JP2004198383A - Apparatus for analyzing sound source - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To identify sound sources by totally determining various characteristics of sound features observed on the frequency or the time waveform which sound posseses. <P>SOLUTION: A sound source analyzing apparatus is provided with a waveform extraction part 1 for computing/extracting the waveform of sound 4 to be evaluated; a spatial distance computing part 2 for computing a spatial distance of the waveform of the sound 4 on a reference space 2a, according to possible cause represented by a plurality of preregistered reference waveforms; and a cause identifying part 3 for identifying the characteristics of the sound source, on the basis of distance computation results that are output from the spatial distance computing part 2. A broadband noise component N is mixed with the plurality of reference waveforms in the spatial distance computing part 2. The sound waveform itself to be evaluated is developed on the multidimensional space for the number of waveform data points, and the sound sources are identified by the statistical distance on the multidimensional space. Accordingly, sound source analysis with high accuracy becomes possible, while synthesizing not only the limited features based on intention of an analyst, such as frequency information or duration, but also a variety of sound features. <P>COPYRIGHT: (C)2004,JPO&NCIPI

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a sound source analyzer that identifies the cause of generation of sound radiated from various devices to be analyzed, and particularly to the case of identifying the cause of generation of impact sound generated from office equipment such as a copying machine or a printer. The present invention relates to an improvement of a sound source analyzer that is effective for a computer.
[0002]
[Prior art]
2. Description of the Related Art In recent years, an office environment has been required to be improved in order to support comfort or intellectual productivity in an office. Office machines such as copiers and printers are also required to be quiet so as not to hinder intellectual work, and product development aimed at noise reduction is being promoted.
[0003]
FIG. 17 shows a time waveform of noise generated during operation of a general electrophotographic printer.
In this figure, the horizontal axis represents time, and the vertical axis represents sound pressure, showing the situation where the sound pressure of noise varies with time.
Here, a portion indicated by a circle is called an impact sound, and a sound having a high sound pressure is generated in a short time of about several tens to several hundreds ms. It has been clarified that such a noise is caused by a collision of a mechanical portion represented by a solenoid or the like, a collision between a paper and a component in a paper alignment portion, or a buckling of the paper itself.
[0004]
FIG. 18 is an enlarged view of the time direction of FIG. 17 showing a time-sound pressure waveform of only an impact sound portion.
Generally, such impulsive sounds are easily recognized by humans because they exhibit large sound pressure fluctuations in a short period of time, and are a major cause of discomfort. Noise countermeasures are being implemented.
However, electrophotographic copiers and printers have many orders of magnitude more mechanical parts than general home appliances, and various sounds operate at various timings. It is very difficult to identify the location and the cause of the generation of an impulsive sound.
[0005]
The method of identifying the sound source position includes measuring the sound intensity taking into account not only the sound pressure but also the particle velocity of the sound, thereby grasping the flow of sound as energy, performing simultaneous measurement with multiple microphones, There is known a method of estimating the position of a sound source based on information such as waveform differences and phase differences between microphones.
However, these methods have a fundamental problem that the spatial resolution depends on the sound wavelength, and the spatial resolution is insufficient for use in identifying a sound source inside a copying machine or a printer.
For example, in recent copiers and printers, each component is placed in a very close position due to the demand for miniaturization, and the resolution required for sound source analysis is often on the order of 10 mm or less, The wavelength of a sound having a frequency of 500 Hz is 680 mm, and it is difficult in principle to obtain a resolution lower than this, which is not practical.
[0006]
In addition, a method such as near-field acoustic holography has been proposed which can obtain a resolution equal to or less than the wavelength of a target sound by measuring an evanescent wave generated near a sound source.
However, in this method, the resolution depends on the distance from the sound source to the measurement position, and is effective when the microphone can be brought close to the vicinity of the sound source. When a large number of machine operating parts are included and it is desired to know the position of the sound source inside the machine operating part, the microphone cannot be brought close to the sound source, so that accurate sound source identification cannot be performed.
[0007]
Due to the problem of such sound source detection technology, at present, the sound source is estimated by comparing the impact sound generation timing with the machine operation timing, taking advantage of the characteristic that the impact sound has a steep peak on the time waveform. The main method is to go.
