Scene estimation method, scene estimation device, and program

Abstract

Provided is a technology that, even when the number of input signals is increased, allows a scene to be inferred with high accuracy. When S represents the number of scenes and M represents the number of input acoustic signals, the present invention includes: an acoustic signal encoding step for generating an integrated acoustic feature quantity from an m-th input acoustic signal (m=1,… , M) and a position at which the m-th input acoustic signal has been acquired by this scene inference device (hereinafter, referred to as m-th input acoustic signal acquisition position) (m=1,…, M); and a scene selection step for selecting a scene in which M input acoustic signals have been acquired, from among the S scenes by the scene inference device using the integrated acoustic feature quantity.

Description

[Technical field]

[0001] The present invention relates to a technique for estimating a scene from which an audio signal or a video signal is acquired. [Background technology]

[0002] As described in Non-Patent Documents 1 and 2, there are conventional techniques for estimating the scene in which an audio signal or a video signal was acquired by using the signal. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] K. Imoto et al., “Spatial Cepstrum as a Spatial Feature Using a Distributed Microphone Array for Acoustic Scene Analysis,” IEEE / ACM Transactions on Audio, Speech, and Language Processing, Vol.25, No.6, JUNE 2017. [Non-Patent Document 2] D. Zhukov et al., “Cross-Task Weakly Supervised Learning from Instructional Videos,” IEEE / CVF Conference on Computer Vision and Pattern Recognition(CVPR) 2019,JUNE 2019. Summary of the Invention [Problem to be solved by the invention]

[0004] In general, the more audio and video signals are used for scene estimation, the more information is available for scene estimation due to the reduction in blind spots, and the more accurate the scene estimation becomes. However, the data handled in the scene estimation process becomes highly dimensional. As a result, the so-called curse of dimensionality occurs, and the problem arises that even if the number of audio and video signals is increased, the accuracy does not increase as much as expected.

[0005] Therefore, an object of the present invention is to provide a technique for estimating a scene with high accuracy even when the number of input signals is large. [Means for solving the problem]

[0006] One aspect of the present invention includes an acoustic signal encoding step in which, where S is the number of scenes and M is the number of input acoustic signals, the scene estimation device generates an integrated acoustic feature from the m-th input acoustic signal (m=1, ..., M) and a position (hereinafter referred to as the m-th input acoustic signal acquisition position) (m=1, ..., M) at which the m-th input acoustic signal was acquired, and a scene selection step in which the scene estimation device uses the integrated acoustic feature to select, from the S scenes, a scene from which M input acoustic signals have been acquired.

[0007] One aspect of the present invention includes an audio signal encoding step in which a scene estimation device generates an integrated audio feature from an m-th input audio signal (m=1, ..., M) and a position where the m-th input audio signal was acquired (hereinafter referred to as the m-th input audio signal acquisition position) (m=1, ..., M), where S is the number of scenes, M is the number of input audio signals, and N is the number of input video signals; a video signal encoding step in which the scene estimation device generates an integrated video feature from an n-th input video signal (n=1, ..., N) and a position where the n-th input video signal was acquired (hereinafter referred to as the n-th input video signal acquisition position) (n=1, ..., N); and a scene selection step in which the scene estimation device uses the integrated audio feature and the integrated video feature to select a scene from S scenes for which M input audio signals and N input video signals have been acquired. Effect of the Invention

[0008] According to the present invention, even if the number of input signals is large, it is possible to estimate a scene with high accuracy. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a block diagram showing an example of a configuration of a scene estimation device 100. [Diagram 2] 11 is a flowchart showing an example of the operation of the scene estimation device 100. [Diagram 3] FIG. 2 is a block diagram showing an example of a configuration of a scene estimation device 200. [Figure 4] 13 is a flowchart showing an example of the operation of the scene estimation device 200. [Diagram 5] FIG. 2 is a block diagram showing an example of a configuration of a scene estimation device 300. [Figure 6] 13 is a flowchart showing an example of the operation of the scene estimation device 300. [Figure 7] FIG. 2 is a diagram illustrating an example of a functional configuration of a computer that realizes each device according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, an embodiment of the present invention will be described in detail. Note that components having the same functions are given the same reference numbers and duplicated explanations will be omitted.

[0011] Before describing each embodiment, the notation used in this specification will be described.

