example. Here, a method using a zero-crossing rate (ZCR) is referred to as Comparative Example 1, and an STE method based on the energy of an audio signal is referred to as Comparative Example 2, and each of the sound sections of the original acoustic data in FIG. The results of the automatic extraction are shown in FIGS. 10B and 10C, respectively. For further comparison, FIG. 10D shows an automatic extraction result obtained by the method according to Example 1 described above. These ZCR and STE methods are typical techniques as a voice section detection method for detecting only a voice section from a voice signal input from the outside.
(ZCR)

First, ZCR is defined by the following equation.

Here, sgn [χ (k)] is expressed by the following equation.

The ZCR is mainly used in a scene where the ZCR of the voiced section to be uttered is considerably smaller than the ZCR of the silent section. This method depends on the type of voice (strong harmony structure).
(STE method)

In the STE method, the STE function of sound is defined as shown in Equation 1 above.

This STE method is mainly used in a scene in which the value of E _k in a voiced section is considerably larger than E _{k in a} silent section, the SNR is high, and E _{k in} a voiced section can be clearly read from background noise. Has been.

As described above, since all methods such as ZCR and STE perform automatic speech interval detection on the premise of an environment with relatively little influence of noise, it is appropriate in an environment where the SNR decreases as shown in FIG. 10A. It is difficult to extract a sound section. In the ZCR of Comparative Example 1 shown in FIG. 10B, it cannot be said that sufficient extraction is performed, and even with the signal energy of Comparative Example 2 shown in FIG. 10C, peaks appear uniformly and separation is difficult. On the other hand, as shown in FIG. 10D, the automatic extraction result by the method according to the above-described first embodiment is clearly extracted as shown in FIG. 10D, and the sleep-related sound data having an extremely small SNR as shown in FIG. 10A. However, by using the sounded section estimation unit according to the example, a sounded section (breathing sound, snoring sound) can be extracted and separated, and its usefulness has been confirmed.

In this example, as shown in Table 6, sleep-related sound data is collected for 10 subjects as original sound data, and the length of the sleep-related sound is 120 s. It has been classified in advance as sound (Breath).

In accordance with the definitions of sensitivity (Sensitivity), specificity (Specificity), and area under the curve (AUC) described above, the area under the curve, the sensitivity, and the specificity for each subject, Example 1, Comparative Example 1, and Comparative Example 2, respectively. Calculated. The obtained results are shown in Tables 9 to 11 below. As a result, in Example 1, it was confirmed that high precision was achieved in all of sensitivity, specificity, and area under the curve.

(Example 3)

Further, screening based on the AIM, that is, sieving is performed by the screening unit 70 of FIG. 1 on the data classified into the snoring sound and the non-snoring sound. Here, the screening part 70 determines whether it is OSAS (obstructive sleep apnea syndrome) or non-OSAS from the snoring sound. Here, Example 3 was performed in order to confirm whether the screening unit 70 can appropriately classify OSAS and non-OSAS. Here, an AE data set extracted from 31 subjects was prepared as a data set for classifying snoring sounds and non-snoring sounds. Of these, 20 were used as learning data sets and 11 were used as test data sets. Here, the data of the subject's sleep for 1 hour was extracted for 2 hours.

AE is pre-labeled into snoring sound or non-snoring sound by hand. Details of the dataset used to classify snoring sounds and non-snoring sounds are shown in the table below.

M: male; F: female; BMI: body mass index; AHI: apnea-hypopnea index

On the other hand, the data set used for OSAS and non-OSAS classification is shown in the following table. Here, 50 subjects were used, 35 of which were used as the learning data set and 15 were used as the test data set.

(Classification of OSAS and non-OSAS based on AIM)

First, the classification performance of the snoring sound or the non-snoring sound by the classifying unit 50 and the discriminating unit 60 of FIG. 1 is extracted from a 6-dimensional feature vector (kurtosis, skewness, spectral centroid, spectral bandwidth, and SSAI extracted from AS. Kurtosis and skewness). As shown in this table, by using these feature vectors, it was confirmed that snoring sounds or non-snoring sounds were classified with extremely high accuracy of sensitivity 98.4% and specificity 94.06%.

