JP4127211B2 - Sound source discrimination device and its discrimination method - Google Patents

Description

  The present invention relates to a sound source discriminating apparatus and a discriminating method for discriminating the cause of a sound source whose cause of occurrence is unknown, and in particular, causes of the generation of a plurality of types of sounds included in mechanical noise radiated to the outside through a housing. The present invention relates to a sound source discriminating apparatus and a discriminating method that are effective for estimating and discriminating the noise.

  In recent years, there has been a demand for improvement in the office environment in order to support improvement in comfort and intellectual productivity in the office. In response to such a movement, for example, an image forming apparatus such as a copying machine or a printer used in an office is required to reduce noise generated from the apparatus.

  FIG. 6 shows a time waveform of noise generated during operation of a printer using a general electrophotographic system. As shown in this figure, it can be seen that the sound pressure of the noise varies with time. A waveform portion indicated by a circle in the drawing indicates a portion where a sound having a high sound pressure for a short time of about several tens to several hundred milliseconds, that is, a so-called impact sound is generated. Such a sound called an impact sound is generated by a collision between parts, a collision between a sheet and a part, and specifically, when a movable part represented by a solenoid, an electromagnetic clutch, a motor or the like is operated. This occurs at the time of a collision between the leading edge of the sheet and the alignment member in the sheet alignment unit.

  Such a sound such as an impact sound generally shows a large fluctuation in sound pressure in a short time, and therefore, it is easily recognized by humans and is likely to cause discomfort among noises.

  For this reason, when developing products that require noise countermeasures, research and development are often conducted with a focus on reducing the impact noise. However, for some of the sound source objects such as the image forming apparatus described above, it is considered extremely difficult to identify the cause of the impact sound.

  This is because sound source objects such as image forming apparatuses such as copying machines and printers have orders of magnitude more moving parts and related parts than ordinary household electrical appliances and mechanical structures. When components or the like operate at various timings, sounds are generated from various sound sources at various timings, and this is due to the feature that a plurality of types of impact sounds are mixed. In addition, in such an image forming apparatus, as the size and functionality of the image forming apparatus advance, the components are arranged in a very close positional relationship, and the components themselves become smaller and new. This is due to the feature that parts are added.

  In view of this, the present applicant has proposed a signal determination apparatus for accurately determining the cause of generation of sounds randomly generated from such many sound sources in a short time (Patent Documents 1 and 2).

  For example, the signal determination device prepares a waveform vector related to an acoustic signal obtained by collecting a plurality of types of sound from a sound source whose cause of occurrence is clear as a known waveform vector (reference space), and becomes a determination target. A waveform vector related to an acoustic signal obtained by collecting sound of a sound source whose occurrence is unknown is taken as a judgment target waveform vector, and a distance measure (for example, Mahalanobis distance) between the waveform vector and the known waveform vector is calculated. Then, by comparing the distance measure of the judgment target waveform vector with a discrimination threshold set based on the distance measure data for each cause of the known waveform vector, the sound source of which known waveform vector is the judgment target waveform vector Judgment is made by estimating whether the material is included in the group.

JP 2002-323370 A JP 2002-318155 A JP-A-9-81180

  However, such a signal determination device has the following problems.

  In other words, if the distance measure related to the sound signal of the sound source to be judged is regularly distributed in specific numerical ranges that differ for each cause of the sound source, the cause of the occurrence can be accurately determined by comparison with the above-mentioned threshold value. It is possible to judge. However, if there is a sound source with characteristics that are distributed so that the distance measure enters a part of the numerical range of the sound source with different causes, the magnitude relationship in which the distance measures between the sound source and other sound sources are reversed. (Overall, they are distributed in an overlapping state in some numerical regions), so that the cause of the sound sources cannot be determined by comparison with the threshold value.

  FIG. 11 shows the measurement result of the distance measure (Mahalanobis distance) of each sound source when the signal determination device performs sound source discrimination of the image forming apparatus. As shown in this figure, each distance measure of three sound sources of which the cause of occurrence is obvious, such as a collision sound of a metal member, a collision sound of a plastic member, and a paper sound (paper front edge collision sound and paper rear edge splash sound). Are regularly and repeatedly distributed in almost the same numerical range for all the samples. On the other hand, when looking at the distance measure of the splash sound of the paper sound, particularly the trailing edge of the paper in the conveyance direction, depending on the sample, it sometimes enters the numerical area of the distance measure of the paper leading edge collision sound. ) May be distributed as follows. For this reason, there is a case where the paper leading edge collision sound and the paper trailing edge splash sound cannot be simply determined using a certain threshold.

  Conventionally, there is a proposal related to a sound source recognition device that measures a difference in duration of sound sources in addition to identifying frequency components of sound sources (Patent Document 3). This recognizing device can recognize a sound source (for example, a short horn and a long horn) by a difference in duration of the sound signal. The duration indicates an elapsed time until the sound signal exceeds a certain voltage level and becomes lower than that.

  However, even with this recognition device, it is difficult to determine a sound source in which the distance measure as described above is randomly distributed. In other words, sound such as impact sound, for example, is known to have large differences in transient characteristics such as the rising curve of the sound pressure amplitude and attenuation polarity when the conditions of the colliding member (surface hardness, etc.) differ. Therefore, simply measuring the duration of the sound signal cannot be determined by comprehensively judging the characteristics appearing in the characteristics over time.

