JP6294747B2 - Notification sound sensing device, notification sound sensing method and program - Google Patents

Description

  The present invention relates to technology for detecting sounds and sounds generated in the vicinity, and more specifically, alarm sounds of clocks and home appliances, chimes at the entrance, alarm sounds of fire alarms, ringtones of phones, car horns, and attention. The present invention relates to a notification sound sensing technique for sensing the occurrence of a notification sound when a notification sound such as a whistling sound is generated.

  When a signal is given to a person, a sound having characteristics different from those of a person's daily movement or a sound existing in nature is often generated. Such sounds include, for example, the sound of “Peepy” when washing of the fully automatic washing machine is finished or cooking of the microwave oven is finished, the sound of “Ping Pong” of the front door chime, and “Pupy” of the fire alarm. Etc. ”. These are collectively referred to as notification sounds.

  However, for a person with hearing impairment, even if a notification sound is generated, it cannot be heard, which not only causes inconvenience in daily life but also may cause danger to the body.

  Non-Patent Document 1 describes a method for presenting a notification sound collected by a microphone in place of this problem instead of vibration. For example, when a user wears a smartphone or dedicated device, the sound captured from the microphone is constantly analyzed by software, and the vibrator is activated when, for example, a sound with a band-limited signal with a high-pass filter is detected. Then, the sound information is converted into vibration and notified to the user. It is also possible to change the vibration pattern according to the sound generation pattern to distinguish different types of notification sounds.

Oda, Furuya, Kataoka, “Method of transmitting sound by vibration for the purpose of supporting the hearing impaired and its effectiveness”, IEICE Transactions. D, Vol. J89-D, No. 12, pp. 2671 -2678

  While the method of Non-Patent Document 1 provides a certain level of convenience to the hearing impaired, it is difficult to adjust the sensitivity with the threshold, and when the sensitivity is increased (that is, the threshold is decreased), it reacts to sounds that the user does not need. If the sensitivity is reduced (that is, the threshold value is increased), the sound required by the user may not be detected even if it is generated in the surroundings.

  In view of such a situation, an object of the present invention is to provide a notification sound detection technique capable of more strictly determining whether a sound generated in the surrounding area is a notification sound.

  In order to solve the above-described problems, the notification sound sensing device of the present invention senses a notification sound when it detects that an input acoustic signal has a predetermined number or more of frequencies where power is concentrated within a predetermined frequency band. A sensing result indicating is output.

  According to the notification sound sensing technology of the present invention, it is possible to more strictly determine whether the sound generated in the surrounding area is a notification sound. That is, it is possible to accurately detect that the notification sound has been sounded in an environment where there is a sound that accompanies everyday human movement or a sound that exists in nature. Accordingly, it is possible to accurately notify the hearing impaired person that the notification sound has been generated by a method other than sound. Further, not only for the hearing impaired service but also for the healthy person, when the user is away from the generation source of the notification sound, the notification sound sensing device is placed near the generation source of the notification sound, for example, wireless It is possible to notify the user who has left the sensing result using a communication means such as.

FIG. 1 is a diagram illustrating a functional configuration of a notification sound sensing device according to the first embodiment. FIG. 2 is a diagram for explaining the characteristics of the notification sound. FIG. 3 is a diagram for explaining the characteristics of the notification sound. FIG. 4 is a diagram for explaining the characteristics of the notification sound. FIG. 5 is a diagram for explaining the characteristics of the notification sound. FIG. 6 is a diagram illustrating a functional configuration of the notification sound sensing device according to the second embodiment. FIG. 7 is a diagram illustrating a processing flow of the notification sound sensing method of the second embodiment. FIG. 8 is a diagram illustrating a functional configuration of the notification sound sensing device according to the third embodiment. FIG. 9 is a diagram illustrating a functional configuration of the notification sound sensing device of the fourth embodiment. FIG. 10 is a diagram for explaining the characteristics of continuous sounds and intermittent sounds. FIG. 11 is a diagram conceptually showing the determination rule of the fourth embodiment.

  Hereinafter, embodiments of the present invention will be described in detail. In addition, the same number is attached | subjected to the component which has the same function in drawing, and duplication description is abbreviate | omitted.

[First embodiment]
1st embodiment of this invention is embodiment which shows the highest concept of this invention. As shown in FIG. 1, in the first embodiment, a voice or an acoustic signal (hereinafter referred to as an acoustic signal) is input, and a detection result indicating that the notification sound is detected (or the notification sound is not detected) is output. An alarm sound sensing device and method.

