JPH0222799A - Acoustic recognizing device - Google Patents

Abstract

PURPOSE:To cause the title device to accurately recognize sounds even when the frequency of the sounds changes as time elapses by recognizing the sounds from the shape of frequency characteristics of the sounds associated with hourly changes. CONSTITUTION:Detected sounds are converted into acoustoelectric signals by means of an acoustoelectric converter M1 and checked by a frequency analyzing means M2 for hourly frequency characteristics. A peak extracting means M3 extracts peaks of the frequency characteristics at every hour from the hourly given frequency characteristics. The peak information is divided into groups at every hourly continuous group by means of a grouping means M4. The grouped peak information is inputted to a feature quantity extracting means M5 and the means 5 extracts the feature quantity of the characteristic shape of the continuous peak groups. The information of the feature quantity is inputted to an identifying means M6 and prescribed sounds are recognized based on the feature quantity and the reference feature quantity of the acoustic signals to be recognized stored in a reference feature quantity storing means M7.

Description

[Detailed description of the invention] [Industrial application field] The present invention relates to an acoustic recognition device that recognizes special sounds from ambient sounds. In particular, it is used in a device mounted on a car or the like to recognize emergency vehicle warning sounds from ambient sounds outside the car and notify the driver. [Prior art]

2. Description of the Related Art Conventionally, the following acoustic recognition devices are known, which are installed in automobiles and which recognize alarm sounds from emergency vehicles, railroad crossing alarms, etc. from ambient sounds outside the vehicle interior and notify the driver. The first is a device that determines whether or not the spectrum of the detected sound includes a fundamental frequency that is characteristic of the alarm sound. Specifically, 2 to 3 characteristic frequency components are extracted from the sound V (ff) using a Vantobayu filter whose center frequency is 2 to 3 reference frequencies that are characteristic of the alarm sound to be recognized. If the magnitude of the component exceeds a predetermined threshold value, 3.) It is determined that an alarm sound is present among all the sounds produced (Japanese Patent Laid-Open No. 58-221500). The second method measures the period, or pitch, of the detected sound, and compares the ratio of each pitch with the pitch of a predetermined warning sound, or uses the stability of the pitch to recognize the warning sound of an emergency vehicle. (Japanese Unexamined Patent Publication No. 60-219521). The third method is to extract multiple peak frequency components of the alarm sound from the detected sound using a plurality of band-pass filters whose passing frequencies are fixed peak frequencies of the spectrum of the alarm sound to be recognized. The alarm sound is recognized by comparing the temporal change shape of a plurality of peak frequency components with the time change shape of the peak frequency component of the alarm sound registered in advance (Japanese Unexamined Patent Publication No. 175238/1983). .

[Problems to be Solved by the Invention] However, since the first device described above makes a judgment by focusing on the absolute values of two to three fixed frequency components of the detected sound at each time, it is difficult to generate the alarm sound. If the spectral shape of the detected sound changes due to individual body differences, noise, and the Doppler effect, there is a problem that it cannot be recognized. Furthermore, in the second device, if other sounds such as the sound of a car running, wind noise, or the sound of other cars are mixed in, the pitch period will be greatly affected, making it difficult to recognize the alarm sound. Further, in extracting the peak frequency component of the spectrum shape of the detected sound, the third device uses a plurality of bandpass filters whose passbands are fixedly set in advance in correspondence with each alarm sound. If the detected sound includes an alarm sound whose frequency has been shifted due to the above-mentioned reason, there is a problem that the shifted peak frequency component is not detected. Furthermore, a large number of bandpass filters are required to detect the peak frequency for each alarm sound, making the device complicated. Furthermore, since the shape of the temporal change of the extracted frequency component is directly compared with the shape of the temporal change of the peak frequency of the alarm sound registered in advance, a large amount of data to be matched is required, and the calculation time is long. Another problem is that it takes time to recognize sounds. In addition, especially for sound sources whose peak frequency changes over time, such as fire engine sirens, the extraction frequency of the above device is fixed, so changes in the peak frequency cannot be detected. , accurate recognition is not possible. By the way, human sound discrimination ability is insensitive to absolute values of frequency and sound pressure, but sensitive to relative changes in frequency and sound pressure. Therefore, while conventional devices classify a given sound by focusing on the absolute value of frequency or sound pressure, the present invention focuses on the characteristics of human sound discrimination ability. The sound 9 recognition is performed by extracting the peak frequency and relative temporal changes in its amplitude as feature quantities.

