Abnormal sound determination device, abnormal sound determination method, and abnormal sound determination program

Abstract

The present application relates to an abnormal sound determination device, an abnormal sound determination method, and an abnormal sound determination program. An abnormal sound is determined with good accuracy using artificial intelligence. An abnormal sound determination device has a computing device. The computing device performs the following steps: determining frequency-time data of a sound recorded by a vehicle; inputting the determined frequency-time data to the learned model, causing the learned model to determine a kind of abnormal sound contained in the sound, causing the learned model to determine a range of reference indicating a frequency range and a time range used to determine the kind of abnormal sound; specifying a specified range indicating a frequency range and a time range in the determined frequency-time data; and determining whether to cause the output device to output the kind of abnormal sound in a determination process including at least one of whether the range of reference and the specified range overlap as a determination element.

Description

TECHNICAL FIELD

[0001] The technology disclosed in this specification relates to an abnormal sound determination device, an abnormal sound determination method, and an abnormal sound determination program. BACKGROUND

[0002] In Japanese Patent Application Publication No. 2017-110919, a technology of determining an abnormal sound from a sound recorded by an image forming apparatus is disclosed. In this technology, the sound recorded by the image forming apparatus is transformed into frequency-time data representing a time variation of a frequency spectrum. Further, the frequency-time data is subjected to a fast Fourier transform in a time axis direction. A user determines a kind of the abnormal sound (for example, a generation source of the abnormal sound) based on an analysis result of the fast Fourier transform and a database of past generated abnormal sounds. SUMMARY

[0003] PROBLEMS TO BE SOLVED BY THE INVENTION

[0004] As described above, in Japanese Patent Application Publication No. 2017-110919, a user (that is, a person) determines a kind of an abnormal sound. In contrast to this, the present inventors conducted an experiment of causing an artificial intelligence to determine an abnormal sound from a sound recorded by a vehicle. In the experiment, there was a case where an abnormal sound that was not actually generated was determined by the artificial intelligence. In this specification, a technology of determining an abnormal sound with good precision using an artificial intelligence is presented.

[0005] SOLUTION TO PROBLEM

[0006] An abnormal sound determination device disclosed in this specification has an operation device capable of accessing a learned model of an artificial intelligence and an output device. The operation device executes the following steps: determining frequency-time data representing a time variation of a frequency spectrum of a sound recorded by a vehicle; inputting the determined frequency-time data to the learned model, causing the learned model to determine a kind of an abnormal sound contained in the sound based on the input frequency-time data, causing the learned model to determine a reference range in the input frequency-time data, the reference range representing a frequency range and a time range used to determine the kind of the abnormal sound; specifying a specified range representing a frequency range and a time range in the determined frequency-time data; and determining whether or not to cause the output device to output the kind of the abnormal sound in a determination process including at least one of the reference range and the specified range as a determination element.

[0007] Note that the above-described "determining frequency-time data representing a time variation of a frequency spectrum of a sound recorded by the vehicle" can be either a step of calculating the frequency-time data by the operation device based on the sound recorded by the vehicle or a step of inputting the frequency-time data calculated by an external device to the operation device.

[0008] In addition, the above-described "learned model" can be located at any position as long as it is accessible by the operation device. For example, the learned model can be stored in a storage device inside the abnormal sound determination device or in a storage device on a network accessible by the operation device.

[0009] In addition, the above-described "designated range" can be designated based on an input from a user or autonomously by the operation device based on a predetermined algorithm.

[0010] In the abnormal sound determination device, the operation device determines the frequency-time data, and inputs the determined frequency-time data to the learned model. Then, the learned model determines a kind of abnormal sound contained in the sound based on the frequency-time data. In this stage, the accuracy of the kind of abnormal sound determined by the learned model is not so high. That is, the learned model can determine a kind of abnormal sound that is not actually generated. In addition, the learned model determines a reference range representing a frequency range and a time range used to determine the kind of abnormal sound in the frequency-time data. In addition, the operation device determines the learned model, and designates a designated range representing a frequency range and a time range in the determined frequency-time data. The designation of the designated range is performed by an operation from a user, a predetermined algorithm, or the like. As the designated range, a frequency range and a time range corresponding to the abnormal sound can be designated. Like this, the designated range is designated separately from the reference range determined by the learned model. The operation device determines whether or not to cause the output device to output the kind of abnormal sound determined by the learned model in a determination process including at least one of the reference range and the designated range as a determination element. In a case where the reference range and the designated range overlap, the likelihood that the kind of abnormal sound determined by the learned model is correct is high, and in a case where the reference range and the designated range do not overlap, the likelihood that the kind of abnormal sound determined by the learned model is incorrect is high. Thus, by determining whether or not to cause the output device to output the kind of abnormal sound in the determination process including at least one of the reference range and the designated range as a determination element, it is possible to suppress a false kind of abnormal sound from being output by the output device. According to the abnormal sound determination device, it is possible to determine the kind of abnormal sound contained in the sound recorded by the vehicle with higher accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1is a block diagram of the abnormal sound determination device 10.