  The notification sound sensing device 1 according to the first embodiment reads a special program into a known or dedicated computer having, for example, a central processing unit (CPU), a main storage device (RAM), and the like. It is a special device constructed. The notification sound sensing device 1 executes each process under the control of the central processing unit, for example. The data input to the notification sound sensing device 1 and the data obtained in each process are stored in, for example, a main storage device, and the data stored in the main storage device is read out as necessary for other processing. Used. Further, at least a part of each processing unit of the notification sound sensing device 1 may be configured by hardware such as an integrated circuit.

  With reference to FIG. 2 to FIG. 5, the characteristics of an acoustic signal that is a target for detecting a notification sound will be described.

  FIG. 2 shows an example of sound recorded for a certain period of time in the office. In the example of FIG. 2, the horizontal axis represents time, the vertical axis represents amplitude, and the waveform in the time domain is illustrated. In the example of FIG. 2, the ringtone of the mobile phone is ringing during the time period indicated by the arrow 2-A. When this ring tone is expressed in characters, it feels like “beep-beep, buzz-beep-beep, blip-beep-beep, blip-beep-beep”. Other times are noise (office noise) that occurs during normal office work, and there is a steady sound, as well as sudden and loud noise as indicated by the arrow 2-B. Yes.

  FIG. 3 shows an example of sound recorded for a certain period of time in the living room. The example of FIG. 3 is a diagram illustrating waveforms in the time domain, as in FIG. At first glance, it seems that only the steady noise (floor noise) does not sound, but in reality, the kitchen timer is sounding in the next kitchen. It can be seen that the timer sound is also recorded.

  In such a situation, the notification sound sensing device 1 senses sensing when a ringtone of a mobile phone sounds in the office example of FIG. 2 and when a kitchen timer sounds in the example of living in FIG. It is required to output a result and not output a sensing result indicating sensing (or outputting a sensing result indicating non-sensing) in other time periods. However, if the prior art is used in the examples of FIGS. 2 and 3, the office example of FIG. 2 senses sudden and loud noise, and the living example of FIG. 3 senses the kitchen timer. In any case, the sensing result cannot be output accurately.

  FIG. 4 (2-A) shows the power spectrum at the time indicated by 2-A in FIG. 2, and FIG. 4 (2-B) shows the power at the time indicated by 2-B in FIG. The spectrum is illustrated. In each case, the horizontal axis represents frequency (Hz) and the vertical axis represents power (dB). FIG. 4 (2-A), which is a notification sound, has a prominent feature compared to FIG. 4 (2-B), which is sudden noise. In FIG. 4 (2-A), it can be seen that the spectrum structure has peaks near 2.8 kHz and 5.6 kHz indicated by arrows. In FIG. 4 (2-B), such a feature is not recognized.

  FIG. 5 (3-A) illustrates the power spectrum at the time indicated by 3-A in FIG. 3, and FIG. 5 (3-B) indicates the power at the time indicated by 3-B in FIG. The spectrum is illustrated. At first glance, it seems that there is no notification sound in FIG. 3, but the spectrum of FIG. 5 (3-B) has a structure having a peak near 4.1 kHz, and FIG. 5 (3-A) has such a feature. It is not allowed. Actually, the kitchen timer is sounding at the time indicated by 3-B in FIG. 3, and is not sounding at the time indicated by 3-A.

  From these observation results, in order to find a notification sound from other ambient noises, it has one or more peaks, that is, one or more frequencies where power is concentrated (hereinafter referred to as having a peak). You can see if you can tell whether or not. Moreover, in general, the notification sound is not a low sound of “boo” but a relatively high sound of “pea” in many cases, so that the second frequency is determined from a predetermined first frequency (for example, 1.0 kHz). When it is judged that the sound has a peak of 1 or more and N or less at a frequency between frequencies (for example, 6.0 kHz), a sensing result indicating sensing is output. Otherwise, a sensing result indicating sensing is output. No or a sensing result indicating no sensing is output. Although it is not essential to set the upper limit N of the number of peaks, for example, it is possible to avoid misdetecting a sound whose power is concentrated on a large number of frequencies such as a human voice as a notification sound, and more accurate notification is made. Sound can be detected. The specific value of N is preferably obtained experimentally, but can be about 6, for example.

  Therefore, the notification sound sensing device 1 of the first embodiment outputs a detection result indicating the detection of the notification sound when it detects that the input acoustic signal has a frequency at which power is concentrated in a predetermined frequency band. It is.

[Second Embodiment]
The second embodiment of the present invention is a notification sound sensing device and method embodying the functional configuration of the notification sound sensing device and method of the first embodiment.