[0012] ^ (caret) represents a superscript. For example, x y^z Yes z is a superscript to x, and x y^z Yes z is a subscript to x. Also, _ (underscore) represents a subscript. For example, x y_z Yes z is a superscript to x, and x y_z Yes zis a subscript on x.

[0013] The superscripts "^" and "~" such as ^x and ~x for a certain letter x should actually be written directly above the "x", but due to the constraints of the description in the specification, they are written as ^x and ~x.

[0014] First, the main point of the present invention will be explained. As described above, as the number of dimensions of the data handled increases, the curse of dimensionality will have an impact. Therefore, we consider removing features extracted from audio and video signals that are not necessary for scene estimation.

[0015] Even with an encoder that is a feature extraction means trained according to the space to be subjected to scene estimation, it is difficult to extract only the minimum necessary features for all audio signals and all video signals. This is because the information that can be acquired differs depending on the position at which the microphone or camera used to acquire the audio signal or video signal is installed, and for example, even if a microphone is installed at a certain position, a feature that takes into account the features contained in the information acquired only by microphones installed at other positions is acquired. By removing such redundant information, the feature is reduced in dimension. In the embodiment of the present invention, as a method for removing redundant information caused by the difference in the installation positions described above, a method of employing an encoder for removing redundant information in the rear stage of an encoder that does not consider the positions at which the microphone or camera is installed will be described. First Embodiment The scene estimation device 100 receives as input M (where M is an integer of 1 or more) sets of input audio signals and positions where the input audio signals are acquired, and N (where N is an integer of 1 or more) sets of input video signals and positions where the input video signals are acquired, and selects and outputs the scenes where the input audio signals and input video signals are acquired from S (where S is an integer of 1 or more) scenes. Here, a scene refers to a scene in which a single event (event) is consecutively connected. For example, a scene in which "a person arrives at the office" can be understood as a scene in which four events, "opening the office door," "saying hello," "walking to one's desk," and "sitting down," are consecutively connected.

[0016] A microphone can be used to obtain the input audio signal, and a camera can be used to obtain the input video signal.

[0017] It is assumed that the input audio signals and the input video signals are synchronized with each other, and that the input audio signals and the input video signals have the same length, which is called the clip length.

[0018] Hereinafter, the scene estimation device 100 will be described with reference to FIG. 1 and FIG. 2. FIG. 1 is a block diagram showing the configuration of the scene estimation device 100. FIG. 2 is a flowchart showing the operation of the scene estimation device 100. As shown in FIG. 1, the scene estimation device 100 includes M acoustic encoders 110 (hereinafter referred to as the first acoustic encoder 110, ..., M-th acoustic encoder 110), M conditional acoustic encoders 120 (hereinafter referred to as the first conditional acoustic encoder 120, ..., M-th conditional acoustic encoder 120), an integrated acoustic encoder 130, N video encoders 140 (hereinafter referred to as the first video encoder 140, ..., N-th video encoder 140), N conditional video encoders 150 (hereinafter referred to as the first conditional video encoder 150, ..., N-th conditional video encoder 150), an integrated video encoder 160, a scene selection unit 170, and a recording unit 190. The recording unit 190 is a component that appropriately records information required for processing by the scene estimation device 100.

[0019] A component including the first acoustic encoder 110, ..., Mth acoustic encoder 110, the first conditional acoustic encoder 120, ..., Mth conditional acoustic encoder 120, and the integrated audio encoder 130 is referred to as an acoustic signal encoder 105. Furthermore, a component including the first video encoder 140, ..., Nth video encoder 140, the first conditional video encoder 150, ..., Nth conditional video encoder 150, and the integrated video encoder 160 is referred to as a video signal encoder 135.

[0020] The operation of the scene estimation device 100 will be described with reference to Fig. 2. Hereinafter, each of various feature amounts generated in the course of the operation of the scene estimation device 100 is a vector of a predetermined dimension determined for each feature amount.

[0021] In S110, the mth audio encoder 110 receives the mth input audio signal as input, generates an mth audio feature from the mth input audio signal, and outputs the mth audio feature. Here, the dimension of the mth audio feature is smaller than the dimension of the mth input audio signal. For example, a multi-layer CNN (Convolutional Neural Networks) can be used as a neural network to configure the mth audio encoder 110. In this case, the mth audio encoder 110 converts the mth input audio signal into the logarithmic absolute value of a short-time Fourier transform (STFT) spectrogram, and inputs the logarithmic mel spectrogram obtained by applying a mel filter bank to the multi-layer CNN.