  The present invention has been made in view of such circumstances, and the main purpose of the present invention is that the generated sound is unstable and different due to various factors as a sound source to be identified whose cause of occurrence is unknown. To provide a sound source discriminating apparatus and a sound source discriminating method capable of appropriately discriminating the cause of sound (mainly impact sound) from such a sound source even when there is a sound source exhibiting such a characteristic. It is in.

  The sound source discriminating apparatus of the present invention includes a sound collecting unit that collects sound from a sound source and captures it as an acoustic signal, and a sound cutout unit that extracts a plurality of acoustic signals of a sound part to be determined from the captured acoustic signal. The similarity calculation unit for calculating an index indicating the similarity of the waveforms of the plurality of cut-out acoustic signals, and the calculated waveform similarity index according to the source of the sound source whose cause of occurrence is clear in advance And a determination unit configured to determine a cause of generation of a sound source to be determined in comparison with a determination threshold set based on a waveform similarity index for each acoustic signal.

  The sound collection unit can be configured by, for example, a microphone that can capture sound emitted from a predetermined sound source as an acoustic signal including an electrical signal. The sound collection unit can be configured by a data storage unit that can store and store an acoustic signal captured by the microphone, in addition to the microphone. In the case of such a configuration, only the sound recording is performed first, and then the processing relating to the analysis and calculation of the recorded acoustic waveform can be collectively performed offline. As the data storage unit, DAT (Digital Audio Tape Recorder), HD (Hard Disc) devices, MDs that can convert acoustic signals with analog / digital (A / D) exchangers and store / store them as digital signals A (Mini Disc) recorder may be used. The sound to be collected is a sound from a sound source whose cause of determination is unknown or a sound from a sound source whose cause of occurrence is clear.

  The sound cutout unit repeatedly cuts out a plurality of sound signals of the sound part to be discriminated from the sound signal taken in by the sound collection unit, and the cutout is the time when the waveform rises or the maximum amplitude part in the sound signal. This is performed for a predetermined range set based on the waveform feature on the axis. The number to be cut out may be a number that can calculate the similarity of the sound part to be discriminated, and may be two, but is preferably at least three or more. Such a sound cut-out unit can be configured by, for example, acoustic analysis software imported into a PC (personal computer), a data logger having a trigger function, or the like.

  The sound part to be determined is not limited as long as the cause of occurrence can be determined by the determination method according to the present invention, and is mainly an impact sound. The impact sound is a sound whose acoustic duration is 1 second (sec) or less, preferably 100 milliseconds (msec) or less. In order to cut out the sound portion to be discriminated, it is necessary to capture in advance (by repeatedly generating) an amount of sound that includes at least a predetermined number of the sound portions in the sound collecting portion. Not too long.

  The sound cutout unit may include an audibility correction unit that corrects the audibility of a sound signal acquired before performing the cutout by a predetermined correction filter. As the correction filter, a high-pass filter for removing low-frequency background noise components and other characteristic filters that are uniquely defined can be used. However, the A characteristic is generally said to be close to human hearing. It is preferable to use a filter. Further, the cut-out acoustic signal may be configured such that the amplitude is normalized so that the maximum value and the minimum value of the amplitude of the time waveform have a predetermined magnitude.

  The similarity calculation unit calculates an index indicating the similarity between the waveforms of a plurality of acoustic signals cut out for the sound portion to be discriminated, and the waveform similarity is determined based on the waveform characteristics of the acoustic signals (sound pressure over time). The degree of similarity may be analyzed by comparing the fluctuation characteristics or the frequency fluctuation characteristics). This similarity calculation unit also calculates a similarity index related to the waveforms of a plurality of acoustic signals obtained by collecting sounds from a sound source whose cause of occurrence is clear.

  The sound pressure fluctuation characteristic can be calculated using an analysis means such as a general acoustic analysis device or acoustic waveform analysis software. On the other hand, the frequency variation characteristic can be calculated using general frequency analysis means such as FFT analysis, wavelet analysis, and generalized harmonic analysis. Further, in the case of analyzing the similarity by the frequency variation characteristic, the time waveform to be analyzed may be subjected to window function processing using window functions such as a Hanning window, a Hamming window, and a flat top window. Good. By performing such processing, problems in frequency analysis such as aliasing error and leakage error can be reduced, and calculation of frequency characteristics in accordance with the actual situation can be performed.

  The similarity of the waveform is calculated by using a general correlation coefficient or a statistical spatial distance as an index.

  When a correlation coefficient is used as an index indicating similarity, a plurality of pieces of data (digital signal: numerical sequence or numerical matrix expressed as data for arithmetic processing) of each acoustic signal obtained by sound collection After obtaining the average, covariance, standard deviation, etc. in advance, the correlation coefficient is calculated, and the value of the correlation coefficient is used as a criterion for the presence or absence of similarity between the acoustic signals (variables). In other words, it can be seen that the closer the correlation coefficient is to 0, the less the correlation between the acoustic signals is, and the lower the degree of similarity is, and the closer the correlation coefficient is to 1, the more the correlation between the acoustic signals is. It can be seen that the degree of similarity tends to be high.