Further, in the present embodiment, the eight-dimensional features of the snoring sound discriminated by the classifying unit 50 and the discriminating unit 60 of FIG. 1 (distortion degree extracted from AS, spectral centroid, spectral roll-off, kurtosis extracted from SSAI, Based on the skewness, spectral bandwidth, spectral roll-off, spectral entropy) vectors, the screening unit 70 classified OSAS and non-OSAS. Further, in the classification of the snoring sound or the non-snoring sound, and the classification of the OSAS and the non-OSAS, the above-described combination of feature amounts can be used. In addition to these feature quantities, spectral asymmetry, band energy ratio, and the like can also be used. The evaluation of the OSAS and non-OSAS classification was performed by a 10-fold cross-validation test. Here, 9fold randomly selected from the data set was used for learning, and the remaining 1fold was used for testing.

As a result, when the threshold value used as the judgment standard of apnea-hypopnea index (AHI) was set to 15 events / h and the OSAS patients were screened by the screening unit 70, as shown in Table 14 above. Excellent results were obtained with a sensitivity of 85.00% ± 26.87 and a specificity of 95.00% ± 15.81, confirming the usefulness of this example. The AHI is not limited to this value, and may be 5 events / h, 10 events / h, or the like. Further, in the analysis in the classification unit 50, the determination unit 60, and the screening unit 70, classification, determination, and sieving in consideration of the characteristics of each gender are possible. For classification and identification of sleep sounds using the classification unit 50 and the determination unit 60, pattern recognition / identification technology, learning machines such as neural networks, deep neural networks, support vector machines (SVM), etc., are used instead of MLR. Can also be used. In the above embodiment, the two-class classification is used. However, a multi-class classification problem can be considered using the learning machine. For example, it can be classified directly into OSAS snoring (1), non-OSAS snoring (2), and non-snoring (3) based on the feature amount. Of course, automatic extraction can be classified into multiple classes such as snoring (1), breathing sound (2), and cough (3).
Example 4

Next, as Example 4, it was verified whether snoring / non-snoring discrimination and OSAS / non-OSAS discrimination were possible with the number of subjects increased, that is, with the subject database expanded. Here, an Audio event (AE) acquired from a 2-hour sleep sound is used. The sleep sound is recorded during the polysomnography (PSG) examination all night. To perform the performance evaluation, two viewers engaged in sleep research are carefully labeling two classes of snore / non-snoring sounds. The subject information obtained during the PSG examination and the average value of the number of snores obtained by labeling are shown in the subject database of Table 15.

(Stepwise method)

A Stabilized auditory image (SAI) was formed for each AE, and the Auditory spectrum (AS) and Summary stabilized auditory (image (SSAI) obtained after 10 frames were used for analysis. Each frame was normalized so that the maximum amplitude of AS and SSAI was 1. In addition, when normalizing with a standard deviation, it is also possible to normalize with one episode. From AS, eight characteristics of Kurtosis, Skewness, Spectral centroid, Spectral bandwidth, Spectral roll-off, Spectral entropy, Spectral contrast, Spectral flatness were used. In SSAI, seven feature values other than Spectral flatness were used. Since these feature amounts are extracted from all frames of the AE, the average value of each feature amount is used as the feature amount obtained from the AE. In order to select an effective feature quantity for discrimination from these feature quantities, a stepwise method, which is a feature selection algorithm, was used for each male, female, and male / female data set.

Learning based on the MLR model is performed using the feature vector selected as described above. Here, classification of snoring sound / non-snoring sound using MLR performed in step S305 of the flowchart showing the bioacoustic extraction method using AIM shown in FIG. 3 (in the case of OSAS screening: classification of OSAS / non-OSAS) Went. In step S307, discrimination based on the threshold value was performed. Here, in order to classify the acoustic feature quantity into a predetermined type by the classifying unit 50 and discriminate the snoring sound or the non-snoring sound by the discriminating unit 60, a multinomial distribution logistic regression using a feature vector extracted from the AE ( Multi-nomial logistic regression (MLR) analysis was used.
(Performance evaluation of automatic snoring classification)

In order to evaluate the performance of automatic snoring classification using AIM, leave-one-out cross validation was performed. Here, sensitivity (Specificity), AUC (area under the curve), accuracy (Accuracy), positive pre-dictive value (PPV), negative appropriateness (Negative) in all test data sets The predictive value (NPV) was calculated, and the average value and standard deviation were calculated to evaluate the classification performance of the system (both automatic classification and screening are evaluated based on strict criteria).
(Performance evaluation of OSAS screening)

For performance evaluation of OSAS screening using AIM, 70% randomly selected from the subject database was used as a learning data set, and the remaining 30% was used as a test data set. Both datasets were divided independently so that the OSAS subjects and Non-OSAS subjects would be equal, and 50 patterns were randomly created, and the sensitivity (Sensitivity) and specificity (Specificity) ), AUC (area under the curve, accuracy, positive pre-dictive value (PPV), negative predictive value (NPV), and then calculate the mean and standard deviation. The OSAS screening performance was evaluated.
(Performance evaluation results of automatic snoring classification)

Table 16 shows the performance evaluation results (Leave-one-out cross-validation) of automatic snoring sound classification using AIM.