  As shown in FIG. 6, the notification sound sensing device 1 according to the second embodiment includes, for example, a cepstrum calculation unit 10, a windowing unit 14, a second FFT unit 15, and a determination unit 16, and inputs an acoustic signal s (n). And the sensing result a is output. The cepstrum calculation unit 10 includes, for example, a first FFT unit 11, a logarithmic power spectrum calculation unit 12, and an inverse FFT unit 13.

  With reference to FIG. 7, the notification sound sensing method of 2nd embodiment is demonstrated.

  The input acoustic signal s (n) is divided into digital formats such as pulse code modulation (PCM) and is divided at regular intervals called frames. An arbitrary value may be used as the sampling frequency, but it is appropriate to sample at 16 kHz or more in order to analyze the frequency characteristic from 1.0 kHz to 6.0 kHz. Hereinafter, the case where the sampling frequency is 16 kHz will be described as an example unless otherwise specified. Any value can be used for the frame length, for example, 5 milliseconds (80 samples for 16 kHz sampling), 10 milliseconds (160 samples for 16 kHz sampling), 20 milliseconds (16 kHz sampling) 320 samples), 32 milliseconds (512 samples for 16 kHz sampling), etc. can be used.

  In step S11, the first FFT unit 11 includes a buffer for storing the input acoustic signal, and converts the input acoustic signal into a frequency spectrum S (k) using a short-time Fourier transform technique. The frequency spectrum S (k) is input to the logarithmic power spectrum calculation unit 12. The window length of the Fourier transform is longer than the frame length, and when the window length of the Fourier transform exceeds the frame length, the signal in the time domain buffer across a plurality of frames is transformed. For example, assuming that the frame length is 32 milliseconds and the Fourier transform window length is 64 milliseconds, two frames are converted together. When the window length of the Fourier transform is longer than the frame length, the time domain is overlapped in the processing of each frame. For example, when the sound signal of the i-th frame is input, the sound signals of the i-th frame and the i-th frame are Fourier transformed, and when the sound signal of the i + 1-th frame is input, the i-th frame and the i-th frame are input. Fourier transform of the acoustic signal of i + 1 frame.

  In step S12, the logarithmic power spectrum calculation unit 12 calculates a logarithmic power spectrum (hereinafter referred to as a logarithmic power spectrum) from the frequency spectrum S (k). The logarithmic power spectrum is input to the inverse FFT unit 13.

  In step S <b> 13, the inverse FFT unit 13 performs inverse Fourier transform on the logarithmic power spectrum output from the logarithmic power spectrum calculation unit 12 to return it to the signal c (n) in the time domain. The output c (n) of the inverse FFT unit 13 is called an FFT cepstrum coefficient (hereinafter simply referred to as a cepstrum). For the details of the cepstrum calculation method, a generally well-known method can be used. The cepstrum calculation method is described, for example, in “Saito, Nakata,“ Basics of Speech Information Processing ”, Ohmsha, 1981, pp. 99-103 (reference document 1)”. Reference 1 also describes a method of windowing a cepstrum described later.

  In step S14, the windowing unit 14 outputs a weighted cepstrum w (n) c (n) obtained by multiplying the cepstrum c (n) by a weight using a predetermined window function. The window function is a weight function having n as a variable, and a window defined using a trigonometric function (for example, a Hamming window, a Hanning window, etc.) can be used in addition to a rectangular window. A specific method for determining the window function will be described later.

  In step S15, the second FFT unit 15 outputs a signal Cw (k) obtained by converting the weighted cepstrum w (n) c (n) into the frequency domain again. In general, it is known that when a weighted cepstrum is converted to the frequency domain, a spectrum in which spectral characteristics are enhanced according to the weight is obtained. In the following description, the signal Cw (k) is defined as an enhanced spectrum. I will call it.

  In step S <b> 16, the determination unit 16 compares the value corresponding to each frequency k in the predetermined frequency band in the enhanced spectrum Cw (k) output from the second FFT unit 15 with a predetermined threshold value Ca and notifies the notification sound. The detection result indicating the detection of is output. Specifically, KL is a predetermined first frequency, KH is a predetermined second frequency, and the value of the enhancement spectrum Cw (k) corresponding to each frequency k in the range of KL ≦ k ≦ KH is set. If there is a value that exceeds the threshold value Ca (or may be the same as the comparison with the threshold value), a detection result a indicating the detection of the notification sound is output, and if there is no value that exceeds the threshold value Ca, the notification is made. Does not output detection result a indicating sound detection. At this time, if there is no value exceeding the threshold value Ca, a detection result a indicating that the notification sound is not detected may be output. For example, when there is a value exceeding the threshold Ca, a = 1 is output as the sensing result, and when there is no value exceeding the threshold Ca, a = 0 is output.