  General distance measures used in statistical discriminant analysis and cluster analysis, such as Euclidean distance, standardized Euclidean distance, Minkowski distance, Mahalanobis distance, etc. Can be used. For example, when the Mahalanobis distance is used as an index indicating similarity, the Mahalanobis distance is calculated after obtaining in advance the average, variance, covariance, etc. of the data for a plurality of acoustic signals obtained by collecting the sound, The value of the Mahalanobis distance is used as an index representing the degree of correlation between acoustic signals (variables) and used as a criterion for the presence or absence of similarity.

  As a pre-process for determining the cause of occurrence of the sound source to be determined, the determination unit obtains a waveform similarity index relating to the sound of the sound source in which the cause of occurrence is clear in advance. A sound source whose cause of occurrence is clear is a sound according to the cause of occurrence that may be radiated from a sound source to be judged (a sound caused by a movable part or a part related to its operation) in the sound collection unit. Each sound is picked up and collected as a plurality of acoustic signals, and an index indicating the waveform similarity between the acoustic signals for each cause is calculated and prepared by the similarity calculation unit. The number of sounds to be collected may be the same as the number of sounds collected from the sound source to be discriminated, or may be a smaller number. Based on all data related to the waveform similarity index of the sound source for each cause obtained in this way, an appropriate index value for distinguishing the sound source for each cause of occurrence is appropriately set as a threshold for determination. For example, when a correlation coefficient or the like is used as the similarity index, an intermediate value between the two correlation coefficients of sounds having different generation causes that indicate values close to each other can be set as a threshold value.

  Then, the determination unit estimates the cause of the generation of the sound source of the determination target by comparing the similarity index regarding the waveform of the sound source of the determination target calculated by the similarity calculation unit with the prepared threshold for determination. Judgment.

  On the other hand, the sound source discrimination method of the present invention is a sound collection step of picking up sound from a sound source and capturing it as an acoustic signal, and sound extraction for extracting a plurality of acoustic signals of a sound part to be determined from the acquired acoustic signal A similarity calculation step of calculating an index indicating the similarity between waveforms of the plurality of cut-out acoustic signals, and a cause of sound source in which the cause of occurrence is clear in advance for the calculated similarity index of waveforms A judgment step of judging the cause of the generation of the sound source to be discriminated by comparing with a discrimination threshold set based on a waveform similarity index for each of the other acoustic signals.

  Each step in this discrimination method can be realized by the operation of each functional unit (sound collection unit, sound cutout unit, similarity calculation unit, discrimination unit) in the sound source discrimination device described above.

  The sound source discriminating apparatus and the discriminating method of the present invention include, among sound sources having a plurality of types of sound sources whose cause of occurrence is unknown, in particular, a sound source exhibiting characteristics that cause the generated sound to be unstable and different due to various factors. It is effective when applied when determining the source of sound source. A sound source that does not stabilize due to various factors and includes a sound source with different characteristics, such as a sound source that has the same cause of occurrence, does not appear to be stable when it is repeatedly generated. Sound source. Examples of the sound source including such a special sound source include a plurality of types of sound sources in which a sound source whose occurrence is unknown is present inside the casing, and specifically includes, for example, the above-described image forming apparatus. When performing sound source determination for an image forming apparatus, it is possible to accurately determine sound relating to the paper used for image formation (such as a collision sound at the front end of the paper, a splashing sound at the rear end of the paper). Become.

According to such a sound source discriminating apparatus and its discriminating method, a sound source (impact sound, etc.) whose cause of occurrence is unknown is repeatedly generated and sampled in plural, and the sound signals of the repeated sounds are between each sound signal. The cause of the sound source is determined from the similarity of waveforms. In this way, for the sound of a sound source whose cause of occurrence is unknown, the cause of the sound is determined not using the waveform characteristics (sound pressure fluctuation characteristics, frequency characteristics, etc.) itself with respect to the time but using the similarity between repetitions of such waveform characteristics as an index. It will be.

  As a result, even when a sound source including a sound source having characteristics that cause the generated sound to be unstable and different due to various factors is determined, the relative characteristics of the waveform when the sound source is repeatedly generated are comparatively performed. A simple waveform comparison process enables accurate sound source discrimination.

  FIGS. 8 to 10 show time-sound pressure waveforms obtained by measuring the three kinds of sounds emitted from the copying machine twice by applying the discrimination apparatus and method of the present invention. FIG. 8 shows the waveform of the operation sound of the solenoid, FIG. 9 shows the waveform of the collision sound at the leading end of the sheet, and FIG. 10 shows the waveform of the splashing sound at the trailing end of the sheet. In addition, the correlation coefficient r is also written as an index indicating the similarity between waveforms in the margin of each figure. As can be seen from these figures, the operation sound of the solenoid has a high similarity between the two waveforms (r = 0.76), whereas the splash sound at the rear end of the paper has a low waveform similarity ( r = 0.43), it is clear that the similarity of the waveform of the collision sound at the leading edge of the sheet is at an intermediate level (r = 0.62) when the similarity of both sounds is used as a reference. It is assumed that such a difference occurs due to the following reasons.

  In other words, the operating sound of moving parts such as solenoids is stable each time the moving parts collide with each other because they are mainly composed of relatively rigid materials such as metal and plastic. As a result, it is considered that the sound generated when the component parts collide tends to be reproduced almost stably every time. On the other hand, the sound related to the paper is not stable due to the low rigidity of the paper itself. As a result, the sound generated when the sound is generated is different. This is probably due to the tendency to occur.