[0012] Figure 2 is an explanatory diagram of the recording device 42.

[0013] Figure 3 is a flowchart showing the abnormal sound determination method.

[0014] Figure 4 is a diagram showing the STFT data 54.

[0015] Figure 5 is a diagram showing the reference range 60.

[0016] Figure 6 is a diagram showing the reference range 60 for determining the contour.

[0017] Figure 7 is a diagram showing the screen for selecting the specified range 64.

[0018] Figure 8 is a flowchart showing the determination processing based on the reference range 60 and the specified range 64.

[0019] Figure 9 is a diagram showing the positional relationship between the reference range 60 and the specified range 64.

[0020] Figure 10 is a diagram showing the positional relationship between the reference range 60 and the specified range 64.

[0021] Figure 11 is a diagram showing the positional relationship between the reference range 60 and the specified range 64.

[0022] Figure 12 is a diagram showing the positional relationship between the reference range 60 and the specified range 64.

[0023] Figure 13 is a diagram showing the display screen of the determination result of the abnormal sound.

[0024] Figure 14 is a flowchart showing the determination processing of the modified example.

[0025] Figure 15 is a flowchart showing the determination processing of the modified example. DETAILED DESCRIPTION

[0026] In the example of the abnormal sound determination device disclosed in this specification, the operation device can also be configured to cause the output device to output the type of the abnormal sound when the reference range and the specified range overlap.

[0027] In one example of the abnormal sound determination device disclosed in this specification, the arithmetic device may not output the type of the abnormal sound when the basis range and the specified range do not overlap.

[0028] The abnormal sound determination device disclosed in this specification may also include an input device. In this case, the arithmetic unit may also specify the specified range based on input from a user using the input device.

[0029] Based on this structure, the range of sounds that the user identifies as abnormal can be defined as a specified range. Therefore, by determining whether the specified range in which the user identifies abnormal sounds overlaps with the range determined by the learned model, it is possible to accurately determine whether the type of abnormal sound determined by the learned model is correct.

[0030] An abnormal sound determination device disclosed in this specification may also include a speaker. In this case, the computing device may also perform the following step: after the specified range is specified and before determining whether the reference range and the specified range overlap, causing the speaker to emit a sound within the specified range.

[0031] According to this structure, users can hear sounds within a specified range when a range is specified. Therefore, users can determine whether the specified range has been correctly specified, and if not, can re-specify the range.

[0032] In one example of the abnormal sound determination device disclosed in this specification, the computing device may also perform a step of calculating and determining the contour determination reference range of the reference range. In this case, the computing device may also determine, in the determination process, whether the contour determination reference range and the specified range overlap.

[0033] The outline of the reference range determined by the model after learning is sometimes blurry. In this case, the computing device calculates the outline of the reference range to determine the reference range, thereby enabling a clear determination of whether the outline of the reference range and the specified range overlap.

[0034] In one example of the abnormal sound determination device disclosed in this specification, the learned model may also be composed of a convolutional network.

[0035] In one example of the abnormal sound determination device disclosed in this specification, the learned model may also use gradient-weighted class activation mapping to determine the basis range.

[0036] [Example]