  The values of the first frequency KL, the second frequency KH, and the threshold value Ca are all determined by the range where the notification sound may exist. Since the specific notification sound is standardized by JIS guidelines and the like, it may be set according to the frequency range of the notification sound described therein. Moreover, you may set according to the notification sound designed uniquely. The first frequency KL can be set to 1.0 kHz, for example. The second frequency KH can be set to, for example, 6.0 kHz.

  In the processing of the windowing unit 14, the cepstrum c (n) is weighted by emphasizing the frequency component in which power is concentrated in the examples of FIGS. 4 (2-A) and 5 (3-B). This is because it is possible to stably determine whether or not the sound has a peak at one or more and N or less frequencies. As for the value of cepstrum c (n), it is known that the region where n is small represents the inclination of the spectrum or a gentle outline, and the fine structure of the spectrum is represented as n increases. It is also known that by multiplying a weighting factor corresponding to the value of n, the outline of the spectrum can be emphasized or removed, and the fine structure of the spectrum can be emphasized or removed.

An object of the present invention is to detect a notification sound, that is, to detect the presence of a signal in which power is concentrated at a specific frequency. In the examples of FIG. 4 and FIG. 5, it is considered that the inclination of the spectrum and the gentle outline are caused by ambient noise other than the notification sound. Therefore, in the second embodiment, a window function is used in which the weight of the region where n is small is smaller than that of the other regions. As an example of the window function,
When n ≦ Nc, w (n) = 0
Otherwise, w (n) = 1
A square window such as can be used. For example, 10 can be used as the value of Nc.

As another example of the window function, Nc <Nh,
When n ≦ Nc, w (n) = 0
When n ≧ Nh, w (n) = 0
Otherwise, w (n) = 1
As described above, the window function may be defined so that the weight of the region where n is large in addition to the region where n is small is also small. The reason for excluding the region where n is large is that the peak frequency of the notification sound has a certain range, and the fine structure of the spectrum that is finer than that is also removed, so that stable notification sound can be detected. This is because it is considered. When the frame length is 32 milliseconds (512 samples) and the Fourier transform window length is 1024 samples, the value of Nh can be set to about 100 to 400, for example.

  In the above example, a rectangular window with w (n) = 0 or w (n) = 1 was used, but for example, a window function in which the value of w (n) changes gradually depending on the value of n may be defined. Good. For example, a window can be defined using a trigonometric function so that w (n) becomes small in a region where n is small and a region where n is large. Typical examples of such window functions include a Hamming window and a Hanning window.

[Third embodiment]
The third embodiment of the present invention is a notification sound sensing device and method embodying the functional configuration of the determination unit 16 of the second embodiment.

  As shown in FIG. 8, the determination unit 16 of the third embodiment includes, for example, a peak detection unit 161, an overall determination unit 162, and a memory unit 163, and receives the enhancement spectrum Cw (k) as an input, and outputs a detection result a. .

  Hereinafter, the process performed by the determination unit 16 according to the third embodiment will be described focusing on differences from the second embodiment.

  The peak detector 161 compares the value of the enhancement spectrum Cw (k) corresponding to each frequency k in the frequency band in the range of KL ≦ k ≦ KH with the threshold value Ca, and when there is a value exceeding the threshold value Ca, A value of 1 or more is output as the peak detection result a0, and when there is no value exceeding the threshold value Ca, a value of 0 is output as the peak detection result a0. The value of the peak detection result a0 may simply be a value indicating the presence or absence of a peak, or may be the number of peaks. When setting the value to indicate the presence or absence of a peak, the value of the emphasized spectrum Cw (k) is examined, and if there is a value exceeding the threshold value Ca, the peak detection result a0 = 1 is output and there is no value exceeding the threshold value Ca The peak detection result a0 = 0 is output. A method for obtaining the number of peaks will be described in detail in the fourth embodiment. The peak detection result a0 is input to the comprehensive determination unit 162 and the memory unit 163.

  The memory unit 163 accumulates the value of the peak detection result a0 over a predetermined time (the number of frames), and sends the time series to the general determination unit 162. The number of frames may be set to a number based on the length of the notification sound that may exist and the frame length.

The overall judgment unit 162 uses the time series of the peak detection result a0 of the current frame and the value of the past peak detection result a0 to determine whether a notification sound that should be detected based on a predetermined rule has occurred. A sensing result a indicating is output. Examples of predefined rules include
1. It is determined that the notification sound is generated when the X detection is continuously performed for X frames or more (for example, when X = 10 and the frame length is 32 milliseconds for 320 milliseconds or more) and the peak detection result a0 is 1 or more.