[Means to solve the problem]

As shown in FIG. 1, the means constituting the invention include a speech-to-speech electrical converter M1, a frequency analysis means M2, and a peak extraction means M3.
, grouping means M4, feature extraction means M5, identification means M6,
It consists of a basic feature amount storage means M7. The above configuration means is achieved by a means configured by hardware or software, and the frequency analysis means M2 calculates the frequency characteristics at each time of the acoustic signal outputted from an acoustoelectric transducer such as a microphone with the time as a variable. It is a means of seeking. Specifically, it is possible to obtain frequency components while scanning frequencies and wave numbers using a digital band-pass filter with a steep passband that is variable in frequency, or to use a Fourier transformer that performs Fourier transform on an acoustic signal. The peak extraction means M3 is means for extracting peaks from the frequency characteristics. Specifically, in terms of frequency characteristics, it can be configured with a differentiator that differentiates with respect to frequency by differential calculation. The grouping means M4 is a means for determining continuity of peaks with respect to time from the time sequence of peaks outputted by the peak extraction means M3, and classifying the peaks into consecutive peak groups. Specifically, it is a means of grouping peaks into consecutive peaks depending on whether the frequency divisions in which the peaks exist are continuous over time, and as is well known, it is similar to a line segment extractor in image processing. It can be configured by the following means. The feature extracting means M5 is a means for extracting feature amounts of the characteristic shapes associated with temporal changes in the grouped peak groups. Specifically, for each peak group and for each partial shape of each peak group, feature quantities such as the amount of increase or decrease in frequency, the amount of increase or decrease in amplitude, and the duration of the partial shape are extracted. It is a means to do so. The reference feature amount storage means M7 is a means for storing the feature amount of a predetermined sound to be recognized as a reference feature amount in correspondence with the feature amount of the detected sound. The identification means M6 is a means for identifying a predetermined sound based on the feature amount extracted by the feature area extraction means M5 and the reference feature amount stored in the base feature storage means M7.

[Effect]

The detected sound is converted into an acoustic signal, which is an electric signal, by the sound alarm electric converter, and the sound U signal is input to the frequency analysis means M2, which determines the frequency characteristics at each time. The frequency characteristics given at each time are input to the peak extraction means M3, and the peak extraction means M
3, the peak of the frequency characteristic at each time is extracted. The peak information is input to the grouping means M4, and the grouping means M4 groups the information into temporally continuous peaks. Information on the divided peak groups is available on special! fiffi extraction means M5, and the characteristic amount of the characteristic shape of the continuous peak group is extracted. Then, the information of the feature m is inputted to the identification means M6, and the A predetermined sound is identified based on the feature amount and the basic feature (2) of the acoustic signal to be recognized, which is recorded in the reference feature storage means M7.