[0008]
[Patent Document 1]
Japanese Patent Application Laid-Open No. 2000-275096 (column of the embodiment of the invention, FIG. 1)
[Patent Document 2]
Japanese Patent Application Laid-Open No. 9-81180 (column of embodiment of the invention, FIG. 1)
[0009]
[Problems to be solved by the invention]
However, for example, in the vicinity of the sheet positioning unit, there are very sounds such as a sound of operation of the mechanical parts for positioning, a sound of the sheet colliding with the positioning member, a sound of the sheet buckling, and a sound of the positioning member returning to the initial position. Since various impact sounds are generated in a short period of time, it is often impossible to identify the sound source only by examining the operation timing. In such cases, removing the components that could be sound sources one by one, pasting a material that absorbs shock, or reducing the speed of the collision, and examining how the sound pressure level of the impact sound changes, It has been a trial-and-error symptomatic treatment, and it takes an enormous amount of time to identify the cause.
[0010]
In order to solve such a problem, there has been proposed a method of estimating a cause of sound generation by analyzing a qualitative characteristic of the sound instead of clarifying a sound generation position.
For example, Patent Literature 1 discloses a sound source type identification device that determines the type of a sound source using power spectrum analysis and a neural network.
Such analysis focusing on sound quality is considered to be a promising approach that can solve the problem of insufficient resolution due to the dependence on the sound wavelength and the distance to the sound source.
[0011]
Certainly, according to this type of sound source type identification device, a sound source having a characteristic in frequency can be determined.
However, this sound source type identification device also has the following disadvantages.
That is, for example, regarding the impact sound generated when metal plates of the same material collide with each other, when the supporting form of the metal plate is changed, the frequency characteristics of the generated sound are not changed, and the reverberation time and the like are not changed. It is known that the characteristics on the time axis change greatly, but it is not possible to identify the difference between these types of sounds.
Further, since the time from the generation of the impact sound to the disappearance thereof is as short as several tens to several hundreds of ms, there is also a problem that it is basically difficult to apply frequency analysis such as FFT analysis.
[0012]
As a solution to this problem, for example, a technique described in Patent Document 2 has been proposed.
Patent Document 2 discloses a sound source recognition device that employs measurement of a difference in duration of a sound source in addition to identification of a frequency component of the sound source.
In this sound source recognition device, by measuring the difference in the duration, it is possible to identify the difference due to the above-described difference in the support form, etc. Rather, for example, it has been reported that even if the surface hardness of the collision member is different even for the same duration, a large difference occurs in transient characteristics such as a rising curve and an attenuation curve of the sound pressure amplitude. , It is not possible to judge various characteristics appearing on the time characteristic comprehensively to identify them with high accuracy.
[0013]
The present invention has been made in order to solve the above technical problem, and comprehensively judges various sound quality characteristics observed on a frequency waveform or a time waveform of a sound to determine a sound source. It is to provide a sound source identification device for performing identification.
[0014]
[Means for Solving the Problems]
That is, according to the present invention, as shown in FIG. 1, a waveform extraction unit 1 for calculating and extracting a waveform of a sound 4 to be evaluated, and a cause-specific reference space 2a represented by a plurality of reference waveforms registered in advance. A spatial distance calculating unit for calculating a spatial distance of a sound waveform to be evaluated; a cause identifying unit for identifying a feature of a sound source from a distance calculating result output from the spatial distance calculating unit; 2 wherein a plurality of reference waveforms are mixed with a wideband noise component N.
The sound 4 to be evaluated is emitted from various devices to be analyzed 5 and the like.
[0015]
In such technical means, the waveform of the sound 4 to be evaluated includes various waveforms such as a time waveform, a frequency waveform, and a waveform using both of them.
Here, the waveform extraction unit 1 may be any unit that calculates and extracts the time waveform and the frequency waveform of the sound 4 to be evaluated from the digitized audio signal. A time waveform calculator is provided for extracting a time waveform (for example, time-sound pressure waveform) of the sound 4 to be evaluated from the signal.
In the time waveform calculation unit, waveform data of a predetermined length is cut out based on a predetermined position determined from a characteristic on the time axis such as a rising portion or a maximum value portion of the waveform.
[0016]
At this time, as a pre-process for cutting out the waveform, the digitized acoustic signal may be subjected to audibility correction such as A-characteristics, or may be passed through a high-pass filter for removing low-frequency background noise components.
By adding such preprocessing, it becomes possible to remove the background noise component and extract the waveform vector of only the sound 4 to be evaluated.
[0017]
As a specific waveform extracting means, a means for extracting a waveform by a data logger having a trigger function or the like, a means for taking a digitized acoustic signal into a computer, and extracting it on software can be used.
Further, the amplitude of the extracted time waveform may be normalized so that the maximum value or the minimum value has a predetermined value.