Or
2. It is determined that a notification sound has been generated when the peak detection result a0 is 1 or more within Y frames or more (for example, Y = 40) within a predetermined past predetermined time T (for example, within the past 6 seconds).

  That is, the comprehensive determination unit 162 does not determine whether the notification sound is generated based on the frequency analysis result of one frame alone, but based on the analysis result (continuity, frequency) within a predetermined past predetermined time. Comprehensive judgment of whether or not has occurred.

[Fourth embodiment]
The fourth embodiment of the present invention is a notification sound sensing device and method embodying the functional configurations of the peak detection unit 161 and the comprehensive determination unit 162 of the third embodiment.

  As shown in FIG. 9, the peak detection unit 161 of the fourth embodiment includes a peak number detection unit 1611 and a peak frequency detection unit 1622, for example, and receives the enhanced spectrum Cw (k) as an input, and the peak number an and the peak center frequency. Output ak.

  As shown in FIG. 9, the overall determination unit 162 of the fourth embodiment includes, for example, a continuous sound determination unit 1621 and an intermittent sound determination unit 1622, and receives a time series of the peak number an and a time series of the peak center frequency ak. The detection result a is output.

  Hereinafter, the process performed by the determination unit 16 according to the fourth embodiment will be described focusing on differences from the third embodiment.

  The peak number detector 1611 compares the value of the enhanced spectrum Cw (k) corresponding to each frequency k in the frequency band in the range of KL ≦ k ≦ KH with the threshold value Ca, and when there is a value exceeding the threshold value Ca. Find the peak number an. Here, the number of peaks is not the number of k exceeding the threshold, but the value of the enhanced spectrum Cw (k) corresponding to each frequency k is examined in order from the smaller (or larger) k, and the threshold Ca is not exceeded. From the state, the value of the emphasized spectrum Cw (k) gradually increases, and the number of peaks with one peak from the frequency ks exceeding the threshold Ca to the frequency ke immediately before the threshold Ca is not exceeded next. is there.

  The peak frequency detector 1622 compares the value of the enhanced spectrum Cw (k) corresponding to each frequency k in the frequency band in the range of KL ≦ k ≦ KH with the threshold value Ca, and when there is a value exceeding the threshold value Ca. The peak center frequency ak is obtained. Here, the peak center frequency is a frequency at an intermediate point between the frequency ks exceeding the threshold value Ca and the frequency ke immediately before the threshold value Ca is not exceeded at each peak obtained by the peak number detection unit 1611, or ks ≦ This is the frequency at which the value of the enhanced spectrum Cw (k) is maximized when k ≦ ke. When the peak number an is 2 or more, a plurality of peak center frequencies ak are obtained corresponding to each peak.

  When there is no value of the enhanced spectrum Cw (k) exceeding the threshold value Ca in the range of KL ≦ k ≦ KH, the peak number an is set to 0 and the peak center frequency ak need not be obtained. For example, the peak center frequency ak may be set to 0 in terms of software implementation.

  The memory unit 163 accumulates the values of the peak number an and the peak center frequency ak over a predetermined time (number of frames), and sends the time series to the comprehensive determination unit 162.

  The continuous sound determination unit 1621 uses the time series of the value of the peak number an in the current frame and the value of the peak number an in the past frame to notify that the continuous sound determination unit 1621 should sense based on a predetermined continuous sound detection rule. Outputs a detection result a indicating whether or not sound is generated. A continuous sound is a notification sound in which a sound whose power is concentrated at a specific frequency continues.

  The intermittent sound determination unit 1622 uses the value of the peak number an and the peak center frequency ak in the current frame and the time series of the peak number an and the value of the peak center frequency ak in the past frame in advance. A detection result a indicating whether a notification sound to be detected is generated based on a rule for detecting the intermittent sound is output. An intermittent sound is a notification sound in which a power-concentrated sound is intermittently continued at a specific frequency.

  The overall determination unit 162 may operate both the continuous sound determination unit 1621 and the intermittent sound determination unit 1622 in parallel, or when the continuous sound determination unit 1621 is first operated and it is not determined that the notification sound has occurred. Alternatively, the intermittent sound determination unit 1622 may be operated.

  If it is determined that at least one of the continuous sound determination unit 1621 and the intermittent sound determination unit 1622 has generated the notification sound, the overall determination unit 162 outputs a sensing result indicating the generation of the notification sound. Further, when it is determined that neither the continuous sound determination unit 1621 nor the intermittent sound determination unit 1622 determines that the notification sound has occurred, a detection result indicating that the notification sound is not detected is output.