[0022] In S120, the m-th conditional acoustic encoder 120 receives the m-th acoustic feature generated in S110 and the position where the m-th input acoustic signal was acquired (hereinafter referred to as the m-th input acoustic signal acquisition position), generates an m-th conditional acoustic feature from the m-th acoustic feature and the m-th input acoustic signal acquisition position, and outputs the m-th conditional acoustic feature. Here, the dimension of the m-th conditional acoustic feature is smaller than that of the m-th acoustic feature. The m-th conditional acoustic encoder 120 may be configured using, for example, a neural network consisting of one linear layer. In this case, the m-th conditional acoustic encoder 120 inputs a vector combining the m-th acoustic feature and the m-th input acoustic signal acquisition position to the neural network.

[0023] In S130, the integrated acoustic encoder 130 receives the m-th conditional acoustic feature (m=1, ..., M) generated in S120 as input, generates an integrated acoustic feature from the m-th conditional acoustic feature (m=1, ..., M), and outputs it. The integrated acoustic encoder 130 can be configured using, for example, a neural network consisting of one linear layer. In this case, the integrated acoustic encoder 130 inputs a vector combining the m-th conditional acoustic features (m=1, ..., M) to the neural network.

[0024] In S140, the nth video encoder 140 receives the nth input video signal, generates an nth video feature from the nth input video signal, and outputs the nth video feature. Here, the dimension of the nth video feature is smaller than the dimension of the nth input video signal. For example, ResNet can be used as a neural network to configure the nth video encoder 140 (see Reference Non-Patent Document 1). (Reference Non-Patent Document 1: D. Tran et al., “A Closer Look at Spatiotemporal Convolutions for Action Recognition,” IEEE / CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2018, JUNE 2018.) The reason why ResNet is preferably used in the configuration of the n-th video encoder 140 will be explained. It is preferable that the n-th video encoder 140 can extract features as a moving image that considers the relationship between frames in addition to the image features of each frame of video. A configuration that satisfies this condition is ResNet, and an example of this is ResNet(2+1)D, a neural network that has achieved high accuracy in human behavior recognition.

[0025] In S150, the nth conditional video encoder 150 receives the nth video feature generated in S140 and the position where the nth input video signal was acquired (hereinafter referred to as the nth input video signal acquisition position), generates an nth conditional video feature from the nth video feature and the nth input audio / video signal acquisition position, and outputs the nth conditional video feature. Here, the dimension of the nth conditional video feature is smaller than that of the nth video feature. The nth conditional video encoder 150 may be configured using, for example, a neural network with one linear layer. In this case, the nth conditional video encoder 150 inputs a vector combining the nth video feature and the nth input video signal acquisition position to the neural network.

[0026] In S160, the integrated video encoder 160 receives the n-th conditional video feature (n=1, ..., N) generated in S150 as input, generates an integrated video feature from the n-th conditional video feature (n=1, ..., N), and outputs the integrated video feature. The integrated video encoder 160 may be configured using, for example, a neural network with one linear layer. In this case, the integrated video encoder 160 inputs a vector combining the n-th conditional video feature (n=1, ..., N) to the neural network.

[0027] In S170, the scene selection unit 170 receives the integrated audio feature generated in S130 and the integrated video feature generated in S160 as input, and selects a scene from among the S scenes from which M input audio signals and N input video signals have been acquired using the integrated audio feature and the integrated video feature, and outputs the selected scene. The scene selection unit 170 may be configured using, for example, a neural network consisting of one linear layer and a Softmax layer. In this case, the scene selection unit 170 inputs a vector combining the integrated audio feature and the integrated video feature to the neural network.

[0028] The operations of the audio signal encoder 105 and the video signal encoder 135 can be described as follows. The audio signal encoder 105 receives an m-th input audio signal (m=1, ..., M) and an m-th input audio signal acquisition position (m=1, ..., M), generates an integrated audio feature from the m-th input audio signal (m=1, ..., M) and the m-th input audio signal acquisition position (m=1, ..., M), and outputs the integrated audio feature. The video signal encoder 135 receives an n-th input video signal (n=1, ..., N) and an n-th input video signal acquisition position (n=1, ..., N), and generates an integrated video feature from the n-th input video signal (n=1, ..., N) and the n-th input video signal acquisition position (n=1, ..., N), and outputs the integrated video feature.