From the above results, it was found that snoring can be automatically classified with high accuracy for both men and women using only the features obtained from AIM.
(Results of OSAS screening performance evaluation using AIM)

Table 17 shows the performance evaluation results of the OSAS screening based on the snoring sound, which was automatically extracted using the AIM by the method described above. For reference, Table 18 shows the performance evaluation results of OSAS screening using only the snoring sound, which is extracted manually without labeling and automatically extracted.

Table 17 suggests that OSAS screening can be performed with high accuracy in any data set based on snoring sounds that are automatically extracted using only AIM, imitating human auditory ability. From Tables 17 and 18, it was found that the performance of OSAS screening based on snoring sounds manually extracted by labeling was higher in all subject sets than in the case of automatic extraction. This result suggests that the OSAS screening performance is improved by further improving the performance of automatic extraction of snoring sound using AIM. In order to improve the snoring automatic extraction performance, it is possible to change the normalization method in the AS and SSAI frames, for example, normalize in one episode instead of normalizing in frames. In addition, addition of feature amounts, for example, signal processing such as pitch information and formant frequency information, speech recognition, feature vectors used for speech signal processing, and the like as described in Non-Patent Document 1 are possible. .

Furthermore, in this embodiment, AIM processing was performed for a high-sounding sound section. In addition, it is possible to classify into categories such as cough, snoring, breathing, vocalization, bed noise, etc. by performing AIM processing on a voiced section including a low SNR sound.

To summarize the method according to the above embodiment, first, the SAI is obtained for every 35 ms frame by the AIM processing. Next, AS and SSAI are obtained from each SAI. Further, feature values (kurtosis, skewness, spectrum bandwidth, spectrum centroid, spectrum entropy, spectrum roll-off, spectrum flatness, etc.) are extracted from AS and SSAI, respectively. Since the feature amount is obtained by the number of frames, each feature amount obtained from one sound of the sounded section can be used as an average value and a standard deviation by averaging. Here, in the case of automatic extraction of snoring sound, the average value and standard deviation of feature values obtained from AS and SSAI were used. On the other hand, in the OSAS screening, only the average value of feature values obtained from AS and SSAI was used. Thus, compared with OSAS screening, more feature values were used and examined in the snoring sound automatic extraction. More feature quantities can be used, and basic statistics (standard deviation, kurtosis, etc.) of feature quantities, feature quantities such as total number of peaks, appearance position, amplitude, center of gravity, slope, increase, decrease, etc. Can be used to evaluate the performance of bioacoustic automatic extraction and OSAS screening. Further, it is not always necessary to use SAI. For example, the calculation speed can be improved by extracting the feature amount up to NAP which is the process before SAI.

In the above example, the snoring sound has been described as an example, but the subject of the present invention is not limited to the snoring sound, but can be used for other sounds (biological sound) generated by biological objects, and from the detected bioacoustics, It can be applied to discovery and diagnosis of various cases. For example, by detecting sleep sound, OSAS screening as described above and discrimination of sleep disorders can be performed. Diagnosis of asthma, pneumonia, etc. can be made from lung sounds, respiratory sounds, coughs, and the like. Alternatively, various heart diseases can be made from heart sounds, and various intestinal diseases such as functional gastrointestinal disorders can be screened by analyzing intestinal sounds. In addition, it can also be applied to detection of fetal movement sounds, muscle sounds, and the like. By appropriately extracting such bioacoustics, it can be suitably used for analysis of cases that can be diagnosed from these bioacoustics. Moreover, this invention can be utilized not only for a human but for other living organisms. For example, it can be suitably used for health examinations of pets and animals bred at zoos.

The bioacoustic extraction device, the bioacoustic analysis device, the bioacoustic extraction program, the computer-readable recording medium, and the recorded device according to the present invention measure snoring sounds together with or in place of a patient's polysomnographic examination, It can utilize suitably as a use which diagnoses.