  As for the sounds related to the paper, first, the collision sound at the front end of the paper is generated by the vibration of the paper itself in the direction of the paper surface due to the collision phenomenon with the paper positioning component, etc., which will be described later. This is presumably because it does not occur when the paper itself is greatly deformed like a splash sound, and therefore tends to be generated as a relatively similar sound. On the other hand, the splashing sound at the rear end of the paper is generated by a behavior phenomenon that is repelled when the rear end is deformed by several mm to several tens mm and passes over a stepped portion or the like. This is because the waveform of the sound generated according to the difference in the degree of deformation tends to change. Actually, when the state of the trailing edge of the sheet conveyed through the copying machine is observed, it is confirmed that the deformation state of the trailing edge of the sheet changes every image forming cycle.

  Even in view of these facts, if the sound from the sound source to be discriminated is generated multiple times, and the similarities of the waveforms of the sound (acoustic signals) of the multiple times are examined and compared, It is clear that the cause of occurrence can be estimated and discriminated.

  As described above, according to the sound source discriminating apparatus and the discriminating method of the present invention, as a discrimination target sound source for which the cause of occurrence is unknown, a sound source exhibiting characteristics that cause the generated sound to be unstable and different due to various factors Even when there is sound, it is possible to appropriately determine the cause of the sound from such a sound source. For example, even if there is a sound source such as a splash sound at the trailing edge of the paper generated in the image forming apparatus, it is possible to properly determine the cause of the occurrence of the splash sound and the collision sound at the leading edge of the paper. It becomes.

Embodiment 1
FIG. 1 illustrates the configuration of a sound source discrimination system 100 according to an embodiment of the present invention. The sound source to be identified by the sound source discrimination system 100 is noise radiated from the copying machine T.

  In the discrimination system 100, the cause of the noise is the cause (1): the operating sound of the movable part (solenoid), the cause (2): the collision sound at the front end of the recording paper conveyance direction, and the cause (3): the recording paper. It is determined which one of the splashing sounds at the rear end in the transport direction. It is determined from the following viewpoint that the sound of the cause is generated. First of all, the causes of about 100 types of impact sounds emitted from these image forming apparatuses were carefully investigated for 20 types of existing copying machines and printers of different manufacturers and models. It has been found that a relatively large number of sounds corresponding to (42), a sound corresponding to cause (2) is “18”, and a sound corresponding to cause (3) is “25”. For this reason, when the occurrence of an impact sound in the noise of such an image forming apparatus becomes a problem, if it becomes clear which of the three types of sound corresponds, it is very important to take measures against the noise. This is because it is useful.

  The entire signal determination system 100 includes a microphone 1 as sound collection means for collecting noise from the copying machine T, a DAT2 connected to the microphone 1, and a personal computer system C connected to the DAT2. . The personal computer system C further includes a computer main body 3, a keyboard 4a as an input means, a mouse 4b, a display device 5 as an output means, and the like.

  The hardware resources in the computer main body 3 include a CPU as a calculation control means, a RAM as a main storage means, a hard disk as an auxiliary storage means, an input / output control device (all not shown), and the like. Software resources in the main body 3 include an operating system, acoustic waveform analysis software, numerical analysis software, and the like (all not shown). By the joint work of the hardware resource and the software resource, the functions of the waveform cutout unit 30, the waveform similarity calculation unit 40, and the determination unit 50 shown in FIG. 2 are realized.

  FIG. 2 is a basic functional block diagram of the discrimination system 100. The basic functions of the discrimination system 100 are, in order along the processing flow, the sound collection unit 10 composed of the microphone 1 and the like, the data storage unit 20 composed of the DAT recorder 2, and the computer main body 3. The waveform cutout unit 30, the waveform similarity calculation unit 40, the determination unit 50, and the display unit 60 configured by the display device 5. Signals input / output between these functional blocks are an analog electric signal AS, a digital signal DS, and a determination result R.

  3 is a functional block diagram showing the configuration of the waveform cutout unit 30 shown in FIG. 2 and FIG. 4 is a functional block diagram showing the configuration of the similarity calculation unit 40 shown in FIG. 2 in more detail.

  As shown in FIG. 3, the waveform cutout unit 30 includes an auditory correction unit 31, a repetitive sound detection unit 32, a repetitive waveform cutout unit 33, and an amplitude standardization unit 34. As shown in FIG. 4, the similarity calculation unit 40 calculates the average value, variance, standard deviation, and the like of the waveform vector calculation unit 41 that generates a vector of repetitive waveforms, and the waveform similarity. A correlation coefficient calculator 42 for calculating a correlation coefficient r as an index. The signals input / output between the functional blocks of the waveform cutout unit 30 and the similarity calculation unit 40 are all digital signals (specifically, a numerical sequence expressed as internal data in the PC).

  The determination unit 50 compares the correlation coefficient r (average value) calculated for the sound to be determined with a threshold value for determination that is prepared and stored in advance, and which sound source is the source of the occurrence. Determine whether. The threshold value is appropriately set based on each correlation coefficient r (average value) obtained as a similarity index regarding the repetitive waveform of the sound source for each sound source whose cause of occurrence is clear. And stored in the determination unit.

  Next, the basic operation of the sound source discrimination system 100 will be described.

  First, as shown in FIG. 5, prior to the use of the sound source discrimination system 100, pre-processing is performed in advance (step S1).