The processing such as the audibility correction, the high-pass filter, and the normalization of the amplitude can be configured as a digital filter or software.
[0018]
It is also possible to provide the time waveform calculation section with a sound pressure level calculation function, calculate a time-sound pressure level waveform from a time-sound pressure waveform, and use this as a time waveform.
In this case, the waveform extracting unit 1 includes a time waveform calculating unit that extracts a time-sound pressure level waveform.
[0019]
Further, the waveform extracting unit 1 can be provided with a frequency waveform calculating unit for calculating a frequency waveform separately from or in addition to the time waveform calculating unit.
Here, the frequency waveform calculator calculates a frequency waveform by performing frequency analysis on the time waveform cut out by the time waveform calculator, for example.
In a mode in which the waveform extracting unit 1 does not include the time waveform calculating unit, the frequency waveform may be calculated using an FFT analyzer or the like.
As the frequency analysis, general frequency analysis means such as FFT analysis, wavelet analysis, and generalized harmonic analysis can be used.
Further, a window function process such as Hanning, Hamming, flat top, etc. may be applied to the time waveform for frequency analysis.
By performing such processing, problems in frequency analysis such as aliasing errors and leakage errors can be reduced, and a frequency waveform according to actual conditions can be calculated.
[0020]
Further, the waveform extracting unit 1 may be provided with a waveform vector creating unit that creates a time waveform or a frequency waveform calculated by the time waveform calculating unit or the frequency waveform calculating unit as a waveform vector.
At this time, as the waveform vector, a single waveform vector may be created for each of the time waveform and the frequency waveform, or a configuration may be made such that a waveform vector obtained by connecting the frequency waveform and the time waveform is created. You may.
For example, when creating a waveform vector by connecting both waveforms, the extracted time waveform and frequency waveform are finally combined into one waveform in the waveform vector creation unit, and the spatial distance calculation unit is used as a waveform vector. 2 is output.
[0021]
The waveform extracting unit 1 may be provided with a data storage unit that stores and stores, for example, an acoustic signal output from a microphone or a digital signal output from an A / D converter.
With such a configuration, it is possible to perform off-line processing in which only recording of sound is performed first, and calculation of sound source analysis is collectively performed later.
[0022]
Further, in the spatial distance calculation unit 2, in the cause-specific reference space 2a defined from a plurality of reference waveforms for each cause registered in advance, how far the waveform vector of the sound 4 to be evaluated is from the center of gravity of the sound 4 The distance in the space of squid is calculated.
As the distance measure for calculating the spatial distance, general distance measures used in statistical discriminant analysis and cluster analysis, for example, Euclidean distance, standardized Euclidean distance, Minkowski distance, Mahalanobis distance Although it is possible to use a Mahalanobis distance, it is preferable to use a Mahalanobis distance because a spatial distance according to the actual situation can be obtained even when a complicated correlation exists between the variable groups.
[0023]
Further, in the spatial distance calculation unit 2, such a reference space is defined and registered in advance for at least one or more causes of occurrence, and how long the waveform of the sound 4 to be evaluated is in the space for each cause. It is calculated whether it has.
Generally, data for defining a reference space is represented using a data group representing the space.
For example, when defining a reference space for a metal collision sound, a metal collision sound is generated and collected under various conditions, and the reference space feature vector is calculated using these shock sound waveform data.
[0024]
On the other hand, the sound source analyzer of the present invention has a feature that data obtained by mixing a wideband noise component N with data representing the reference space is used as data for defining the reference space.
For example, when defining the reference space of a metal collision sound, the reference space feature vector is obtained by using data obtained by mixing and superimposing a broadband noise component waveform on a metal collision sound waveform generated and collected under various conditions. It is calculated.
This enables highly accurate sound source analysis even in an environment where various background sounds exist.
[0025]
For example, when analyzing the impact sound generated in office equipment such as a copying machine or a printer using the sound source analyzer according to the present invention, (1) a fan sound or a rubbing noise during paper transport, (2) office equipment is installed. Background sounds such as the noise of the surrounding environment and the air-conditioning sound.
As a result of examining the effects of these mixed sounds, if the background sound has an amplitude of 20% or less with respect to the amplitude of the impact sound to be evaluated, the effect of the mixed sound is negligibly small, but the background sound exceeds this. It has been confirmed that, when mixed, there is a great effect on the calculation result of the spatial distance.
As a result of examining the characteristics of these background sounds, it is clear that each of them is close to a noise component having a component in a relatively wide band. By mixing such a wide band noise component N in the reference space data in advance, We found that the effect of background sound could be reduced.