  The reason why the detection rules for the continuous sound and the intermittent sound are set differently is to detect the notification sound without any error.

  Examples of continuous sounds include a notification sound (for example, an operation end sound of a washing machine, a microwave oven, etc.) that has the same sound "Pee" for a certain period of time, an interphone ring tone that changes the pitch of "Pean Pawn", A fire alarm sound whose pitch continuously changes from “Pew Pew”, a ring tone of a telephone with a complicated sound quality such as “Tururu Lulu” or “Chiri Lillin” is assumed. Even if these sounds do not necessarily continue the same sound, and the sound changes in pitch, a person recognizes that the sound is a notification sound by continuing the sound in which power is concentrated at a specific frequency. Note that the change in pitch means that the frequency at which power is concentrated and the number of peaks change over time.

  As an example of the intermittent sound, an alarm clock, a kitchen timer, a ringtone of a mobile phone, etc. are assumed. In these cases, the time from the beginning of the sound to the end of the sound continues for a certain time or more, but the time when the sound is emitted once is very short. However, when the same sound sounds regularly and intermittently, a person recognizes it as a notification sound as well as a continuous sound.

  In the method of the third embodiment, when trying to detect a continuous sound and an intermittent sound with the same rule, “a notification sound is generated when the number of peaks a0 is 1 or more over a Y frame or more within a predetermined past predetermined time T. The value of T and Y when setting the rule “determined to have been determined” must be set to a value at which intermittent sounds can be detected, that is, Y with respect to the value of T (that is, the number of frames corresponding to T) The value of must be set small enough. However, if the setting is made in this manner, the daily sound may be erroneously determined as the notification sound and the erroneous detection result may be output even though the notification sound is not sounding. In particular, tableware and television program sounds are often mistaken as notification sounds. In order to prevent this, it is necessary to determine whether or not the same sound is regularly and intermittently sounding.

  On the other hand, in continuous sound detection, if the restriction that the same sound is continuously played is applied, it will not be possible to detect a sound whose pitch changes like an intercom or fire alarm sound. However, anyway, if a sound whose power is concentrated at a specific frequency is continuously sounding, it may be determined as a notification sound.

An example of a rule for detecting continuous sounds is:
1. It is determined that the notification sound is generated when the number of peaks an is equal to or greater than X frames continuously (for example, when X = 10 and the frame length is equal to or greater than 320 milliseconds when the frame length is 32 milliseconds).

Or
2. It is determined that a notification sound is generated when the number of peaks an is 1 or more within Yc frames (for example, Yc = 20) within a predetermined past fixed time Tc (for example, within the past 1 second).

  That is, the continuous sound determination unit 1621 does not determine whether a continuous sound is generated based on the frequency analysis result of one frame alone, but continuously based on analysis results (continuity, frequency) within a predetermined past predetermined time. Judgment is made as to whether or not sound has occurred.

Examples of rules for detecting intermittent sounds include
1. Within a predetermined past predetermined time Tb (for example, within the past 6 seconds), there are frames having Yb frames or more (for example, Yb = 10), the number of peaks an and the peak center frequency ak being the same value other than 0. Sometimes it is determined that a notification sound has occurred.

Or
2. Within a predetermined past predetermined time Tb (for example, within the past 6 seconds), the number of peaks an and the peak center frequency ak are not the same value, or the difference between the values is within a predetermined tolerance (hereinafter, generic name) If the time difference between the frames (frame number difference) is less than or equal to a predetermined value, the same sound is played continuously between the frames. The generation of the notification sound is determined by applying the above-described rule for detecting continuous sounds. The above “once it is assumed that the same sound is continuously played between the frames” means, for example, the number of peaks an and the center frequency ak of the peaks of all frames between the frames. The processing may be replaced with the number an and the center frequency ak of the peak. Or, there are actually only two of the frames in which the number of peaks an is 1 or more, but when counting X and Yc of the rule for detecting the continuous sound, the difference of the frame number to the number of the frames Strictly speaking, it may be a process of adding 1 to the number of the frames and subtracting 1 by 1 (to overlap one frame). Note that the thresholds X, Tc, and Yc when applying the rule for detecting continuous sounds can be set to values different from those of the continuous sound determination unit 1621.

  That is, the intermittent sound determination unit 1622 does not determine whether or not an intermittent sound has occurred based on the frequency analysis result of one frame, but intermittently based on analysis results (continuity and frequency) within a predetermined past predetermined time. Judgment is made as to whether or not sound has occurred.