In this pre-processing, each cause (1) to (n) for which the cause is obvious is obtained as a waveform vector for a predetermined number of repetitions (m), and each cause (1) to (n) is obtained. The same occurrence occurs after calculating the average vector μ (1) to (n), the covariance vectors s (1) to (n), and the standard deviation vectors σ (1) to (n) for each repeating unit. Correlation coefficients r (1) to (n) between each repeating unit of the cause sound are obtained, and the result is stored in the determination unit 50. Here, when the correlation coefficient is expressed as a correlation coefficient r _xy of the data string (x _i , y _i ), r _xy = s _xy / (σ _x · σ _y ). Further, the determination unit 50 sets a threshold value as an intermediate value between adjacent average correlation coefficients r ′ based on the average values (r ′) of the correlation coefficients r (1) to (n) for each cause. Each is set as TH and stored.

  In this embodiment, n = 3 because three types of causes including cause (1), cause (2), and cause (3) described above are discriminated. In addition, m = 10 because 10 sampled sound signals are captured for one cause. The sound for each of the above causes (1) to (3), whose cause of occurrence is clear, was generated as follows. The cause (1) is generated by sequentially operating a plurality of solenoids having different sizes and driving forces under various conditions. Cause (2) is caused by causing various types of paper with different basis weights and sizes to collide from the front end of the paper to one side (the surface along the axial direction) of the flat plate or roll peripheral surface while changing the paper conveyance speed. It was. As for cause (3), various papers having different basis weights and sizes were passed through paper transport paths having various curvatures and steps by changing the paper transport speed, and a splash sound was generated at the rear end of the paper. .

  The sound for each cause was generated as a sound under a total of 100 conditions by variously changing the above conditions. Each sound was picked up by the repetitive quantity. In addition, the similarity calculation unit 40 performs each calculation until the correlation coefficient of each sound for each cause is obtained. The correlation coefficient of the sound of each cause is the average of the correlation coefficients of 100 sounds collected for each cause.

  When the preprocessing (step S1) is completed, the cause of the noise radiated from the sound source whose generation cause is unknown, that is, the copier is determined (steps S2 to S6 in FIG. 5).

  First, the microphone 1 collects an acoustic signal to be determined (unknown cause) (step S2). In this embodiment, the noise from the copying machine T is measured by the microphone 1 installed close to the copying machine T (at a position 10 cm away from the exterior cover which is the casing of the copying machine). This sound collection is performed so as to collect sound generated when a copying operation of one cycle of the copying machine T is executed a plurality of times. The sound picked up by the microphone 1 is converted into an electric signal and taken into the DAT 2 as an analog acoustic signal AS.

  The analog signal AS is converted into a digital signal DS by an A / D conversion circuit built in DAT2 (step S3). The digital signal at this time is once recorded on the cassette type magnetic tape of DAT2. FIG. 6 is a graph showing the digital signal DS recorded in the DAT2. In this embodiment, since the resolution (sampling frequency) of the A / D conversion circuit of DAT2 is 48 kHz, the obtained digital signal DS is obtained as 48000 points of sampling data per second. Here, when picking up an image forming apparatus such as a copying machine, it is desirable to apply a sampling frequency of at least 10 kHz or more, preferably 20 kHz or more.

  Next, after the digital signal DS recorded in the DAT 2 is taken into the main body 3 (hard disk) of the personal computer C, the waveform cutout unit 30 cuts out the sound part (impact sound) to be discriminated. (Step S4).

  First, the audibility correction unit 31 corrects the audibility of the captured digital signal DS. In this embodiment, correction is performed by an A characteristic filter that is close to human hearing in order to cut signals of low frequency components of several tens of kHz or less that are often included in background noise.

  Subsequently, for the new digital signal DS ′ whose auditory sensation is corrected, the waveform of the impact sound to be determined is detected by the repeated sound detection unit 32, and the waveform data of the impact sound is detected by the repeated waveform cutout unit 33. A predetermined number m of repetitions are cut out. In this embodiment, the position on the time axis where the sound pressure value P is the maximum value (MAX) is used as a reference, and it consists of 32 points of data existing on the front side and 31 points of data existing on the rear side. Waveform data consisting of a total of 64 points is extracted in the same manner for a predetermined number m of waveform). FIG. 7 is a graph showing the cut out one digital signal ΔDS ′. The vertical axis of this graph represents the sound pressure, and the horizontal axis represents the sample number.

Subsequently, the extracted m pieces of waveform data are standardized by the amplitude standardization unit 34 so that the maximum value of the amplitude of each waveform becomes a constant value. This is done by standardizing the digital signal ΔDS ′ based on the reference value S preset in the standardization unit 34 and the maximum value P (MAX) of the absolute value of the sound pressure included in the digital signal ΔDS ′. Obtained digital signals S _m1 , S _m2 ,..., S _mk (see FIG. 3). Here, m in “S _mk ” is the number of repetitions described above, and indicates that the signal has the m-th waveform. k indicates the number of setting conditions selected for each cause. In this example, k = 64.

Next, the digital signals S _m1 , S _m2 ,..., S _m64 cut out in this way are output to the similarity calculation unit 40 (FIG. 3), and the similarity index between the repeated waveforms is calculated ( Step S5). First, the digital signals S _m1 , S _m2 ,..., S _m64 are converted into a repetitive waveform vector of 64 rows × m columns in the waveform vector calculation unit 41. In this example, the number m of repetitions is set to m = 4, so that the waveform vector is a numerical matrix of 64 rows × 4 columns.