[0026]
As the broadband noise component N to be mixed with the reference space data, any one of white noise, pink noise, brown noise and the like can be used, but the amplitude is 1 / f due to the high similarity to the background sound. It is desirable to use pink noise in which high frequency components are attenuated.
The mixing ratio is also arbitrary, but since the ratio of the amplitude of the background sound to the main impact sound in a general copying machine or printer was measured to be 10 to 40%, in order to obtain higher robustness, It is desirable to change the mixing ratio in the range of 10 to 40%.
[0027]
Such distance calculation can be configured as software such as a computer.
Specifically, each reference space feature vector may be stored in a memory or an external storage device, and a matrix operation with a waveform vector of the sound 4 to be evaluated may be performed.
The calculated spatial distance calculation results are output to the cause identification unit 3.
[0028]
The cause identification unit 3 estimates the cause of occurrence based on the distance data output from the spatial distance calculation unit 2.
Basically, it is appropriate to determine that the sound belongs to the cause-specific space having a short distance.
At this time, it may be simply determined that the data belongs to the space having the shortest distance, or the respective distance data may be grouped in advance into higher-order groups and sequentially determined for each group.
[0029]
For example, in the spatial distance calculation unit 2, there are four collisions: (1) metal collision, (2) plastic collision, (3) sheet material (for example, paper) collision, and (4) sheet material (for example, paper) buckling. When the distances D1, D2, D3, and D4 to the cause-specific space are calculated and output, it may be simply determined that the cause belongs to the cause system indicating the smallest distance, or the cause system related to the collision of the mechanical parts may be determined. First, the group of distances D1 and D2 and the group of distances D3 and D4, which are causal systems related to the sheet material, are roughly classified, and the average distances are calculated and compared to estimate which group the group belongs to. It is also possible to use a simple estimation process.
[0030]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, the present invention will be described in detail based on embodiments shown in the accompanying drawings.
FIG. 2 shows an embodiment of a sound source analyzer to which the present invention is applied.
In FIG. 1, a sound source analyzer 10 includes a microphone 11 that captures noise from office equipment 20 such as a copying machine or a printer, an A / D (analog / digital) converter 12 connected to the microphone 11, A data storage 13 for temporarily storing the data converted by the A / D converter 12 and a computer system 19 for performing a predetermined arithmetic processing based on the data from the data storage 13 are provided.
The computer system 19 includes a computer body 19a, a keyboard 19b as input means, a mouse 19c, a display device 19d as output means, and the like.
[0031]
The hardware resources in the computer main body 19a include a CPU as operation control means, a RAM as main storage means, a hard disk as auxiliary storage means, an input / output control device (all not shown), and the like. The software resources in the main body 19a include an operating system, acoustic analysis software, numerical analysis software, and the like (all not shown).
Then, through the work of these hardware resources and software resources, the respective functional units shown in FIG. 3, specifically, the waveform vector extracting unit 14, the spatial distance calculating unit 15, the cause identifying unit 16, and the result display The unit 17 is realized.
[0032]
In the present embodiment, the microphone 11 detects a sound radiated from the office equipment 20 such as a copying machine or a printer and converts the sound into an electric signal.
The electric signal output of the microphone 11 is input to the A / D converter 12 and is converted into a digital signal.
The output of the A / D converter 12 is input to the data storage 13 and is temporarily stored.
Here, the data storage unit 13 may be appropriately selected as long as it can temporarily store data. For example, as shown in FIG. 2, a DAT (Digital Audio) having the A / D converter 12 built therein is used. In addition to using a tape recorder, an MD (Mini Disc), a hard disk device connected to a personal computer, a magnetic recording device, or the like can be used.
When the data storage device 13 has a configuration in which the A / D converter 12 is built in, the A / D converter 12 does not need to be separately prepared.
With such a configuration, it is possible to perform off-line processing in which only recording of sound is performed first, and calculation of sound source analysis is collectively performed later.
[0033]
The acoustic data stored in the data storage 13 is input to the waveform vector extraction unit 14.
The waveform vector extraction unit 14 includes a time waveform calculation unit 31, a frequency waveform calculation unit 32, and a waveform vector creation unit 33, as shown in FIG.
In the present embodiment, the time waveform calculation unit 31 includes an audibility correction unit 311, a waveform cutout unit 312, and an amplitude standardization unit 313, as shown in FIG.
[0034]
First, in the audibility correction unit 311, for example, an A-characteristic audibility correction process is performed to cut low-frequency components of several tens Hz or less that are particularly included in background noise.