[Examples] The present invention will be described below based on specific examples. FIG. 2 is a block diagram showing the configuration of the embodiment device. The vehicle alarm device 2 of this embodiment includes an audio signal input unit 6 that samples and stores an audio signal from a microphone 4 that picks up sounds from outside the vehicle for a predetermined period of time, and a unit that analyzes the sampled audio signal. A high-speed arithmetic processing unit 8 performs predetermined arithmetic processing at high speed, and the audio signal sampled by the audio signal input unit 6 is input to the high-speed arithmetic processing unit 8 to perform predetermined arithmetic processing, and the result of the calculation is input to the high-speed arithmetic processing unit 8. an alarm sound identification section 10 that identifies various alarm sounds from emergency vehicles, railroad crossing gates, etc. from external sounds picked up by the microphone 4 based on the above information, and outputs an identification signal representing the identification result; and an alarm sound identification section. an output unit 74 that outputs a control signal to an alarm device 72 provided in the vehicle interior in response to the identification signal from the alarm sound identification unit 10 and notifies the vehicle driver of the identification results of various alarm sounds, and the identification results of the alarm sound identification unit 10; is transmitted to the vehicle control device 76,
A transmitter 78 that executes vehicle control according to the identification result.
It is composed of and. Here, in the acoustic signal input section 6, the acoustic signal from the microphone 4 is first input to the preprocessing circuit 20, and the sound V (ff number is A/D:] that has passed through the preprocessing circuit 20).
A/D conversion is performed by the converter 22. The preprocessing circuit 20 is
This circuit processes the sound 94i so that the A/D converter 22 can perform A/D conversion well, and includes an amplifier that amplifies the sound 7 signal, an anti-aliasing filter,
It is equipped with a sample hold circuit, etc. Also, A/
The D converter 22 is controlled by the control circuit 24 and automatically converts the audio signal into A at a predetermined sampling period.
/D conversion and transfer the result to RAMa 26 or RAMb
28. The control circuit 24 first
The output of the D converter 22 is connected to R via the switch circuit 30.
Connect to AMa26 and send A/D conversion results to RAMa26.
Thereafter, when the storage area of the RAMa 26 becomes full, a storage signal indicating this is output to the CPU 40 of the alarm sound identification section 10, and at the same time, the switch circuit 30 is switched to switch the A/D converter 22. output to RAM
b28, and then A/D converter 22
/D conversion data is sequentially stored in RAMb28, and the A/D conversion results by the A/D converter 22 are stored in RAMa26. The data is stored alternately in RAMb28. 1. This allows the alarm sound identification unit 10 to read the A/D conversion data without stopping the operation of the A/D converter 22. :Become. Incidentally, the switch circuit 30 is actually composed of a TTL or CMO8 logic circuit. Also, RAMa26.
The RAMb 28 has a capacity capable of continuously storing the A/D conversion results for a period of time more than twice the period of the envelope of the alarm sound to be 858. Next, the alarm sound identification unit 10 includes a CPU 40, a ROM 42,
and RAM 44, and executes an alarm sound recognition process described below in accordance with a control program described later stored in ROM 42. Note that the alarm sound recognition process is performed by controlling the A/D converter 22 through the control circuit 24, and by controlling the R
The A/D conversion data stored in the AMa 26 or the RAMb 28 is output to the high-speed arithmetic processing unit 8 to perform predetermined arithmetic processing, and based on the arithmetic result, the external sound picked up by the microphone 4 is selected. , emergency car alarm sound,
Crosswalk warning sounds, warning sounds from barrier gates at railroad crossings,
This is a process for identifying alarm sounds (i.e. horn sounds) etc. from other vehicles. Also, in order to execute this process, R
The OM 42 stores in advance a reference frame trace amount extracted from the characteristic shape accompanying the temporal change of the peak of the frequency characteristic obtained by frequency analysis of the various alarm sounds to be identified. Next, the high-speed arithmetic processing section 8 processes the sound signal inputted through the acoustic signal input section 6 at high speed based on the input data from the alarm sound identification section 10, and calculates the frequency of the acoustic signal at each time. DSP (Digital Signal Processor) for processing large amounts of numerical calculations at high speed and for obtaining characteristics 50. A RAM 52 for storing input data from the W alarm identification unit 10 and data after calculation processing, and a ROM 54 in which a control program for high-speed calculation execution is stored in advance.
, and the DSP according to the calculation command from the alarm sound identification section 10.
It is constituted by a control circuit 56 that controls execution or stop of the operation of 50. Therefore, the CPU 40 of the alarm sound identification unit 10 stops the calculation process of the DSP 50 by the control circuit 56, transfers the data to be processed to the RAM 52, and then causes the control circuit 56 to execute the calculation process of the DSP 50. Thus, desired arithmetic processing can be executed as needed. Next, the output section 74 is for giving a signal to the alarm device 72 when a warning must be issued to the driver based on the result of the identification of the alarm sound by the alarm sound identification section 10.