[0029] According to the embodiment of the present invention, it is possible to estimate a scene with high accuracy even if the number of input signals is large. Specifically, by using information on the position where the signal was acquired, it is possible to generate conditional features with smaller dimensions related to what should be paid special attention to within the signal acquisition position, and the use of the conditional features makes it possible to estimate a scene with high accuracy. <Second embodiment> In the first embodiment, both the audio signal and the video signal are used as input, but only the audio signal may be used. That is, the scene estimation device 200 receives as input a set of M (where M is an integer equal to or greater than 1) input audio signals and the positions at which the input audio signals are acquired, selects and outputs the scenes from which the input audio signals are acquired from among S (where S is an integer equal to or greater than 1) scenes.

[0030] The scene estimation device 200 will be described below with reference to Figs. 3 and 4. Fig. 3 is a block diagram showing the configuration of the scene estimation device 200. Fig. 4 is a flowchart showing the operation of the scene estimation device 200. As shown in Fig. 3, the scene estimation device 200 includes M acoustic encoders 110 (hereinafter referred to as the first acoustic encoder 110, ..., M-th acoustic encoder 110), M conditional acoustic encoders 120 (hereinafter referred to as the first conditional acoustic encoder 120, ..., M-th conditional acoustic encoder 120), an integrated acoustic encoder 130, a scene selection unit 270, and a recording unit 190. The recording unit 190 is a component that appropriately records information necessary for processing by the scene estimation device 200.

[0031] The operation of the scene estimation device 200 will be described with reference to Fig. 4. Since the processes from S110 to S130 are the same as those in the first embodiment, only the process of S270 will be described here.

[0032] In S270, the scene selection unit 270 receives the integrated acoustic feature generated in S130 as an input, and selects, from among the S scenes, a scene from which M input acoustic signals have been acquired, using the integrated acoustic feature, and outputs the selected scene. The scene selection unit 270 can be configured using, for example, a neural network consisting of one linear layer and a Softmax layer.

[0033] According to the embodiment of the present invention, it is possible to estimate a scene with high accuracy even if the number of input signals is large. Specifically, by using information on the position where the signal was acquired, it is possible to generate conditional features with smaller dimensions related to what should be paid special attention to within the signal acquisition position, and the use of the conditional features makes it possible to estimate a scene with high accuracy. <Third embodiment> In the first embodiment, both the audio signal and the video signal are used as input, but only the video signal may be used. That is, the scene estimation device 300 receives as input a set of N (where N is an integer equal to or greater than 1) input video signals and the positions at which the input video signals are acquired, and selects and outputs the scenes from which the input video signals are acquired from S (where S is an integer equal to or greater than 1) scenes.

[0034] Hereinafter, the scene estimation device 300 will be described with reference to Figs. 5 and 6. Fig. 5 is a block diagram showing a configuration of the scene estimation device 300. Fig. 6 is a flowchart showing an operation of the scene estimation device 300. As shown in Fig. 5, the scene estimation device 300 includes N video encoders 140 (hereinafter referred to as the first video encoder 140, ..., the Nth video encoder 140), N conditional video encoders 150 (hereinafter referred to as the first conditional video encoder 150, ..., the Nth conditional video encoder 150), an integrated video encoder 160, a scene selection unit 370, and a recording unit 190. The recording unit 190 is a component that appropriately records information required for processing by the scene estimation device 300.

[0035] The operation of the scene estimation device 300 will be described with reference to Fig. 6. Since the processes from S140 to S160 are the same as those in the first embodiment, only the process of S370 will be described here.

[0036] In S370, the scene selection unit 370 receives the integrated video feature generated in S160 as an input, and selects and outputs a scene from which N input video signals have been acquired from among the S scenes using the integrated video feature. The scene selection unit 370 may be configured using, for example, a neural network consisting of one linear layer and a Softmax layer.