  The difference between the third embodiment and the fourth embodiment will be described in more detail. FIG. 10 illustrates the center frequency of peaks of (A) continuous sounds whose pitches change, (B) intermittent sounds, and (C) sounds other than notification sounds (life noise, etc.). These are examples in which the number of peaks an is 1, the horizontal axis represents time, and the vertical axis represents the peak center frequency. When the number of peaks an is 2 or more, there are two or more lines (horizontal lines in the figure) indicating the peak center frequency at the same time. Here, for simplicity, an is 1 Only an example will be described. A time zone without a line indicating the peak center frequency indicates that there is no peak. In the third embodiment, when the peak detection result a0 is a value indicating the presence or absence of a peak (a0 = 1), in the example of FIG. 10, when it is determined that a notification sound has occurred in (A) and (B). , (C), it is determined that a notification sound has occurred. This is because within a certain time T, the number of frames with a0 = 1 is larger in (C) than in (B). FIG. 11 conceptually shows the determination rule of the intermittent sound determination unit 1622 in the fourth embodiment, and (B) is regarded as a continuous sound such as (B ′). In the fourth embodiment, only the continuous sound determination unit 1621 determines (A) as the notification sound, and the intermittent sound determination unit 1622 determines (B) as the notification sound. Therefore, in the case of (A) and (B), it is determined that the notification sound is generated, and (C) is not determined as the notification sound.

[Fifth embodiment]
The fifth embodiment of the present invention is a modification of the peak detection unit 161 and the comprehensive determination unit 162 of the fourth embodiment.

  In general, the notification sound is a high frequency sound of 1.0 kHz or more. Therefore, in the peak detection unit 161 of the fourth embodiment, it is further checked whether or not there is a value of the enhanced spectrum Cw (k) that exceeds the threshold value Ca in the range of k <KL, and a frame that has a value that exceeds the threshold value in the range. Is excluded from the determination target. To exclude from the determination target, for example, the number of peaks an in the frame is set to 0.

  As a result, the probability that a sound that is not a notification sound is mistakenly detected as a notification sound is further reduced.

  The present invention is not limited to the above-described embodiment, and it goes without saying that modifications can be made as appropriate without departing from the spirit of the present invention. The various processes described in the above embodiment may be executed not only in time series according to the order of description, but also in parallel or individually as required by the processing capability of the apparatus that executes the processes or as necessary.

[Program, recording medium]
When various processing functions in each device described in the above embodiment are realized by a computer, the processing contents of the functions that each device should have are described by a program. Then, by executing this program on a computer, various processing functions in each of the above devices are realized on the computer.

  The program describing the processing contents can be recorded on a computer-readable recording medium. As the computer-readable recording medium, for example, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory may be used.

  The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its own storage device. When executing the process, the computer reads a program stored in its own recording medium and executes a process according to the read program. As another execution form of the program, the computer may directly read the program from a portable recording medium and execute processing according to the program, and the program is transferred from the server computer to the computer. Each time, the processing according to the received program may be executed sequentially. Also, the program is not transferred from the server computer to the computer, and the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes the processing function only by the execution instruction and result acquisition. It is good. Note that the program in this embodiment includes information that is used for processing by an electronic computer and that conforms to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer).

  In this embodiment, the present apparatus is configured by executing a predetermined program on a computer. However, at least a part of these processing contents may be realized by hardware.

DESCRIPTION OF SYMBOLS 1 Notification sound detection apparatus 10 Cepstrum calculation part 11 1st FFT part 12 Logarithmic power spectrum calculation part 13 Inverse FFT part 14 Windowing part 15 Second FFT part 16 Determination part 161 Peak detection part 1611 Peak number detection part 1612 Peak frequency detection part 162 General determination unit 1621 Continuous sound determination unit 1622 Intermittent sound determination unit 163 Memory unit

Claims (6)