Next, in the waveform cutout unit 312, a point having the maximum sound pressure value is searched from the acoustic data, and a plurality of data are respectively arranged on the front side and the rear side with reference to the position. ) Data is extracted. At this time, the numbers of front data and rear data may be appropriately allocated.
Then, the extracted waveform is normalized by an amplitude reference unit 313 so that the maximum value becomes a constant value. For example, the frequency waveform calculation unit 32 and the waveform vector creation unit 33 are used as time waveform data at n points. Output to
[0035]
In FIG. 4, the frequency waveform calculator 32 performs a window function process of applying a Hanning window function or the like to the time waveform data output from the time waveform calculator 31, and then uses the frequency analysis such as FFT analysis to perform frequency power analysis. A spectrum waveform is calculated.
The obtained predetermined number n / 2 (for example, 128) frequency power spectrum waveforms are output to the waveform vector creation unit 33.
The waveform vector creating unit 33 calculates the total number (n + n) of the predetermined number n of the time waveform data obtained from the time waveform calculating unit 31 and the predetermined number n / 2 of the frequency waveform data obtained by the frequency waveform calculating unit 32. A composite waveform of / 2 = (3/2) n = k) (for example, 384) is created and output to the spatial distance calculation unit 15 as a total k-dimensional waveform vector.
[0036]
As shown in FIG. 3, the space distance calculation unit 15 includes six individual space distance calculation units 41 to 46 corresponding to the following six types of cause-specific spaces.
As a result of carefully examining the causes of the impulsive noises that occur in ordinary copiers and printers, the causes were the first to be loud, with (a) sound caused by collision between mechanical parts and (b) sound related to paper. Can be roughly divided into
Then, regarding the collision between the mechanical components, from the material of the colliding member, it can be further classified into (a1) collision between the metal member and the plastic member, (a2) collision between the metal members, and (a3) collision between the plastic members. . In addition, regarding the sound related to paper, (b1) collision between paper and mechanical parts, (b2) buckling of paper itself, (b3) when paper passes through a step or gap during paper running, It turned out that it could be divided into popping sounds.
[0037]
It is very useful if it becomes clear which of these six types of sounds corresponds to when the generation of an impact sound becomes a problem.
For example, the sound generated around the paper positioning unit described above, (1) the sound of the mechanism mechanism for positioning, (2) the sound of the paper colliding with the positioning member, (3) the sound of the paper buckling, (4) Regarding the sound of the positioning member returning to the initial position, first, the sound of (1) corresponds to (a1) because the mechanical component (metal) collides with the positioning member (plastic), and (2) The sound of (b1) is the sound of (b), the sound of (3) is (b2), and the sound of (4) is a sound that collides with the stopper made of metal when the solenoid driving the mechanical parts returns to its original position. If it is clear which of the above six causes the sound in question is the sound corresponding to a2), the sound source can be identified with necessary and sufficient accuracy.
[0038]
Therefore, for the above-mentioned six causes, impulsive sounds are generated by simulating various generating conditions generated inside the machine, and several hundred kinds (for example, 300 kinds) of sample data of the impulsive sounds representative of the causes are collected. did.
Here, a pink noise signal whose maximum amplitude is appropriately changed in the range of 10 to 40% is artificially generated with respect to the sample data, and mixed with the sample data (mix paste) to generate reference space data. did.
The same processing as that of the waveform vector extracting unit 14 is performed on all the reference space data groups created in this way, and a predetermined number k-dimensional (for example, 384-dimensional) feature vectors Y and k × The characteristic matrix A of k (the inverse matrix of the correlation coefficient matrix) is obtained.
[0039]
This is performed for all six cause systems, and six feature vectors Ya1, Ya2,..., Yb2, Yb3 characterizing the six types of cause-specific spaces and feature matrices Aa1, Aa2,. Ab3 was determined.
The feature vector and the feature matrix are calculated in advance and stored in the memory. Each of the individual spatial distance calculation units 41 to 46 calculates, for example, a Mahalanobis distance from the feature vector and the feature matrix and the waveform vector input from the waveform vector extraction unit 14.
[0040]
Here, the Mahalanobis distance is supplementarily described.
From a plurality of reference waveforms registered in advance, a correlation coefficient matrix R of k rows and k columns corresponding to the number of variable groups k (the number of points constituting one waveform) is obtained, and its inverse matrix A is calculated.
The waveform vector of the sound to be evaluated is (x ₁ , ..., x _k ), The distance D of Mahalanobis ^Two Is calculated from equation (1) in FIG.