DESCRIPTION OF SYMBOLS 100 ... Bioacoustic extraction apparatus 110 ... Bioacoustic analysis apparatus 10 ... Input part 20 ... Sound section estimation part 21 ... Preprocessor 22 ... Squarer 23 ... Downsampling device 24 ... Median filter 30 ... Auditory image generation part 40 ... Sound Feature amount extraction unit 50 ... classification unit 60 ... discrimination unit 70 ... screening unit

Claims (27)

1. A bioacoustic extraction device for extracting necessary bioacoustic data from original acoustic data including bioacoustic data,
   An input unit for acquiring original acoustic data including bioacoustic data;
   From the original sound data input from the input unit, a voiced section estimation unit that estimates a voiced section;
   An auditory image generation unit that generates an auditory image according to an auditory image model based on the voiced segment estimated by the voiced segment estimation unit;
   An acoustic feature amount extraction unit that extracts an acoustic feature amount from the auditory image generated by the auditory image generation unit;
   A classification unit that classifies the acoustic feature amount extracted by the acoustic feature amount extraction unit into a predetermined type;
   A bioacoustic extraction apparatus comprising: a discrimination unit that discriminates whether or not the acoustic feature quantity classified by the classification unit is bioacoustic data based on a predetermined threshold value.
2. The bioacoustic extraction device according to claim 1,
   The auditory image generation unit is configured to generate a stabilized auditory image using an auditory image model,
   A bioacoustic extraction apparatus in which the acoustic feature amount extraction unit extracts an acoustic feature amount based on the stabilized auditory image generated by the auditory image generation unit.
3. The bioacoustic extraction device according to claim 2,
   The auditory image generation unit is configured to generate an overall stabilized auditory image and an auditory spectrum from the stabilized auditory image,
   A bioacoustic extraction apparatus in which the acoustic feature quantity extraction unit extracts an acoustic feature quantity based on the overall stabilized auditory image generated by the auditory image generation unit and an auditory spectrum.
4. The bioacoustic extraction device according to claim 3,
   The acoustic feature quantity extraction unit includes at least kurtosis, skewness, spectrum centroid, spectrum bandwidth, spectrum flatness, spectrum roll-off, spectrum entropy, and octave-based spectrum contrast of the auditory spectrum and / or the overall stabilized auditory image. A bioacoustic extraction apparatus that extracts either as an acoustic feature.
5. The bioacoustic extraction device according to claim 1,
   The auditory image generation unit is configured to generate a neural activity pattern using an auditory image model,
   The bioacoustic extraction apparatus which the said acoustic feature-value extraction part extracts an acoustic feature-value based on the neural activity pattern produced | generated by the said auditory image production | generation part.
6. A bioacoustic extraction device for extracting necessary bioacoustic data from original acoustic data including bioacoustic data,
   An input unit for acquiring original acoustic data including bioacoustic data;
   From the original sound data input from the input unit, a voiced section estimation unit that estimates a voiced section;
   An auditory image generation unit that generates an auditory image according to an auditory image model based on the voiced segment estimated by the voiced segment estimation unit;
   An auditory spectrum generator that generates an auditory spectrum for the auditory image generated by the auditory image generator;
   An overall stabilized auditory image generator for generating an overall stabilized auditory image for generating an overall stabilized auditory image for the auditory image;
   An acoustic feature amount extraction unit that extracts an acoustic feature amount from the auditory spectrum generated by the auditory spectrum generation unit and the overall stabilized auditory image generated by the overall stabilization auditory image generation unit;
   A classification unit that classifies the acoustic feature amount extracted by the acoustic feature amount extraction unit into a predetermined type;
   A bioacoustic extraction apparatus comprising: a discrimination unit that discriminates whether or not the acoustic feature quantity classified by the classification unit is bioacoustic data based on a predetermined threshold value.
7. The bioacoustic extraction device according to any one of claims 1 to 6,
   The bioacoustic extraction apparatus comprised so that the area which has a period may be extracted among original sound data.
8. The bioacoustic extraction device according to any one of claims 1 to 7,
   The voiced section estimation unit,
   A pre-processor for differentiating or differentiating the original acoustic data and pre-processing;
   A squarer for squaring the preprocessed data preprocessed by the preprocessor;
   A downsampler for downsampling the squared data squared by the squarer;
   A median filter for obtaining a median value from the downsampled data downsampled by the downsampler;
   A bioacoustic extraction apparatus.
9. The bioacoustic extraction device according to any one of claims 1 to 8,
   The bioacoustic extraction device in which the input unit is a non-contact microphone that is installed in a non-contact manner with a patient to be examined.
10. The bioacoustic extraction device according to any one of claims 1 to 9,