A first FFT unit that converts an input acoustic signal into a frequency spectrum;
A logarithmic power spectrum calculator for calculating a logarithmic power spectrum from the frequency spectrum;
An inverse FFT unit for converting the logarithmic power spectrum into a cepstrum;
A windowing unit that generates a weighted cepstrum weighted to the cepstrum using a predetermined window function;
A second FFT unit for generating an enhanced spectrum obtained by converting the weighted cepstrum into a frequency domain;
A determination unit that outputs a sensing result indicating sensing of the notification sound by comparing the value of the enhanced spectrum corresponding to each frequency within a predetermined frequency band with a predetermined threshold;
Including
The determination unit is
The value of the enhanced spectrum corresponding to each frequency in the frequency band is compared with the threshold value in ascending or descending order of frequency, and the frequency from the frequency exceeding the threshold to the frequency immediately before the threshold value is not exceeded is 1 A peak number detector for obtaining the number of peaks as one peak;
A peak frequency detection unit for obtaining a peak center frequency from a frequency between a frequency that exceeds the threshold for the peak and a frequency immediately before the threshold is not exceeded, and
A memory unit for accumulating the number of peaks and the peak center frequency and outputting a time series of the number of peaks and the peak center frequency;
A continuous sound determination unit that verifies the time series of the peak number and the peak center frequency, and determines whether or not a continuous sound in which the power concentrates at a specific frequency is generated;
An intermittent sound determination unit that verifies the time series of the peak number and the peak center frequency, and determines whether or not an intermittent sound in which power concentrates at a specific frequency is intermittently generated;
Including
The intermittent sound determination unit is
When the number of peaks an and the peak center frequency ak are the same value except for 0 in a predetermined number of frames within a predetermined time,
Or
If the same number of frames are extracted with the peak number an and the peak center frequency ak being other than 0 within a predetermined time, and the time difference between the frames is equal to or less than a predetermined value, the frames are continuously transmitted over a predetermined number of frames. When the peak number an is 1 or more,
Or
When the same number of frames are extracted with the peak number an and the peak center frequency ak being other than 0 within a predetermined time, and the time difference between the frames is equal to or less than a predetermined value, the predetermined number or more is exceeded within the predetermined time. When the number of peaks an is 1 or more in a frame,
A notification sound sensing device for determining that the intermittent sound has occurred. The notification sound sensing device according to claim 1,
The continuous sound determination unit
When the peak number an is 1 or more continuously in a predetermined number of frames or more,
Or
When the peak number an is 1 or more in a predetermined number of frames within a predetermined time,
It is determined that the above continuous sound has occurred.
Notification sound sensing device. The notification sound sensing device according to claim 1 or 2,
The determination unit determines whether the intermittent sound has occurred in the intermittent sound determination unit and outputs the sensing result when the continuous sound determination unit does not determine that the continuous sound has occurred. A comprehensive judgment unit
Notification sound sensing device. The notification sound sensing device according to claim 3 ,
The said comprehensive determination part is excluded from the object of verification when the value of the said enhancement spectrum exceeds the said threshold value in the frequency outside the range of the said frequency band. A first FFT step of converting the input acoustic signal into a frequency spectrum;
A logarithmic power spectrum calculating step for calculating a logarithmic power spectrum from the frequency spectrum;
An inverse FFT step of converting the log power spectrum into a cepstrum;
A windowing step for generating a weighted cepstrum weighted to the cepstrum using a predetermined window function;
A second FFT step for generating an enhanced spectrum obtained by converting the weighted cepstrum into a frequency domain;
A determination step of comparing the value of the enhanced spectrum corresponding to each frequency within a predetermined frequency band with a predetermined threshold value and outputting a sensing result indicating sensing of the notification sound;
Including
The determination step includes
The value of the enhanced spectrum corresponding to each frequency in the frequency band is compared with the threshold value in ascending or descending order of frequency, and the frequency from the frequency exceeding the threshold to the frequency immediately before the threshold value is not exceeded is 1 A peak number detection step for obtaining the number of peaks as one peak,
A peak frequency detection step for obtaining a peak center frequency from a frequency between a frequency that exceeds the threshold for the peak and a frequency immediately before the threshold is not exceeded, and
A memory step for accumulating the number of peaks and the peak center frequency and outputting a time series of the number of peaks and the peak center frequency;
A continuous sound determination step that verifies the time series of the peak number and the peak center frequency, and determines whether or not a continuous sound in which power concentrates on a specific frequency is generated;
An intermittent sound determination step for verifying the time series of the peak number and the peak center frequency and determining whether or not an intermittent sound in which power concentrates on a specific frequency continues intermittently occurs;
Including
The intermittent sound determination step includes
When the number of peaks an and the peak center frequency ak are the same value except for 0 in a predetermined number of frames within a predetermined time,
Or
If the same number of frames are extracted with the peak number an and the peak center frequency ak being other than 0 within a predetermined time, and the time difference between the frames is equal to or less than a predetermined value, the frames are continuously transmitted over a predetermined number of frames. When the peak number an is 1 or more,
Or
When the same number of frames are extracted with the peak number an and the peak center frequency ak being other than 0 within a predetermined time, and the time difference between the frames is equal to or less than a predetermined value, the predetermined number or more is exceeded within the predetermined time. When the number of peaks an is 1 or more in a frame,
Determine that the intermittent sound has occurred
Notification sound detection method. Program for causing a computer to function as the notification sound sensing device according to any one of claims 1 to 4.