The system notifies the driver of the presence or absence of a warning and its type using a speaker, warning lamp, display, etc. Further, the transmitter 78 is for transmitting the recognition result of the alarm sound to the vehicle control device 76 that controls the running of the vehicle, and the vehicle control device 76 side uses this information as a single-handed element of control. do. That is, for example, the vehicle control device 7
If the engine control device 6 is an engine control device U, the vehicle will be decelerated when an alarm sound from a barrier is recognized, or if the vehicle control device 76 is a steering control device U, the vehicle will be slowed down by an emergency vehicle in an automated driving vehicle. When a warning sound is recognized, the vehicle can be controlled to pull over to the shoulder of the road. The above is the hardware configuration of this device.The acoustoelectric transducer component is composed of the microphone 4, and the frequency analysis means M2 is the main component of the high-speed arithmetic processing section 8 and the high-speed arithmetic processing section 8.
The peak extracting means M3. Grouping means 144. Feature extraction means M5
．． The identification means M6 is constituted by the alarm sound identification section 10, and the reference frame trace storage means M7 is constituted by the ROM 42 of the alarm sound identification section 10. Next, FIG. 3 shows the processing procedure of the CPU 40 of the alarm sound recognition unit 10 to explain the operation of the vehicle alarm device 2. Figure 4 and Figure 5 illustrating data processing ~'fP, 12
This will be explained with reference to the explanatory diagram of the figure. As shown in FIG. 3, when the alarm sound recognition process is started, first, in step 100, an initialization process is executed to initialize the memory and various peripheral elements, and then the process moves to step 110. In step 110, a drive signal is output to the control circuit 24 to start the operation of the audio signal input section 6.
The A/D conversion operation of the D converter 22 is started. Then, as mentioned above, in the acoustic signal input section 6,
The sound tube signal input through the microphone 4 is converted into A/D converter 22 at a predetermined sampling period.
D conversion, and the A/D conversion data is first transferred to RAMa 2.
6 sequentially, and when RAMa 26 is full,
That is, when the data for the time necessary for recognizing the alarm sound is obtained, the control circuit 24 outputs a storage signal to the CPU 40. Then, the subsequent sampling data by the A/D converter 22 is sequentially stored in the RAMb28. With this procedure, the acoustic signal from the microphone 4 is transferred to the RAMa 26, RA which functions as a buffer memory.
Sequentially stored in Mb28, RAMa26 or RAMb
Each time 28 becomes full, a storage signal is input to CPU 40. Then, in synchronization with this storage signal,
The time-varying characteristics of the frequency characteristics within a certain period of time are determined. Therefore, in the subsequent step 120, the input of the storage signal from the control circuit 24 is waited, and when the storage signal is input and the RAMa 26 or RAMb 28 is full, the process moves to step 130, and the A/D The conversion data is read from RAMa 26 or RAMb 28 and temporarily stored in RAM 44 of alarm sound identification section 10. In the following step 140, frequency analysis processing is executed based on the stored A/D conversion data. This frequency analysis process is executed as shown in FIG. As shown in FIG. 4, in the frequency analysis process, first, in step 300, the A/D conversion data stored in the RAM 44 is transferred to the RAM 52 of the high-speed arithmetic processing section 8, and then in step 310, it is transferred via the control circuit 56. DSP
50 frequency analysis program is started. Then, the DSP 50 performs filter processing to calculate the time change characteristics of the amplitude (power) of a specific frequency component over a certain period of time from the A/D conversion data stored in the RAM 52 for a certain period of time according to the program procedure stored in the ROM 54. is executed, the processing result is stored in a free area of the RAM 52, and the CPU 40. Let me know. Therefore, in the subsequent step 320, we wait for the end signal of the program to be input, and when the end signal is input, the process moves to step 330, and the time change characteristic data for one frequency base of the filter processing result is stored in the RAM 52. is read out and stored in an empty area of the RAM 44. Then, in the following step 340, it is determined whether the filter processing has been completed for all preset extraction frequencies, and if it has not been completed, the process returns to step 310 and the filter processing program is started again. conduct. The DSP 50 changes the extraction frequency little by little each time the filter processing is activated and executes frequency analysis processing for the entire frequency range. As a result, time change characteristic data of each frequency component within the same time period is stored in the RAM 44. That is, for example, the acoustic signal shown in FIG.
After frequency analysis by P50, as shown in Figure 6,
Time-varying characteristics of frequency characteristics within a certain period of time can be obtained. This certain period of time is the time required to recognize the alarm sound, that is, about twice the cycle of the envelope of the alarm sound. In addition, DS