    A bioacoustic extraction apparatus in which the discrimination of bioacoustic data by the discrimination unit is non-language processing.
11. The bioacoustic extraction device according to any one of claims 1 to 10,
    The original acoustic data is a bioacoustic acquired when the patient sleeps,
    A bioacoustic extraction apparatus configured to extract necessary bioacoustic data from bioacoustic data acquired during sleep.
12. The bioacoustic extraction device according to any one of claims 1 to 11,
    The original sound data are sleep-related sounds collected during patient sleep,
    The bioacoustic data is snoring sound data,
    The bioacoustic extraction apparatus in which the predetermined type is different from a snoring sound and a non-snoring sound.
13. A bioacoustic analyzer for extracting and analyzing necessary bioacoustic data from original acoustic data including bioacoustic data,
    An input unit for acquiring original acoustic data including bioacoustic data;
    From the original sound data input from the input unit, a voiced section estimation unit that estimates a voiced section;
    An auditory image generation unit that generates an auditory image according to an auditory image model based on the voiced segment estimated by the voiced segment estimation unit;
    An acoustic feature amount extraction unit that extracts an acoustic feature amount from the auditory image generated by the auditory image generation unit;
    A classification unit that classifies the acoustic feature amount extracted by the acoustic feature amount extraction unit into a predetermined type;
    A discrimination unit that discriminates whether or not the biometric data is based on a predetermined threshold with respect to the acoustic feature amount classified by the classification unit;
    A screening unit that performs screening on true value data determined as bioacoustic data by the determination unit;
    A bioacoustic analyzer.
14. A bioacoustic analyzer for extracting and analyzing necessary bioacoustic data from original acoustic data including bioacoustic data,
    An input unit for acquiring original acoustic data including bioacoustic data;
    From the original sound data input from the input unit, a voiced section estimation unit that estimates a voiced section;
    An auditory image generation unit that generates an auditory image according to an auditory image model based on the voiced segment estimated by the voiced segment estimation unit;
    An auditory spectrum generator that generates an auditory spectrum for the auditory image generated by the auditory image generator;
    For the auditory image, a generalized stabilized auditory image generation unit that generates a generalized stabilized auditory image;
    An acoustic feature amount extraction unit that extracts an acoustic feature amount from the auditory spectrum generated by the auditory spectrum generation unit and the overall stabilized auditory image generated by the overall stabilization auditory image generation unit;
    A classification unit that classifies the acoustic feature amount extracted by the acoustic feature amount extraction unit into a predetermined type;
    A discrimination unit that discriminates whether or not the biometric data is based on a predetermined threshold with respect to the acoustic feature amount classified by the classification unit;
    A screening unit that performs screening on true value data determined as bioacoustic data by the determination unit;
    A bioacoustic analyzer.
15. The bioacoustic analyzer according to claim 13 or 14,
    The said screening part is a bioacoustic analyzer comprised so that a disease screening may be performed with respect to the bioacoustic data extracted from original acoustic data.
16. The bioacoustic analyzer according to claim 15,
    The screening unit is a bioacoustic analysis apparatus configured to perform screening for obstructive sleep apnea syndrome on bioacoustic data extracted from original acoustic data.
17. A bioacoustic extraction method for extracting necessary bioacoustic data from original acoustic data including bioacoustic data,
    Obtaining original acoustic data including bioacoustic data;
    Estimating the voiced section from the acquired original acoustic data;
    Generating an auditory image according to an auditory image model based on the estimated voiced interval;
    Extracting an acoustic feature from the generated auditory image;
    Classifying the extracted acoustic features into a predetermined type;
    And a step of determining whether or not the classified acoustic feature amount is bioacoustic data based on a predetermined threshold.
18. A bioacoustic extraction method for extracting necessary bioacoustic data from original acoustic data including bioacoustic data,
    Obtaining original acoustic data including bioacoustic data;
    Estimating the voiced section from the acquired original acoustic data;
    Generating a stabilized auditory image according to an auditory image model based on the estimated voiced section;
    Generating an overall stabilized auditory image from the stabilized auditory image;
    Extracting a predetermined acoustic feature obtained from the generated generalized stabilized auditory image;
    And a step of discriminating whether or not the extracted acoustic feature quantity is bioacoustic data based on a predetermined threshold value.
19. The bioacoustic extraction method according to claim 18,
    Generating an auditory spectrum from the stabilized auditory image;
    In the step of extracting the predetermined acoustic feature amount, a bioacoustic extraction method of extracting a predetermined acoustic feature amount obtained from the generated auditory spectrum in addition to the overall stabilized auditory image.