[0037] Figure 1The abnormal sound detection device 10 shown determines abnormal sounds based on sounds recorded from a vehicle. The abnormal sound detection device 10 is constructed as a computer. The abnormal sound detection device 10 includes a processing unit 12, a storage device 16, a monitor 18, a speaker 20, an input device 22, and a network interface 24. The processing unit 12 consists of a CPU (central processing unit), memory, etc. The processing unit 12 is connected to the storage device 16, monitor 18, speaker 20, input device 22, and network interface 24. The storage device 16 consists of a hard disk drive, solid-state drive, etc. The storage device 16 stores an abnormal sound detection program 50. The processing unit 12 executes the abnormal sound detection program 50. The input device 22 consists of a mouse, keyboard, etc. Signals are input to the processing unit 12 by the user operating the input device 22. The processing unit 12 controls the monitor 18 and the speaker 20. Furthermore, the processing unit 12 is connected to a network line 30 via the network interface 24. The network line 30 can be the Internet or an intranet. The storage device 32, etc., is connected to the network line 30. At least one of storage devices 16 and 32 stores the learned model 52. Regardless of which storage device 16 or 32 stores the learned model 52, the computing device 12 can access the learned model 52. The learned model 52 is artificial intelligence, constructed from a convolutional neural network (CNN). The learned model 52 is a model that has been learned in a way that determines abnormal sounds based on frequency-time data (data representing the time-varying frequency distribution of sound). In this embodiment, short-time Fourier transform data (hereinafter referred to as STFT) data is used as the frequency-time data. If STFT data is input to the learned model 52, the learned model 52 determines the type of abnormal sound contained in the sound based on the STFT data. Furthermore, the learned model 52 has a function such as gradient-weighted class activation mapping (Grad-CAM). After learning Model 52, when determining the type of abnormal sound based on STFT data, Grad-CAM is used to determine the frequency range and time range from the STFT data that are the basis for determining the type of abnormal sound.

[0038] Storage device 16 is capable of storing audio data 56. Audio data 56 is generated by... Figure 2The recording device 42 shown generates sound. The recording device 42 is mounted on the vehicle 40. The recording device 42 records sounds generated by the vehicle 40 during its operation. The recording device 42 can record both inside and outside the vehicle (e.g., in the engine compartment). The recording device 42 stores the recorded sound as sound data 56 in a portable storage device. Sound data 56 is data representing the waveform of sound vibrations (i.e., air vibrations). Sound data 56 includes road noise, engine noise, and operating sounds of components (e.g., alternator, water pump, turbine, VSV (vacuum switching valve), etc.). Additionally, if an abnormal sound is generated in the vehicle 40 for some reason, the abnormal sound is also included in the sound data 56. By connecting the portable storage device to the abnormal sound detection device 10, the sound data 56 can be input to the abnormal sound detection device 10. Alternatively, sound data 56 can also be input to the abnormal sound detection device 10 via network line 30. Figure 1 As shown, the sound data 56 input to the abnormal sound determination device 10 is stored in the storage device 16.

[0039] Next, the abnormal sound determination method executed by the abnormal sound determination device 10 will be described. The abnormal sound determination device 10 executes the abnormal sound determination procedure 50 to determine... Figure 3 The abnormal sound determination method is shown. When the user performs a predetermined operation using the input device 22, the arithmetic unit 12 starts the abnormal sound determination procedure 50 (i.e., Figure 3 Methods for determining abnormal sounds).

[0040] In step S2, the processing unit 12 instructs the user to select sound data 56 by displaying a window on the monitor 18, etc. The user can select any sound data 56 from the sound data 56 stored in the storage device 16 using the input device 22.

[0041] In step S4, the processing unit 12 performs a short-time Fourier transform on the audio data 56 selected in step S2. As a result, the processing unit 12 calculates... Figure 4 The STFT data 54 is shown. The vertical axis of the STFT data 54 represents the frequency of the sound, and the horizontal axis represents time. Additionally, the color of each pixel in the STFT data 54 represents the sound pressure level (dB). In other words, the STFT data 54 is data representing the time-varying frequency spectrum of the sound. The processing unit 12 calculates the STFT data 54 as image data.

[0042] In step S6, the computing device 12 inputs the STFT data 54 calculated in step S4 into the learned model 52. The learned model 52 then extracts features from the input STFT data 54 and determines the abnormal sounds contained in the STFT data 54 (more specifically, the abnormal sounds contained in the sounds represented by the STFT data 54) based on the extracted features. That is, the learned model 52 determines the type of abnormal sound while determining the abnormal sound. The learned model 52 determines abnormal sound A, abnormal sound B, etc., as types of abnormal sounds. In a more specific example, the learned model 52 determines abnormal sounds such as alternator abnormal sounds, water pump abnormal sounds, turbine abnormal sounds, and VSV abnormal sounds. That is, the type of abnormal sound indicates the source of the abnormal sound. Furthermore, while determining the type of abnormal sound, the learned model 52 uses Grad-CAM to determine the frequency range and time range, i.e., the basis range, which are the criteria for determining the type of abnormal sound. That is, after model 52 is learned, it assigns importance to each pixel in the STFT data 54, determining the type of abnormal sound by prioritizing pixels with higher importance that have a greater impact on the output result (i.e., the determined type of abnormal sound). The scope is the set of pixels with high importance in the STFT data 54. For example, for... Figure 4 The STFT data 54 shown is as follows: Figure 5 The reference range 60 is shown. Since the importance of each pixel differs, the outline of reference range 60 is blurred. It should be noted that in step S6, after model 52 has been learned, it sometimes identifies multiple types of abnormal sounds. In this case, multiple reference ranges 60 are also identified.