The time interval of the frequency analysis data output by P50 is equal to the sampling period, but the data shown in FIG. The data has been averaged. Next, the CPU 40 moves to step 150 in FIG. 3, extracts peak information from the data as shown in FIG.
Create data as shown in Figure 7. In other words, if the frequency characteristics at a certain time are the characteristics shown in FIG.
That is, the peak is set data of frequency and amplitude (fo, Po)
= (fo, Pz). Such processing is performed at each time 1. ,1. . t 5-4. The calculation is performed for each frequency characteristic, and as a result, peak data as shown in FIG. 7 is obtained. Next, the CPU 40 moves the execution to step 160, and the seventh
The continuity of the extracted peaks is determined as shown in the figure. This means that if the peak frequency f in the frequency characteristic at an arbitrary time t is close to the peak frequency fl-1 in the frequency characteristic at the previous time tl-1 by a certain width,
The extracted peak fl is grouped as an extension of continuous line B. Also, conversely, the peak frequency ft and the peak frequency f
If l-1 are not close to each other with a constant width, that peak is used as the starting point of a new continuous line for grouping. By performing such processing on the peaks of the frequency characteristics at each time, the extracted peaks are grouped into continuous lines B, continuous lines C, etc., as shown in FIG. Note that in an actual road environment, peaks that should originally be extracted may not be extracted due to noise, the passage of sound-insulating objects, and the like. In such cases, silence occurs between the peaks before and after the peak, and the sound is not detected as continuous. Therefore, the determination of continuity with respect to time is relaxed, and even if the peak frequency is continuous with the peak frequency a certain time ago, it is determined to be continuous. You may also do this. Next, in steps 170 and 180, the grouped peak groups are divided into blocks for each common feature amount of the characteristic shape that changes over time. Among them, step 170
Now, as shown in FIG. 8, attention is paid to the time change in the frequency of peak data grouped into continuous lines B, continuous lines C, etc., and the blocks are divided. That is, as shown in FIG. 9, from the perspective of the time change in frequency, the peak series of continuous line B can be divided into block X where the frequency is increasing and block Y where the frequency is decreasing. can. On the other hand, since the peak sequence of the continuous line C has a constant frequency throughout, it is determined as one block Z as it is. Next, in step 180, block division is further performed focusing on temporal changes in amplitude. Block X in FIG. 9 is determined to have constant amplitude, block Y to constant amplitude, and block Z to attenuated amplitude. In this case, no further block division is performed. In the example shown in FIG. 12, in the block division focusing on the temporal change in frequency in step 170, the frequency of the entire block is constant and is determined as one block, but in this step 180, the frequency change over time is determined as one block. Since we are focusing on changes, it is divided into two blocks, block V and block W, and both blocks are determined to be amplitude attenuation. Next, in step 190, the features of each block x, y, z divided as described above are stored in the RAM in the following format.
44. (Block start time, block end time. Temporal change form of amplitude, block start frequency. Time change form of block end frequency 1 frequency) Therefore, in the data of Fig. 10, specifically, X = ( T,, T,, constant, fo, f+, rise)
'1' = ('r+, T2. constant, f,, f,,
descending) Z = (T3. T4. descending, fs, fz,
constant). As described above, a special ftk focusing on the time change in the frequency and the time change in the amplitude of the sound input from the microphone 4 is performed.
This means that ffi is extracted for each block. The program then proceeds to step 200, where step 1
It is determined whether or not any of the features 1fiffi obtained in step 90 match the reference feature amount of the recognition target sound. Each recognition target sound is divided into reference blocks for each reference time@quantity, similar to the above-mentioned blocking. The reference block name constructed for each recognition target sound and the reference time 1fiffi of each reference block are stored in the ROM 42 in the same format as described above. For example, a warning sound from a railroad crossing barrier is divided into two reference blocks α and β. Then, the reference feature amount of each reference block is α=(O, t, ±Δ8. descending. f, ±Δ2.f, ±Δ8. constant) β=(0+Δ4.t, ±Δ3. descending. ±Δ6. ±Δ7.constant). In addition, in the actual railroad crossing warning sound, the reference frame trace amounts α and β are related by a certain relationship unique to the sounding body, so
The reference characteristic@quantity β is defined by the allowable deviation from the reference frame trace amount α. That is, the start time of the second sound corresponding to the reference weight β is within Δ4 after the end of the first sound with respect to the first sound corresponding to the reference frame trace amount α, and the frequency is within Δ4 relative to the frequency of the first sound. The allowable range is ±ΔG or ±Δ〒. Thus, the relative difference Δ between the frequencies of the first and second sounds. ,Δ. Also, since the shape of time change of frequency and amplitude, and the sounding time L+jf of the first and second sounds are limited, the allowable range Δ2 of the absolute value f of the frequency of the first and second sounds is limited. Δ. Even if it is set to a very large value or to infinity (that is, there is no limit to the absolute value of the frequency), other sounds will not be mistakenly recognized as a railroad crossing warning sound, and there will be no difference in frequency due to individual differences in the sounding body or Doppler. Recognition becomes possible without being affected by the frequency shift caused by the effect. Also, in the case of an ambulance siren, the ``pee'' and ``po'' of ``pee-po-pee-bo'' are used.