  Subsequently, in the correlation coefficient calculation unit 42 based on the waveform vector data, an average vector μ, a covariance vector s, and a standard deviation vector σ are calculated for each repetition unit of the waveform vector. After the correlation coefficient matrix A is calculated, an average value (average correlation coefficient) r ′ of the correlation coefficient r is obtained from the matrix A as a similarity index. The calculated average correlation coefficient r ′ is output to the determination unit 50 for each sound source to be determined.

  Next, the determination unit 50 compares the average correlation coefficient r ′ relating to the sound source to be determined output from the similarity calculation unit 40 with a preset threshold value TH. Based on the comparison result, it is determined which of the above three generation causes (1) to (3) corresponds to the sound source to be determined. The determination unit 50 determines the cause of the generation of the sound source to be determined, and obtains a determination result R relating to the cause (step S6).

In this embodiment, the result of calculating the average correlation coefficient r ′ relating to the sound of each occurrence cause (1) to (3) in the preprocessing is r ₁ ′ = 0.64 in the cause (1), and the cause (2). In this case, r ₂ ′ = 0.48, and in cause (3), r ₃ ′ = 0.32. For the threshold value TH, an intermediate value [(r ₁ ′ + r ₂ ′) / 2, (r ₂ ′ + r ₃ ′) / 2]] of the average correlation coefficient of each occurrence cause is used as the first threshold value TH1 and the second threshold value TH2. Set as each. That is, the first threshold value TH1 was (0.64 + 0.48) /2=0.56, and the second threshold value TH2 was (0.48 + 0.32) /2=0.40.

  Subsequently, in the determination unit 50, if the average correlation coefficient r ′ of the sound source to be determined input from the waveform similarity calculation unit 40 is larger than the first threshold value TH1, the sound source is the sound of the cause (1). If the average correlation coefficient r is between the first threshold TH1 and the second threshold TH2 (0.40 to 0.56), it is determined that the sound source is the sound of the cause (2). When the average correlation coefficient r is smaller than the second threshold TH2, it is determined that the sound source is the sound of the cause (3).

  Finally, the cause determination result R is displayed on the display device 5 (step S7). For example, when the sound source to be determined is the “operation sound of the movable part” of the cause (1), that effect is displayed on the screen of the display device 5 (see FIG. 2). Thereafter, if the next sound source discrimination is necessary, the above steps (S2 to S7) are similarly repeated, but if there is no subsequent sound source discrimination, the process is ended (step S8).

  Hereinafter, the result of examining the discrimination capability of the sound source discrimination system 100 according to Embodiment 1 will be described.

  First, the sound source discriminating system 100 picks up 10 types of sounds corresponding to the above causes (1) to (3) from the noise of the copying machine, and collects four repeated sounds of the respective sounds. It cut out and calculated | required the average value r 'of the correlation coefficient between each repetitive waveform, respectively. Thereafter, the respective correlation factors r ′ were compared with the first threshold value TH1 and the second threshold value TH2, thereby determining the cause of each sound.

  The discrimination results at this time are shown in Table 1. In Table 1, for example, the top column will be described as an example. The number of outputs (the number of correct answers) correctly determined as the cause (1) as the determination result of the sound (all 10 types) corresponding to the cause (1) is “8”. The number of outputs erroneously determined as the cause (2) is “1”, and the number of outputs erroneously determined as the cause (3) is “1”.

  When the reproducibility and accuracy of this discrimination result were examined, the results shown in Table 2 were obtained.

  The recall rate indicates that there are few oversights when correctly determined (correct answer). For example, in Table 2, the reproduction rate of the cause (1) is a percentage of “number of outputs correctly identified as the cause (1) (number of correct answers) / total number of sounds corresponding to the cause (1) (total number of correct answers)”. It is shown.

  On the other hand, the accuracy indicates the number of errors in the output result. For example, in Table 2, the accuracy of the cause (1) is expressed as a percentage of “number of outputs correctly determined as cause (1) (number of correct answers) / total number of outputs determined as cause (1) (8 + 1 + 0)”. ) Is indicated as a percentage of “number of outputs correctly determined as cause (2) (number of correct answers) / total number of outputs determined as cause (2) (1 + 7 + 3)”.

  From the results shown in Table 2, it was confirmed that the reproducibility exceeded 70% and the accuracy exceeded 60% regardless of the cause of occurrence, and there was sufficient discrimination ability. In particular, it has been found that it is possible to properly discriminate between a collision sound at the leading end of the paper and a splashing sound at the trailing end of the paper, which has been difficult to distinguish.

<< Embodiment 2 >>
The sound source discrimination system according to Embodiment 2 is the same as that of Embodiment 1 except that a spatial distance (for example, Mahalanobis distance) is used instead of the correlation coefficient used in Embodiment 1 as a waveform similarity index. The sound source discrimination system 100 has the same configuration.

  That is, in the sound source discrimination system 100 according to this embodiment, the correlation coefficient calculation unit 42 (FIG. 4) in the waveform similarity calculation unit 40 (FIG. 2) described in the first embodiment is as shown in FIG. The spatial distance calculation unit 43 is changed. The spatial distance calculation unit 43 calculates a spatial distance as a similarity index of the waveform of the waveform vector obtained from the waveform beltle calculation unit 41 of the waveform similarity calculation unit 40 as described later.