[0043] In step S8, the processing unit 12 binarizes the importance set for each pixel of the STFT data 54 based on a predetermined threshold. Thus, the processing unit 12... Figure 6 The outline of the reference range 60 is clearly defined as shown. After step S8, the arithmetic unit 12 uses the reference range with the defined outline as the reference range 60.

[0044] In step S10, as Figure 7 As shown, the arithmetic unit 12 displays the STFT data 54 on the monitor 18. The user can select a frequency range and a time range on the STFT data 54 displayed on the monitor 18 by operating the input device 22. Hereinafter, the range selected in step S10 will be referred to as the specified range 64. For example, as... Figure 7 As shown, a specified range 64 is selected. In step S10, the user can select a range within the STFT data 54 that is considered equivalent to an abnormal sound as the specified range 64. The user can select the specified range 64 based on their own experience while visually recognizing the STFT data 54. Additionally, as... Figure 7As shown, a replay button 90 and an OK button 92 are displayed next to the STFT data 54. If the replay button 90 is pressed while the specified range 64 is selected, the arithmetic unit 12 replays the sound within the specified range 64 through the speaker 20. Therefore, by hearing the sound within the specified range 64, the user can determine whether an abnormal sound is contained within the specified range 64. Thus, the range considered equivalent to an abnormal sound can be easily selected as the specified range 64. The user can press the OK button 92 while the specified range 64 is selected. If the OK button 92 is pressed, the arithmetic unit 12 determines the selected specified range 64 and proceeds to the next step. In this way, in step S10, the range that the user determines is equivalent to an abnormal sound is designated as the specified range 64. It should be noted that in step S10, multiple ranges can also be designated as the specified range 64.

[0045] In step S12, a determination process based on the basis range 60 determined in step S6 and the specified range 64 specified in step S10 is performed. Then, the type of abnormal sound is displayed on the monitor 18 according to the determination process. Figure 8 This describes the details of step S12. For example... Figure 8 As shown, in step S12, the arithmetic device 12 executes steps S20 to 28.

[0046] In step S20, the computing device 12 selects one abnormal sound from the abnormal sounds (i.e., the types of abnormal sounds) determined by the learned model 52 in step S6. If there is only one abnormal sound determined in step S6, then that abnormal sound is selected.

[0047] In step S22, the computing device 12 determines whether the basis range 60 corresponding to the selected abnormal sound overlaps with the specified range 64.

[0048] For example, such as Figure 9 As shown, when the selected abnormal sound range 60 overlaps with the specified range 64, the arithmetic device 12 determines it as YES in step S22.

[0049] In addition, such as Figure 10 As shown, if the selected abnormal sound range 60 does not overlap with the specified range 64, the arithmetic device 12 determines it as NO in step S22.

[0050] in addition, Figure 11This indicates the existence of multiple reference ranges 60a and 60b. In this case, if the selected reference range 60 for the abnormal sound is reference range 60a, then reference range 60a overlaps with the specified range 64, and therefore the calculation device 12 determines it as YES in step S22. Conversely, if the selected reference range 60 for the abnormal sound is reference range 60b, then since reference range 60b does not overlap with the specified range 64, the calculation device 12 determines it as NO in step S22.

[0051] in addition, Figure 12 This indicates the existence of multiple specified ranges 64a and 64b. In this case, if the selected abnormal sound reference range 60 overlaps with either of the specified ranges 64a or 64b, the processing unit 12 determines it as YES in step S22. Conversely, if the selected abnormal sound reference range 60 does not overlap with either of the specified ranges 64a or 64b, the processing unit 12 determines it as NO in step S22.

[0052] If the result in step S22 is YES, then the processing unit 12 executes step S24. In step S24, the processing unit 12 displays the type of abnormal sound selected in step S20 on the monitor 18. For example, if the selected abnormal sound is abnormal sound A, in step S24, if... Figure 13 As shown, "Abnormal sound A detected." is displayed on monitor 18. On the other hand, if the determination is NO in step S22, the arithmetic unit 12 executes step S26. In step S26, the arithmetic unit 12 does not display the type of abnormal sound selected in step S20 on monitor 18.