By limiting the relative relationship between the two reference blocks corresponding to , it becomes possible to obtain the same knowledge as in the case of railroad crossings. The feature amount of the detected sound extracted as described above is compared with the layered feature amount in the following manner. The corresponding reference frame trace amount is selected by determining whether or not the time variation form of the amplitude and the time variation form of the frequency match between the feature amount and the reference frame trace amount. Then, between the selected reference frame trace amount and the feature amount, we check whether the duration is within the allowable range, whether the frequency change is within the allowable range, the starting frequency and the ending frequency. A determination is made as to whether or not the value is within the permissible range. In this way, the base quantity that satisfies all the conditions is selected. Next, when one grouped peak group is composed of multiple blocks or when two blocks are close to each other,
It is determined whether the relationship between the blocks is equal to the relationship between the reference blocks, and when the relationship between the two is equal,
Ultimately, it is recognized as an alarm sound made up of reference blocks. Specifically, the continuous line segment B in Figure 9 is the sound of a fire engine siren whose frequency changes over time, and the continuous line segment in Figure 12 is the sound of two consecutive sounds with the frequency unchanged and the amplitude attenuated. , recognized as the warning sound of a railroad crossing barrier. Next, the process proceeds to step 210, and based on the recognition result in step 200, a recognition signal is output to the output unit 74 and transmission unit 78 in FIG. This allows the sound to be heard inside the vehicle. In this way, one cycle of sound recognition processing is completed, and the process returns to step 120, and the next cycle of sound v12 Ta processing is executed in synchronization with the next stored signal. [Effects of the Invention] The acoustic recognition processing of the present invention obtains the frequency characteristics of the acoustic signal at each time, extracts the peak from the frequency characteristics, and
The continuity of the peaks with respect to time is determined, the peaks are divided into consecutive peak groups, and the feature amount of the characteristic shape accompanying the temporal change of the peak group is extracted, and the feature amount (!:
A predetermined sound is identified based on the M quasi-feature quantities. Therefore, since sounds are recognized based on the shape of their frequency characteristics as they change over time, it is possible to accurately recognize sounds that have a frequency shift or whose frequency changes over time. becomes. In addition, the characteristic shapes are not compared as they are, but are compared based on the features of the shapes, which shortens the matching calculation and improves the recognition speed.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing the concept of the present invention. FIG. 2 is a block diagram showing the configuration of an apparatus according to a specific embodiment of the present invention. FIG. 3 and FIG. 4 are flowcharts showing the processing procedure of the CPU of the device of the same embodiment. FIG. 5 is a waveform diagram showing an acoustic signal. FIG. 6 is an explanatory diagram showing the frequency analysis results. FIG. 7 is an explanatory diagram showing peak extraction. FIG. 8 is an explanatory diagram showing grouping of peaks by continuous line segments. FIG. 9 is an explanatory diagram showing the division of peak groups into blocks for each feature. FIG. 10 is an explanatory diagram showing extraction of special @ quantities. FIG. 11 is a characteristic diagram showing frequency characteristics at one time. FIG. 12 is an explanatory diagram showing the blocking of peak groups. 2-Vehicle alarm device 4' Microphone 6 - Sound and thunder signal input section 8 High-speed calculation processing section 10 Alarm sound identification section Patent applicant Nippondenso Co., Ltd.

Claims (1)

[Scope of Claims] An acoustic recognition device that recognizes a predetermined sound from a detected sound detected by an acoustoelectric transducer, comprising: frequency analysis means for determining the frequency characteristics at each time of an acoustic signal output from the acoustoelectric transducer; peak extracting means for extracting peaks from the frequency characteristics at each time obtained by the frequency analyzing means; and determining the continuity of the peaks extracted by the peak extracting means with respect to time, and extracting the peaks for each group of consecutive peaks. a grouping means for grouping; a feature extracting means for extracting a feature amount of a characteristic shape accompanying a temporal change of the peak group grouped by the grouping means; corresponding to the feature amount of the detected sound, a reference feature amount storage means that stores the feature amount of a predetermined sound to be recognized as a reference feature amount, based on the feature amount extracted by the feature amount extraction means and the reference feature amount stored in the reference feature amount storage means. hand,
A sound recognition device comprising: identification means for identifying a predetermined sound.