That is, the distance D of Mahalanobis ^Two Is calculated as a matrix product of an inverse matrix (reference space feature vector) of a correlation coefficient matrix previously obtained from a waveform vector of a sound to be evaluated and a plurality of reference waveforms.
The Mahalanobis distance calculated in this way represents the distance from the position of the center of gravity in the reference space defined by the reference waveform, and the correlation state between all the variable groups constituting the reference space is comprehensively evaluated. Therefore, even when there is a complicated correlation between the variable groups, a spatial distance that matches the actual situation can be obtained.
[0041]
As described above, when the Mahalanobis distance is calculated based on the equation (1) in FIG. 6, the six Mahalanobis distances Da1, Da2,... Calculated by the individual space distance calculation units 41 to 46 are calculated. , Db2, Db3 are output to the cause identification unit 16.
The cause identification unit 16 estimates the cause of occurrence from the six Mahalanobis distances Da1, Da2,..., Db2, Db3 output from the spatial distance calculation unit 15 as follows.
[0042]
First, there are two groups, namely, groups Da1, Da2, Da3 relating to the collision sound of the mechanical parts, and groups Db1, Db2, Db3 relating to the paper. (Db1 + Db2 + Db3) / 3 is calculated.
Then, Da and Db are compared to identify the sound belonging to the smaller group.
Next, for the group having a smaller value in the comparison between Da and Db, detailed cause identification is performed from three values of the Mahalanobis distance.
For example, if Da <Db, Da1, Da2, and Da3 are compared and identified as a sound belonging to the cause group having the smallest distance value.
The cause identification result obtained by the cause identification unit 16 in this way is output to the result display unit 17 and displayed to the user of the present apparatus.
[0043]
In the present embodiment, the waveform vector extraction unit 14 calculates and outputs a waveform vector using both a time waveform and a frequency waveform. However, the present invention is not limited to this. A time waveform vector representing pressure fluctuation may be used, a time waveform vector representing sound pressure level fluctuation may be used, or both of them may be used.
Furthermore, the waveform vector extraction unit 14 may extract only a time waveform or a frequency waveform.
[0044]
【Example】
◎ Example 1
The present embodiment is a further embodiment of the sound source analyzer according to the embodiment.
In the present embodiment, a DAT (Digital Audio Taperecorder) in which the A / D converter 12 is incorporated is used as the data storage 13, and audio data with a sampling frequency of 48 kHz is recorded.
The acoustic data stored in the data storage 13 is input to the waveform vector extraction unit 14.
At this time, in the time waveform calculation unit 31 of the waveform vector extraction unit 14, the audibility correction (A-characteristic audibility correction processing) is performed by the audibility correction unit 311. In this case, for example, as shown in FIG. 7, low frequency components of several tens Hz or less are cut.
Thereafter, in the waveform cutout unit 312, a point having the maximum sound pressure value is searched from the acoustic data, and a total of 256 data, 100 data in front and 155 data in back, are extracted based on the position. FIG. 8 shows the state at this time.
[0045]
The extracted waveform is normalized by an amplitude reference unit 313 so that the maximum value becomes a constant value, and is output to the frequency waveform calculation unit 32 and the waveform vector creation unit 33 as 256-point time waveform data. You.
Thereafter, the frequency waveform calculator 32 performs a process of applying a Hanning window function to the time waveform data output from the time waveform calculator 31, and then calculates a frequency power spectrum waveform using FFT analysis. The state at this time is shown in FIG.
The obtained 128 frequency power spectrum waveforms are output to the waveform vector creation unit 33.
Thereafter, the waveform vector creation unit 33 calculates 256 + 128 = 384 composite waveforms from the 256 time waveform data obtained from the time waveform calculation unit 31 and the 128 frequency waveform data obtained from the frequency waveform calculation unit 32. And outputs it to the spatial distance calculation unit 15 as a 384-dimensional waveform vector. FIG. 10 shows an example of the waveform vector at this time.
When a sound pressure level waveform is extracted as a time waveform, for example, the waveform after the audibility correction by the audibility correction unit 311 may be square-smoothed. For example, a sound pressure level waveform as shown in FIG. Is obtained.
[0046]
Further, in the present embodiment, with respect to the six causes of the impulsive sound, impulsive sounds are generated by simulating various generating conditions that occur inside the machine, and 300 kinds of sample data of the impulsive sound representative of the cause are generated. Collected. Here, a pink noise signal whose maximum amplitude is changed to four types of 10, 20, 30, and 40% is artificially generated with respect to the sample data, mixed with the sample data (mix paste), and mixed with a reference space. Data was created.