  Next, the basic operation of the sound source discrimination system 100 will be described. Since the operation is basically the same as that of the system according to the first embodiment (for example, up to step S4 shown in FIG. 5), the different parts will be mainly described below.

First, similarly to the case of the first embodiment, the reference-completed digital signals S _m1 , S _m2 ,..., Which are finally cut out by the waveform cutout unit 30 through the steps S1 to S4 as described above. S _m64 is input to the similarity calculation unit 40. And the said digital signal input into this similarity calculating part 40 is calculated about the similarity parameter | index between the repetitive waveforms (step S5).

First, the digital signals S _m1 , S _m2 ,..., S _m64 are converted into a repetitive waveform vector of 64 rows × m columns in the waveform vector calculation unit 41. In this example, the number m of repetitions is set to m = 4, so that the waveform vector is a numerical matrix of 64 rows × 4 columns.

Subsequently, an average vector μ, a covariance vector s, and a standard deviation vector σ are calculated for each repetition unit of the waveform vector based on the waveform vector data, and a space is calculated for each repetition unit. As the distance, Mahalanobis distance: D ² (i) is calculated for i = 1 to m, and its variation width: DR (ratio between the maximum value and the minimum value of the spatial distance value) is obtained as the similarity index. The calculated spatial distance variation DR: is output to the determination unit 50 for each sound source to be determined.

  As the calculation of the spatial distance, the Mahalanobis general distance calculation as shown below was used. In order to calculate this spatial distance, a reference space serving as a reference for calculating the spatial distance is necessary. This is set in advance as follows.

First, as a reference space, 100 sounds of the cause (1) whose cause of occurrence is obvious (moving sound of moving parts) are recorded, and the sound pressure value P of each sound is on the time axis that is the maximum value (MAX). S _i1 , S _i2 ,..., S are waveform data consisting of a total of 64 points consisting of 32 points of data existing on the front side and 31 points of data existing on the rear side. _i64 (i = 1 to 100) is created.

Next, the waveform data is subjected to frequency analysis by fast Fourier transform to obtain 32 points of frequency waveform data y _i1 , y _i2 ,..., Y _i32 . This is similarly performed for the necessary number of data i = 100 sounds to obtain y = (y _i1 , y _i2 ,..., Y _i32 ) as frequency waveform data. The frequency waveform data y (= y _i1 , y _i2 ,..., Y _i32 ) is normalized from i = 100 data. Normalization is performed based on Equation 1 below. In the formula, i represents an integer of 1 to 32, and σ represents a standard deviation.

Subsequently, the correlation coefficient matrix A and the inverse matrix R ⁻¹ of the correlation coefficient matrix A are calculated. In order to obtain a correlation coefficient matrix, a correlation coefficient r is calculated. The correlation coefficient r was calculated by the following formula 2. In the formula, y displayed as a superscript bar represents an average value of y, and p and q both represent integers of 1 to 32.

  The correlation coefficient matrix A was obtained by the following formula 3. In the formula, a represents an element obtained by calculation of an inverse matrix, and k represents an integer of 1 to 32.

The correlation coefficient matrix A in the reference space related to the cause (1) thus determined is stored as a digital data in a file that can be read on the PC. Using the correlation coefficient matrix A obtained in advance, the Mahalanobis distance D ^{2 of the} digital signals S _m1 , S _{m2 ,.} Is calculated based on Y ′ _i1 , Y ′ _i2 ,..., Y ′ _ik ) in the _equation is a frequency analysis of the digital signals S _m1 , S _{m2 ,.} Later, it is frequency waveform data obtained by further normalizing it.

By calculating more steps Mahalanobis distance, such as (distance measure) to 4 times (m = 4) repeatedly, the spatial distance (D ² 1 to D ² 4) is obtained.

Next, the determining section 50, the variation width DR of the sound source on the spatial distance of the discrimination object output from the similarity calculation unit ^{40 (D 2 1~D 2 4)} , is compared with a threshold value TH which is set in advance . Based on this comparison result, it is determined whether the sound source to be determined corresponds to one of the above three causes (1) to (3). The determination unit 50 determines the cause of the generation of the sound source to be determined, and obtains a determination result R relating to the cause (step S6).

In this embodiment, the pretreatment cause (Step S1) (1) ~ are as variation result of calculating the width DR is less than the spatial distance in the sound of each ^{(3) (D 2 1~D 2} 4) there were. For cause (1), DR ₁ = (D ² max / D ² min) = 4.3 / 2.5 = 1.7, and for cause (2), DR ₂ = (D ² max / D ² min) = 52350 / 10020 = 5.2, and in cause (3), DR ₃ = (D ² max / D ² min) = 185360/9020 = 21. D ² max represents the maximum value of the spatial distance ^{(D 2 1~D 2 4),} D 2 min denotes the minimum value of the spatial distance (D ² 1~D ² 4).

As for the threshold value TH, intermediate values [(DR ₁ + DR ₂ ) / 2, (DR ₂ + DR ₃ ) / 2]] of the variation width DR of each occurrence cause are set as the first threshold value TH1 and the second threshold value TH2, respectively. did. That is, the first threshold value TH1 was (1.7 + 5.2) /2=3.5, and the second threshold value TH2 was (5.2 + 21) / 2 = 13.