[0037] According to the embodiment of the present invention, it is possible to estimate a scene with high accuracy even if the number of input signals is large. Specifically, by using information on the position where the signal was acquired, it is possible to generate conditional features with smaller dimensions related to what should be paid special attention to within the signal acquisition position, and the use of the conditional features makes it possible to estimate a scene with high accuracy. <Additional Notes> 7 is a diagram showing an example of a functional configuration of a computer 2000 that realizes each of the above-mentioned devices. The processing in each of the above-mentioned devices can be implemented by having a recording unit 2020 load a program for causing the computer 2000 to function as each of the above-mentioned devices, and having the control unit 2010, the input unit 2030, the output unit 2040, and the like operate.

[0038] The device of the present invention has, as a single hardware entity, an input section to which a keyboard or the like can be connected, an output section to which a liquid crystal display or the like can be connected, a communication section to which a communication device (e.g., a communication cable) capable of communicating with the outside of the hardware entity can be connected, a CPU (which may also have a central processing unit, cache memory, registers, etc.), memories such as RAM and ROM, an external storage device such as a hard disk, and buses connecting the input section, output section, communication section, CPU, RAM, ROM, and external storage device so that data can be exchanged between them. If necessary, the hardware entity may also be provided with a device (drive) capable of reading and writing a recording medium such as a CD-ROM. A physical entity equipped with such hardware resources is, for example, a general-purpose computer.

[0039] The external storage device of the hardware entity stores the programs required to realize the above-mentioned functions and the data required in the processing of these programs (not limited to the external storage device, but for example the programs may be stored in a ROM, which is a read-only storage device). Data obtained by the processing of these programs is stored appropriately in the RAM, the external storage device, etc.

[0040] In the hardware entity, each program stored in an external storage device (or ROM, etc.) and the data required for processing each program are loaded into memory as necessary, and interpreted, executed, and processed by the CPU as appropriate, so that the CPU realizes a predetermined function (each component represented as the above, "unit," "means," etc.).

[0041] The present invention is not limited to the above-described embodiment, and can be modified as appropriate without departing from the spirit of the present invention. In addition, the processes described in the above-described embodiment are not limited to being executed in chronological order according to the order described, but may be executed in parallel or individually depending on the processing capacity of the device that executes the processes or as necessary.

[0042] As described above, when the processing functions of the hardware entities (the devices of the present invention) described in the above embodiments are realized by a computer, the processing contents of the functions that the hardware entities should have are described by a program. Then, by executing this program on a computer, the processing functions of the hardware entities are realized on the computer.

[0043] The program describing the processing contents can be recorded on a computer-readable recording medium. The computer-readable recording medium may be, for example, a magnetic recording device, an optical disk, a magneto-optical recording medium, a semiconductor memory, or any other suitable recording medium. Specifically, for example, a hard disk drive, a flexible disk, a magnetic tape, or the like can be used as the magnetic recording device, a DVD (Digital Versatile Disc), a DVD-RAM (Random Access Memory), a CD-ROM (Compact Disc Read Only Memory), a CD-R (Recordable) / RW (ReWritable), or the like can be used as the optical disk, an MO (Magneto-Optical disc), or the like can be used as the magneto-optical recording medium, and an EEP-ROM (Electronically Erasable and Programmable-Read Only Memory) or the like can be used as the semiconductor memory.

[0044] The program may be distributed, for example, by selling, transferring, lending, etc. portable recording media such as DVDs and CD-ROMs on which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of a server computer and transferring the program from the server computer to other computers via a network.

[0045] A computer that executes such a program, for example, first stores the program recorded on a portable recording medium or the program transferred from a server computer in its own storage device. Then, when executing a process, the computer reads the program stored in its own storage device and executes a process according to the read program. As another execution form of the program, the computer may directly read the program from the portable recording medium and execute a process according to the program, or may execute a process according to the received program each time a program is transferred from the server computer to the computer. The server computer may not transfer the program to the computer, but may execute the above-mentioned process by a so-called ASP (Application Service Provider) type service that realizes the processing function only by issuing an execution instruction and obtaining the result. Note that the program in this embodiment includes information used for processing by an electronic computer that is equivalent to a program (data that is not a direct instruction to the computer but has a nature that specifies the processing of the computer, etc.).

[0046] In addition, in this embodiment, a hardware entity is configured by executing a specific program on a computer, but at least a part of the processing contents may be realized by hardware.