20. The bioacoustic extraction method according to claim 18 or 19, further comprising:
    Prior to the step of extracting the predetermined acoustic feature amount, a bioacoustic extraction method including a step of selecting an acoustic feature amount contributing to identification from the extracted acoustic feature amount.
21. The bioacoustic extraction method according to any one of claims 18 to 20,
    The bioacoustic extraction method, wherein the step of determining whether or not the bioacoustic data is classification of snoring sound or non-snoring sound using a multinomial distribution logistic regression analysis.
22. A bioacoustic analysis method for extracting and analyzing necessary bioacoustic data from original acoustic data including bioacoustic data,
    Obtaining original acoustic data including bioacoustic data;
    Estimating the voiced section from the acquired original acoustic data;
    Generating a stabilized auditory image according to an auditory image model based on the estimated voiced section;
    Generating an auditory spectrum and an overall stabilized auditory image from the stabilized auditory image;
    Extracting a predetermined acoustic feature obtained from the generated auditory spectrum and the overall stabilized auditory image;
    Determining whether or not the extracted acoustic feature amount is bioacoustic data based on a predetermined threshold;
    Screening for true value data determined as bioacoustic data in the determination step;
    A bioacoustic analysis method including:
23. The bioacoustic analysis method according to claim 22,
    The bioacoustic extraction method, wherein the screening is screening for obstructive sleep apnea syndrome or non-occlusive sleep apnea syndrome using multinomial logistic regression analysis.
24. A bioacoustic extraction program for extracting necessary bioacoustic data from original acoustic data including bioacoustic data,
    An input function for acquiring original acoustic data including bioacoustic data;
    A voiced section estimation function for estimating a voiced section from the original sound data input by the input function,
    Auditory image generation function for generating an auditory image according to an auditory image model based on the voiced segment estimated by the voiced segment estimation function;
    An acoustic feature amount extraction function for extracting an acoustic feature amount from the auditory image generated by the auditory image generation function;
    A classification function for classifying the acoustic feature quantity extracted by the acoustic feature quantity extraction function into a predetermined type;
    The bioacoustic extraction program which makes a computer implement | achieve the discrimination | determination function which discriminate | determines whether it is bioacoustic data based on the predetermined threshold value with respect to the acoustic feature-value classified by the said classification function.
25. A bioacoustic extraction program for extracting necessary bioacoustic data from original acoustic data including bioacoustic data,
    An input function for acquiring original acoustic data including bioacoustic data;
    A voiced section estimation function for estimating a voiced section from the original sound data input by the input function,
    A stabilized auditory image generation function for generating a stabilized auditory image according to an auditory image model based on the voiced segment estimated by the voiced segment estimation function;
    A function of generating an overall stabilized auditory image from the stabilized auditory image;
    An acoustic feature amount extraction function for extracting a predetermined acoustic feature amount for the generated overall stabilized auditory image;
    A classification function for classifying the predetermined acoustic feature amount extracted by the acoustic feature amount extraction function into a predetermined type;
    The bioacoustic extraction program which makes a computer implement | achieve the discrimination | determination function which discriminate | determines whether it is bioacoustic data based on the predetermined threshold value with respect to the acoustic feature-value classified by the said classification function.
26. A bioacoustic analysis program for extracting and analyzing necessary bioacoustic data from original acoustic data including bioacoustic data,
    An input function for acquiring original acoustic data including bioacoustic data;
    A voiced section estimation function for estimating a voiced section from the original sound data input by the input function,
    A stabilized auditory image generation function for generating a stabilized auditory image according to an auditory image model based on the voiced segment estimated by the voiced segment estimation function;
    A function of generating an overall stabilized auditory image from the stabilized auditory image;
    An acoustic feature amount extraction function for extracting a predetermined acoustic feature amount for the generated overall stabilized auditory image;
    A classification function for classifying the predetermined acoustic feature amount extracted by the acoustic feature amount extraction function into a predetermined type;
    A discrimination function for discriminating whether or not bioacoustic data is based on a predetermined threshold with respect to the acoustic feature amount classified by the classification function;
    A function of performing screening on true value data determined as bioacoustic data by the determination function;
    A bioacoustic analysis program that enables a computer to realize
27. A computer-readable recording medium or a recorded device storing the program according to claim 25 or 26.

Images