FIG. 12 shows an example of the impact sound sample data used here, FIG. 13 shows an example of the mixed pink noise signal, and FIG. 14 shows an example of the reference space data obtained by mixing these.
The same processing as that performed by the waveform vector extraction unit 14 is performed on all 300 pieces of data created in this manner, and a 384-dimensional feature vector Y and a 384 × 384 feature matrix (correlation coefficient) that characterize the cause space in advance. A) was obtained.
[0047]
In order to investigate the cause identification ability of the sound source analyzer according to the present embodiment, an attempt was made to identify the cause of two types of sounds whose causes are known in advance.
Specifically, among the six causes used in the present embodiment, particularly, the auditory impression and the sound pressure waveform are similar, and it is difficult to identify the cause. (1) The collision sound between the metal member and the plastic member And, (2) for the collision sound between the plastic members, each of the 100 sample sounds whose cause is apparent are collected by 100, and 80 out of the 100 collision sounds of the former metal member and the plastic member are randomly selected, A feature vector and a feature matrix representing a collision sound between a metal member and a plastic member were calculated.
[0048]
In addition, in order to verify that high-accuracy sound source analysis can be performed even in an environment where a loud background sound exists, this verification test is frequently performed by a person. The test was conducted in a general office environment.
For each of the remaining 20 collision sounds between the metal member and the plastic member and the 100 collision sounds between the plastic members, respective waveform vectors are obtained. Mahalanobis distance was determined.
[0049]
FIG. 15 shows a calculation result when a sound source analyzer (comparative example) in which pink noise is not mixed into the reference space data is shown in FIG. 15, and a calculation result according to the present embodiment in which pink noise is mixed into the reference space data is shown in FIG. .
In the former comparative example, the distribution range of the Mahalanobis distance is widened for 20 sounds for verifying the collision sound between the metal member and the plastic member and 100 sounds for verifying the collision sound between the plastic members, and the two overlap and sufficient identification is possible. In contrast, in the present embodiment, the distance between the Mahalanobis of the 20 sounds for verifying the collision sound between the metal member and the plastic member is all 2 or less, and the sound is similar to the center of gravity of the reference space. Indicated.
On the other hand, the 100 sounds for collision sound verification between the plastic members all have a Mahalanobis distance exceeding 50, which indicates that they are far from the center of gravity of this space. It was verified that it could be completely identified.
[0050]
Using such a sound source analyzer, a sound source analysis of an impact sound radiated from an actual copier / printer resulted in the occurrence of the above-mentioned six types of impact sounds with a recall rate and accuracy exceeding 80%. Confirmed that it can be identified.
[0051]
【The invention's effect】
As described above, according to the sound source analyzer of the present invention, the sound waveform itself to be evaluated is developed in a multidimensional space corresponding to the number of waveform data points, and the sound source is calculated based on the statistical distance in the multidimensional space. , It is possible to perform high-accuracy sound source analysis that combines not only limited features such as frequency information and duration but also various features of sound.
Further, according to the present invention, by mixing the wideband noise component waveform with the reference spatial data, highly accurate sound source analysis can be performed even in an environment where various background sounds exist.
[Brief description of the drawings]
FIG. 1 is an explanatory diagram showing an outline of a sound source analyzer according to the present invention.
FIG. 2 is an explanatory diagram showing one embodiment of a sound source analyzer to which the present invention is applied.
FIG. 3 is a block diagram showing each functional unit of the sound source analyzer according to the present embodiment.
FIG. 4 is an explanatory diagram illustrating a configuration example of a waveform vector extraction unit used in the present embodiment.
FIG. 5 is an explanatory diagram illustrating a configuration example of a time waveform calculation unit used in the present embodiment.
FIG. 6 is an explanatory diagram showing a mathematical expression for calculating a Mahalanobis distance used in the present embodiment.
FIG. 7 is an explanatory diagram illustrating an example of a waveform after the audibility correction processing is performed by the audibility correction unit in the example.
FIG. 8 is an explanatory diagram showing an example of a waveform around a peak portion cut out by a waveform cutout unit in the example.
FIG. 9 is an explanatory diagram showing an example of a frequency waveform obtained by performing a frequency analysis by a frequency waveform calculation unit in the embodiment.
FIG. 10 is an explanatory diagram showing an example of a waveform vector obtained by joining a time waveform and a frequency waveform in the embodiment.
FIG. 11 is an explanatory diagram showing an example of a sound pressure level waveform obtained by square-smoothing a sound pressure waveform in the example.