Subsequently, when the determination unit 50 has a variation width (DR ₁ , DR ₂ , DR ₃ ) regarding the spatial distance of the sound source to be determined input from the waveform similarity calculation unit 40 larger than the first threshold value TH1, If it is determined that the sound source is the sound of cause (1) and the variation range is between the first threshold TH1 and the second threshold TH2 (3.5 to 13), the sound of the sound source is cause (2) If the average correlation coefficient r is smaller than the second threshold value TH2, it is determined that the sound source is the sound of the cause (3).

  Then, the cause determination result R is displayed on the display device 5 as in the case of the first embodiment (step S7). Thereafter, if the next sound source discrimination is necessary, the above steps (S2 to S7) are similarly repeated, but if there is no subsequent sound source discrimination, the process is ended (step S8).

  Hereinafter, the result of examining the discrimination capability of the sound source discrimination system 100 according to the second embodiment will be described.

First, the sound source discriminating system 100 picks up 10 types of sounds corresponding to the above causes (1) to (3) from the noise of the copying machine, and collects four repeated sounds of the respective sounds. cut out the variation width DR of the spatial distance between each repeating waveform (D ² 1~D ² 4) were determined, respectively. Thereafter, each variation width (DR ₁ , DR ₂ , DR ₃ ) was compared with the first threshold value TH1 and the second threshold value TH2, thereby determining the cause of occurrence of each sound. Table 3 shows the discrimination results at this time. The meaning of the display in Table 3 is the same as in Table 1 in the first embodiment.

  When the current rate and accuracy of this discrimination result were examined in the same manner as in the case of Embodiment 1 (Table 2), the results shown in Table 4 were obtained.

  From the results shown in Table 4, it was confirmed that the reproducibility was 80% or more and the accuracy exceeded 70% regardless of the cause of occurrence, and that there was sufficient discrimination ability. In particular, it has been found that it is possible to properly discriminate between a collision sound at the leading end of the paper and a splashing sound at the trailing end of the paper, which has been difficult to distinguish.

1 is a conceptual diagram showing a configuration of a sound source discrimination system according to Embodiment 1. FIG. It is a functional block diagram which shows the basic function of the discrimination | determination system of FIG. It is a functional block diagram which shows the structure of the waveform cutout part shown in FIG. 2 in detail. It is a functional block diagram which shows the structure of the waveform similarity calculating part shown in FIG. 2 in detail. It is a flowchart which shows the usage and operation | movement of a sound source discrimination | determination system of FIG. It is a graph which shows the time-sound pressure waveform from a copying machine. It is a graph which shows the waveform which cut out a part of FIG. It is a time-sound pressure waveform diagram when the sound (solenoid) radiated from the copying machine is repeatedly measured twice. FIG. 6 is a time-sound pressure waveform diagram when sound radiated from a copying machine (collision sound at the leading end of a sheet) is measured twice repeatedly. FIG. 6 is a time-sound pressure waveform diagram when sound radiated from a copying machine (splash sound at the rear end of a sheet) is measured twice. It is a graph which shows the measurement result of the distance measure (Mahalanobis distance) of each sound source. FIG. 10 is a functional block diagram showing in more detail the configuration of a waveform similarity calculation unit in a sound source discrimination system according to Embodiment 2.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 10 ... Sound collection part, 30 ... Waveform cut-out part, 40 ... Waveform similarity calculation part, 50 ... Judgment part, 100 ... Sound source discrimination system (sound source discrimination device), T ... Copy machine (sound source).

Claims (10)

A similarity calculation unit that collects a plurality of sounds of sound sources whose cause of occurrence is unknown as an acoustic signal to be discriminated, and calculates an index indicating the similarity of the waveforms of the plurality of collected acoustic signals;
The calculated waveform similarity index is compared with the discrimination threshold set based on the waveform similarity index for each acoustic signal for each sound source of the sound source whose cause is clear in advance. A sound source discriminating apparatus comprising: a judgment unit that judges the cause of occurrence of the sound source. The discrimination device according to claim 1,
A sound source discriminating apparatus that uses a correlation coefficient as a waveform similarity index calculated by the similarity calculation unit. The discrimination device according to claim 1,
A sound source discriminating apparatus that uses a statistical spatial distance as a waveform similarity index calculated by the similarity calculation unit. In the discrimination device according to claim 3,
A sound source discriminating apparatus in which the statistical spatial distance is Mahalanobis distance. In the discrimination device according to any one of claims 1 to 4,
A sound source determination device in which the similarity calculation unit calculates similarity based on sound pressure fluctuation characteristics or frequency fluctuation characteristics of an acoustic signal. The sound discriminating apparatus according to claim 1, wherein the sound to be discriminated is an impact sound . The sound source determination apparatus according to claim 1, wherein the sound source is an image forming apparatus. A similarity calculation step of sampling a plurality of sound source sounds whose cause of occurrence is unknown as an acoustic signal to be discriminated, and calculating an index indicating the similarity of the waveforms of the collected acoustic signals;
The calculated waveform similarity index is compared with the discrimination threshold set based on the waveform similarity index for each acoustic signal for each sound source of the sound source whose cause is clear in advance. Judgment process to determine the cause of
A sound source discrimination method characterized by comprising: 9. The sound source determination method according to claim 8, wherein the sound to be determined is an impact sound . 10. The sound source determination method according to claim 8, wherein the sound source is an image forming apparatus.

Images