[0053] In step S28, the processing unit 12 determines whether the processing of all abnormal sounds determined by the learned model 52 is complete. If the result in step S28 is NO, step S20 is executed again. In this case, in step S20, abnormal sounds that were not processed are selected. Therefore, steps S20 to S28 are repeated until the processing of all abnormal sounds is complete. Therefore, if multiple abnormal sounds are determined by the learned model 52, only the abnormal sounds that were determined to be YES in step S22 are displayed on the monitor 18. In addition, if there are no abnormal sounds that were determined to be YES in step S22, "No abnormal sound detected." is displayed on the monitor 18. If the processing of all abnormal sounds determined by the learned model 52 is complete, the processing unit 12 ends the abnormal sound determination process.

[0054] As explained above, when the selected abnormal sound's reference range 60 overlaps with the user-specified range 64, the processing unit 12 displays the type of the selected abnormal sound on the monitor 18; when the selected abnormal sound's reference range 60 does not overlap with the user-specified range 64, it does not display the type of the selected abnormal sound on the monitor 18. Therefore, abnormal sounds determined by the learned model 52 that match the abnormal sounds identified by the user are displayed on the monitor 18, while abnormal sounds determined by the learned model 52 that the user has not identified are not displayed on the monitor 18. Thus, the user can confirm the type of abnormal sound they have identified as an abnormal sound through the monitor 18.

[0055] The abnormal sound determination device 10 according to the embodiment is capable of selecting only the abnormal sounds identified as abnormal by the user from the abnormal sounds determined by the learned model 52 and displaying the type of the abnormal sound on the monitor 18. Therefore, it is possible to determine the type of abnormal sound actually generated in the vehicle 40 with good accuracy.

[0056] Furthermore, the abnormal sound determination device 10 of this embodiment can replay the sound within the specified range 64 after the user selects the specified range 64. Therefore, the user can easily determine the range corresponding to the abnormal sound as the specified range 64.

[0057] Furthermore, for the abnormal sound characteristic device of the embodiment, since the outline of the reference range 60 determined by the learned model 52 is determined by the computing device 12, it is possible to accurately determine whether the reference range 60 overlaps with the specified range 64.

[0058] It should be noted that, in Figure 8 In step S22, if it is determined that the range 60 overlaps with the specified range 64, the selected abnormal sound is displayed on the monitor 18 in step S24. However, as... Figure 14 As shown, there may also be a step S23 between steps S22 and S24 where other determinations are made. Figure 14 Even when the range 60 overlaps with the specified range 64, if the selected abnormal sound does not meet the reference condition 1 (i.e., it is NO in step S23), the selected abnormal sound will not be displayed on the monitor 18. Thus, even if it is determined to be YES in step S22, there may still be a situation where the selected abnormal sound is not displayed on the monitor 18.

[0059] In addition, Figure 8 In step S22, if it is determined that the range 60 and the specified range 64 do not overlap, the selected abnormal sound will not be displayed on the monitor 18 in step S26. However, as...Figure 15 As shown, there may also be a step S25 between steps S22 and S26 where other determinations are made. Figure 15 Even when the range 60 and the specified range 64 do not overlap, the selected abnormal sound will still be displayed on the monitor 18 if the selected abnormal sound meets the reference condition 2 (i.e., YES in step S25). In this way, even if NO is determined in step S22, the selected abnormal sound can still be displayed on the monitor 18. Alternatively, combinations are also possible. Figure 14 and Figure 15 .

[0060] Furthermore, in the above embodiment, the computing device 12 causes the monitor 18 to display the type of abnormal sound. However, the computing device 12 can also cause other devices to output the type of abnormal sound. For example, other devices can also output the type of abnormal sound via the network line 30.

[0061] Furthermore, in the above embodiment, the STFT data 54 is image data. That is, after the learning model 52 is completed, it determines the type and scope of the abnormal sound based on the STFT data 54, which is image data. However, the data format of the STFT data 54 is not limited to image data. For example, the STFT data 54 can also be data representing sound pressure level in terms of time and frequency.