[0047] The foregoing description of the embodiments of the invention has been presented for purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Modifications and variations are possible in light of the above teachings. The embodiments have been chosen and depicted to provide a best illustration of the principles of the invention and to enable those skilled in the art to utilize the invention in various embodiments and with various modifications as may be suitable for the practical use contemplated. All such modifications and variations are within the scope of the invention as defined by the appended claims interpreted in accordance with the breadth to which they are fairly, legally, and equitably entitled.

Claims

1. Let S be the number of scenes and M be the number of input audio signals. an audio signal encoding step in which the scene estimation device generates an integrated audio feature from an m-th input audio signal (m=1, ..., M) and a position where the m-th input audio signal is acquired (hereinafter referred to as an m-th input audio signal acquisition position) (m=1, ..., M); a scene selection step of selecting, by the scene estimation device, from among S scenes, scenes from which M input acoustic signals have been acquired, using the integrated acoustic feature; A scene estimation method comprising: The acoustic signal encoding step includes: an m-th acoustic encoding step (m=1, …, M) for generating an m-th acoustic feature from an m-th input acoustic signal; an m-th conditional acoustic encoding step (m=1, ..., M) of generating an m-th conditional acoustic feature from the m-th acoustic feature and the m-th input acoustic signal acquisition position; An integrated acoustic encoding step of generating the integrated acoustic feature from an m-th conditional acoustic feature (m=1, ..., M) is included. Scene estimation methods.

2. 2. The scene estimation method according to claim 1, The dimension of the m-th conditional acoustic feature is smaller than the dimension of the m-th acoustic feature.

13. A scene estimation method comprising:

3. Let S be the number of scenes, M be the number of input audio signals, and N be the number of input video signals. an audio signal encoding step in which the scene estimation device generates an integrated audio feature from an m-th input audio signal (m=1, ..., M) and a position where the m-th input audio signal is acquired (hereinafter referred to as an m-th input audio signal acquisition position) (m=1, ..., M); a video signal encoding step of generating an integrated video feature from an n-th input video signal (n=1, ..., N) and a position (hereinafter referred to as an n-th input video signal acquisition position) (n=1, ..., N) where the scene estimation device acquired the n-th input video signal; a scene selection step of selecting, from among S scenes, a scene from which M input audio signals and N input video signals have been acquired, by the scene estimation device, using the integrated acoustic feature and the integrated video feature; A scene estimation method comprising: The acoustic signal encoding step includes: an m-th acoustic encoding step (m=1, …, M) for generating an m-th acoustic feature from an m-th input acoustic signal; an m-th conditional acoustic encoding step (m=1, ..., M) of generating an m-th conditional acoustic feature from the m-th acoustic feature and the m-th input acoustic signal acquisition position; an integrated acoustic encoding step of generating the integrated acoustic feature from an m-th conditional acoustic feature (m=1, ..., M); The video signal encoding step includes: an n-th video encoding step (n=1, ..., N) of generating an n-th video feature from an n-th input video signal; an n-th conditional video encoding step (n=1, ..., N) of generating an n-th conditional video feature from the n-th video feature and the n-th input video signal acquisition position; The method includes an integrated video encoding step of generating the integrated video feature from an n-th conditional video feature (n=1, ..., N). Scene estimation methods.

4. The scene estimation method according to claim 3, The dimension of the m-th conditional acoustic feature is smaller than the dimension of the m-th acoustic feature, The dimension of the n-th conditional video feature is smaller than the dimension of the n-th video feature.

13. A scene estimation method comprising:

5. Let S be the number of scenes and M be the number of input audio signals. an audio signal encoder that generates an integrated audio feature from an m-th input audio signal (m=1, ..., M) and a position where the m-th input audio signal was acquired (hereinafter referred to as the m-th input audio signal acquisition position) (m=1, ..., M); a scene selection unit that selects, from among S scenes, scenes from which M input acoustic signals have been acquired, using the integrated acoustic feature; A scene estimation device comprising: The audio signal encoder includes: an m-th acoustic encoder (m=1, …, M) that generates an m-th acoustic feature from an m-th input acoustic signal; an m-th conditional acoustic encoder (m=1, …, M) that generates an m-th conditional acoustic feature from the m-th acoustic feature and the m-th input acoustic signal acquisition position; An integrated acoustic encoder for generating the integrated acoustic feature from an m-th conditional acoustic feature (m=1, ..., M) Scene estimation device.

6. A program for causing a computer to execute the scene estimation method according to any one of claims 1 to 4.

Images