JP-2026075996-A - Substitute model proposal device, substitute model proposal method, and substitute model proposal system

Abstract

[Problem] To propose a substitute model for analyzing new detection targets. [Solution] The substitute model proposal device comprises: an acquisition unit that acquires an audio signal in which a sound containing the target sound to be detected is recorded; an analysis unit that generates multiple frames by dividing the audio signal into predetermined intervals and analyzes the distance between the deep features of the multiple trained models and the deep features of the frame sounds included in the multiple frames for each frame using multiple trained models that detect sounds different from the target sound; and a proposal unit that determines and outputs one or more substitute models from the multiple trained models to be used for detecting the target sound based on the analyzed distance. [Selection Diagram] Figure 8

Inventors

• 溝渕 翔平

Assignees

• パナソニックＩＰマネジメント株式会社

Dates

Publication Date

20260511

Application Date

20241023

Claims (5)

1. An acquisition unit that acquires an audio signal containing the target sound that is to be detected, An analysis unit generates multiple frames by dividing the sound signal into predetermined intervals, and uses multiple trained models to detect sounds different from the target sound, analyzing the distance between the deep features of the multiple trained models and the deep features of the frame sounds included in the multiple frames for each frame. The system includes a proposal unit that, based on the analyzed distance, determines and outputs one or more surrogate models from among the plurality of trained models to be used for detecting the target sound, A device for proposing a substitute model.
2. The aforementioned proposal section is, Based on the analyzed distance, a candidate surrogate model is selected from the multiple trained models to be used for detecting the target sound. Based on the deep features of the trained model corresponding to the candidate substitute model and the deep features of the frame sound, the system determines and outputs the substitute model from among the candidate substitute models. The alternative model proposed device according to claim 1.
3. The aforementioned proposal section is, The system generates and outputs a suggestion screen containing sound information detected by the pre-trained model corresponding to the aforementioned substitute model. The alternative model proposed device according to claim 2.
4. A method for proposing a substitute model performed by at least one processor, The audio signal containing the target sound to be detected is acquired. Multiple frames are generated by dividing the aforementioned sound signal into predetermined sections. Using multiple trained models that detect sounds different from the target sound, the distance between the deep features of the multiple trained models and the deep features of the frame sounds included in the multiple frames is analyzed for each frame. Based on the analyzed distance, one or more surrogate models from the plurality of trained models to be used for detecting the target sound are determined and output. A method for proposing a substitute model.
5. A device that proposes one or more surrogate models used for detecting target sounds, A substitute model proposal system comprising a database that can communicate with the aforementioned device, The aforementioned database is Multiple trained models that detect sounds different from the target sound are transmitted to the device. The aforementioned device is A sound signal is acquired in which a sound including the aforementioned target sound is recorded. Multiple frames are generated by dividing the aforementioned sound signal into predetermined sections. Using the aforementioned multiple trained models, the distance between the deep features of the multiple trained models and the deep features of the frame sounds included in the multiple frames is analyzed for each frame. Based on the analyzed distance, one or more surrogate models from the plurality of trained models to be used for detecting the target sound are determined and output. A system for proposing substitute models.

Description

This disclosure relates to a substitute model proposal device, a substitute model proposal method, and a substitute model proposal system. Patent Document 1 discloses a method for determining the classification of input data. This method assumes that there are n classifications A and m classifications X (where n is an integer greater than or equal to 2, m is an integer greater than or equal to 1, and m ≤ nC2), that classification X has relatively fewer training data points than classification A, and that classification X shares common characteristics with two or more different classifications A. The method includes a data input step for inputting data, an initial discrimination step for determining whether the input data belongs to an arbitrary classification A or a classification X having characteristics common to classification A, a final discrimination step for determining whether the input data belongs to an arbitrary classification A or classification X using a second pre-trained model different from the first pre-trained model, and a discrimination result output step for outputting the discrimination result of the discrimination step. The method is characterized in that the second pre-trained model reuses part or all of the training data of classification A having characteristics common to classification X other than arbitrary classification A as training data for classification X. Japanese Patent Publication No. 2021-163483 A diagram showing an example of the system configuration of the proposed substitute model system according to the embodiment.Block diagram showing an example of the internal configuration of an A/D converter and terminal device.Block diagram showing an example of the functional configuration of the signal processing unit and memory of a terminal device.A flowchart showing an example of the operation procedure of a terminal device in the embodiment.A diagram showing an example of audio signal processing.This figure shows how the deep features of the target sound are mapped to the feature space.A diagram showing an example of the process for proposing candidate substitute models.A diagram showing an example of a screen for suggesting a substitute model. The following describes in detail embodiments of the substitute model proposal apparatus, substitute model proposal method, and substitute model proposal system disclosed herein, with reference to the drawings as appropriate. However, unnecessarily detailed explanations may be omitted. For example, detailed explanations of already well-known matters and redundant explanations of substantially identical configurations may be omitted. This is to avoid unnecessarily redundancy in the following explanation and to facilitate understanding for those skilled in the art. The accompanying drawings and the following explanation are provided to enable those skilled in the art to fully understand this disclosure and are not intended to limit the subject matter described in the claims. Referring to Figure 1, an example of the system configuration of the substitute model proposal system 100 according to this first embodiment will be described. Figure 1 is a diagram showing an example of the system configuration of the substitute model proposal system 100 according to this first embodiment. Note that, in order to simplify the diagram, the input unit INP and the display DP are omitted from the illustration of the substitute model proposal system 100 shown in Figure 1. The substitute model proposal system 100 captures sounds, including the target sound emitted by the object to be detected, using a microphone MC. It then proposes a pre-trained model (hereinafter referred to as the "substitute model") from among multiple pre-trained models designed to detect sounds different from the target sound, i.e., to substitute for detecting the object to be detected. The substitute model proposal system 100 includes a microphone MC, an A/D converter CV, a terminal device P1, a cloud server S1, and a pre-trained model database DB. The proposed alternative system 100 shown in Figure 1 is merely an example and is not limited thereto. For example, the microphone MC or A/D converter CV may be integrated with the terminal device P1. Furthermore, the terminal device P1 and the cloud server S1 may be integrated. The microphone MC captures sound, including the target sound emitted by the object being detected, and converts the captured sound into an audio signal (analog signal). The microphone MC outputs the audio signal (analog signal) to the A/D converter CV. Note that the audio signal in this disclosure may include sounds other than the target sound. The A/D converter CV is connected to the microphone MC for data (signal) communication and acquires the audio signal (analog signal) output from the microphone MC. The A/D converter CV converts the acquired audio signal (analog signal) into an audio signal (digital signal) in a format that can be processed by the terminal