FIG. 12 is an explanatory diagram showing a time-sound pressure waveform of an impact sound sample for reference space data in the example.
FIG. 13 is an explanatory diagram showing a time-sound pressure waveform of a pink noise sample mixed with reference space data in the example.
FIG. 14 is an explanatory diagram showing a time-sound pressure waveform of reference space data in which pink noise has been mixed in the example.
FIG. 15 is a scatter diagram illustrating a result of discrimination between a collision sound between a metal member and a plastic member and a collision sound between plastic members by a sound source analyzer according to a comparative example.
FIG. 16 is a scatter diagram showing a result of identification of a collision sound between a metal member and a plastic member and a collision sound between plastic members by the sound source analyzer according to the embodiment.
FIG. 17 is an explanatory diagram showing an example of a time-sound pressure waveform of noise generated from a general printer.
FIG. 18 is an explanatory diagram showing a time-sound pressure waveform obtained by enlarging only the impact sound portion of FIG. 17;
[Explanation of symbols]
DESCRIPTION OF SYMBOLS 1 ... Waveform extraction part, 2 ... Spatial distance calculation part, 2a ... Reference space by cause, 3 ... Cause identification part, 4 ... Sound, 5 ... Analysis target, N ... Broadband noise component, 11 ... Microphone, 12 ... A / D Converter 13 Data storage unit 14 Waveform vector extraction unit 15 Spatial distance calculation unit 16 Cause identification unit 17 Result display unit 31 Time waveform calculation unit 32 Frequency waveform calculation unit 33 ... Waveform vector creation unit, 311 audibility correction unit, 312 waveform extraction unit, 313 amplitude reference unit, 41 to 46 individual space distance calculation unit

Claims (11)

A waveform extracting unit for calculating and extracting a waveform of a sound to be evaluated;
A spatial distance calculation unit that calculates a spatial distance of a sound waveform to be evaluated on a cause-specific reference space represented by a plurality of pre-registered reference waveforms,
A cause identification unit that identifies a feature of the sound source from the distance calculation result output from the spatial distance calculation unit,
A sound source analyzer, wherein a plurality of reference waveforms in the spatial distance calculator are mixed with a wideband noise component. The sound source analyzer according to claim 1,
A sound source analyzer, wherein the broadband noise component is pink noise. The sound source analyzer according to claim 1,
A sound source analyzer wherein a mixing ratio of the wideband noise component is changed in a range of 10 to 40% for each reference waveform. The sound source analyzer according to claim 1,
A sound source analyzer, wherein a statistical distance measure is used as the distance calculated by the spatial distance calculation unit. The sound source analyzer according to claim 4,
A sound source analyzer, wherein the distance calculated by the spatial distance calculation unit is a Mahalanobis distance. The sound source analyzer according to claim 1,
The sound source analyzer according to claim 1, wherein the spatial distance calculation unit includes two causes of spatial distance calculation units: (1) a sound generated by a collision of a mechanical component, and (2) a sound related to a sheet material. The sound source analyzer according to claim 6,
The spatial distance calculation unit is (1) a collision sound between a metal member and a plastic member, (2) a collision sound between metal members, (3) a collision sound between plastic members, (4) a collision sound between a sheet material and a mechanical component. And (5) a buckling sound of the sheet material itself, and (6) a sound distance analysis unit for each of the six causes: a sound of the end of the sheet material being repelled. The sound source analyzer according to claim 1,
The waveform extracting unit includes a time waveform calculating unit that extracts a time-sound pressure waveform, and calculates and outputs a time waveform vector representing a sound pressure variation in a time domain of a sound to be evaluated. Sound source analyzer. The sound source analyzer according to claim 1,
The waveform extracting section includes a time waveform calculating section for extracting a time-sound pressure level waveform, and calculates and outputs a time waveform vector representing a sound pressure level variation in a time domain of a sound to be evaluated. Sound source analyzer. The sound source analyzer according to claim 1,
The waveform extracting unit includes a frequency waveform calculating unit that extracts a frequency-sound pressure level waveform, and calculates and outputs a frequency waveform vector representing a sound pressure level variation in a frequency domain of a sound to be evaluated. Sound source analyzer. The sound source analyzer according to claim 1,
The waveform extracting unit includes a time waveform calculating unit that extracts a time-sound pressure waveform or a time-sound pressure level waveform, and a frequency waveform calculating unit that extracts a frequency-sound pressure level waveform, and a sound to be evaluated is provided. A sound source analyzer that calculates and outputs a waveform vector using both the time waveform and the frequency waveform.

Images