[0062] Furthermore, in the above embodiment, the specified range 64 is specified by the user's operation. However, the arithmetic device 12 may also specify the specified range 64 according to a predetermined algorithm. For example, the range with high sound pressure level may be automatically specified as the specified range 64. In such a structure, the type of abnormal sound in the final output can be selected using the range 60 and the specified range 64 determined by different algorithms, thus enabling the determination of the type of abnormal sound with good accuracy.

[0063] The embodiments have been described in detail above, but these are merely illustrative and do not limit the scope of the claims. The technology described in the claims includes techniques obtained by various modifications and alterations to the specific examples described above. The technical elements described in this specification or drawings are useful individually or in various combinations, and are not limited to the combinations described in the claims at the time of application. Furthermore, the technology illustrated in this specification or drawings may achieve multiple objectives simultaneously, and achieving even one of these objectives is itself technically useful.

Claims

1. An abnormal sound detection device, wherein, The abnormal sound detection device has the following features: A computing device capable of accessing a learned model of artificial intelligence; and Output device The computing device performs the following steps: Determine frequency-time data representing the temporal variation of the spectrum of sound recorded from the vehicle; The determined frequency-time data is input into the learned model, so that the learned model determines the types of abnormal sounds contained in the sound based on the input frequency-time data, and the learned model determines a basis range in the input frequency-time data, which represents the frequency range and time range used to determine the types of abnormal sounds. In the absence of the stated range, the user's operation or a predetermined algorithm specifies the range of frequencies and times that are judged to be equivalent to abnormal sounds in the determined frequency-time data. as well as In a determination process that includes at least one of the determination elements, namely whether the basis range and the specified range overlap, it is determined whether the output device should output the type of the abnormal sound.

2. The abnormal sound detection device according to claim 1, wherein, When the calculation device causes the output device to output the type of the abnormal sound when the reference range and the specified range overlap.

3. The abnormal sound detection device according to claim 1 or 2, wherein, When the calculation device does not overlap the reference range and the specified range, it will not cause the output device to output the type of the abnormal sound.

4. The abnormal sound detection device according to claim 1 or 2, wherein, It also has an input device. The computing device specifies the specified range based on input from a user who has used the input device.

5. The abnormal sound detection device according to claim 1 or 2, wherein, It also has a speaker. The computing device further performs the following steps: after the specified range is specified and before determining whether the reference range and the specified range overlap, the speaker emits sound within the specified range.

6. The abnormal sound detection device according to claim 1 or 2, wherein, The computing device also performs a step of calculating and determining the contour of the defined range. In the determination process, the computing device determines whether the contour determination basis range and the specified range overlap.

7. The abnormal sound detection device according to claim 1 or 2, wherein, The learned model consists of a convolutional network.

8. The abnormal sound detection device according to claim 7, wherein, The learned model uses gradient-weighted class activation mapping to determine the range of criteria.

9. A method for determining abnormal sounds, comprising using a computing device and an output device capable of accessing a learned artificial intelligence model to determine abnormal sounds based on sounds recorded from a vehicle, wherein, The method for identifying abnormal sounds includes the following steps: The computing device determines frequency-time data representing the time-varying spectrum of the sound; The computing device inputs the determined frequency-time data into the learned model, so that the learned model determines the types of abnormal sounds contained in the sound based on the input frequency-time data, and the learned model determines a reference range in the input frequency-time data, which represents the frequency range and time range used to determine the types of abnormal sounds. The computing device, without showing the aforementioned range, uses user operation or a predetermined algorithm to specify a range in the determined frequency-time data that represents the frequency and time range deemed equivalent to an abnormal sound; and In a determination process that includes at least one of the determination elements, namely whether the basis range and the specified range overlap, the computing device determines whether to cause the output device to output the type of the abnormal sound.

10. A storage medium storing an abnormal sound detection program, wherein, The abnormal sound detection procedure enables a computing device that has access to the learned model of artificial intelligence and is connected to the output device to perform the following steps: Determine frequency-time data representing the temporal variation of the spectrum of sound recorded from the vehicle; The determined frequency-time data is input into the learned model, so that the learned model determines the types of abnormal sounds contained in the sound based on the input frequency-time data, and the learned model determines a basis range in the input frequency-time data, which represents the frequency range and time range used to determine the types of abnormal sounds. In the absence of the stated range, the user's operation or a predetermined algorithm specifies a range in the determined frequency-time data that represents the frequency and time range that are judged to be equivalent to abnormal sounds. In a determination process that includes at least one of the determination elements, namely whether the basis range and the specified range overlap, it is determined whether the output device should output the type of the abnormal sound.

Images