# US 12,015,421 B2 · Training and learning model for recognizing acoustic signal

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,015,421 B2
- Document ID: US12015421B2
- Status: Granted
- Kind: B2
- Application number: 17/484,284
- Filed: 2021-09-24
- Published: 2022-07-07
- Granted: 2024-06-18
- Earliest claimed: 2021-01-05
- Assignees: Electronics and Telecommunications Research Institute
- Applicants: Electronics and Telecommunications Research Institute
- Inventors: Young Ho Jeong; Soo Young Park; Tae Jin Lee
- Web page: https://patentgrep.com/patent/US12015421B2

## CPC

- G06N 3/08: Learning methods
- H03M 1/60: with intermediate conversion to frequency of pulses
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/09: Supervised learning
- G06N 20/20: Ensemble learning
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- H03M: CODING; DECODING; CODE CONVERSION IN GENERAL (using fluidic means F15C4/00; optical analogue/digital converters G02F7/00; coding, decoding or code conversion, specially adapted for particular applications, see the relevant subclasses, e.g. G01D, G01R, G06F, G06T, G09G, G10L, G11B, G11C, H04B, H04L, H04M, H04N; ciphering or deciphering for cryptography or other purposes involving the need for secrecy G09C)

## Abstract

Disclosed are a training method for a learning model for recognizing an acoustic signal, a method of recognizing an acoustic signal using the learning model, and devices for performing the methods. The method of recognizing an acoustic signal using a learning model includes identifying an acoustic signal including an acoustic event or acoustic scene, determining an acoustic feature of the acoustic signal, dividing the determined acoustic feature for each of a plurality of frequency band intervals, and determining the acoustic event or acoustic scene included in the acoustic signal by inputting the divided acoustic features to a trained learning model.

## Claims

### Claim 1

A method of recognizing an acoustic signal using a learning model, the method comprising:
identifying an acoustic signal comprising an acoustic event or acoustic scene;
determining an acoustic feature of the acoustic signal;
dividing the determined acoustic feature for each of a plurality of frequency band intervals; and
determining the acoustic event or acoustic scene included in the acoustic signal by inputting the divided acoustic features to a trained learning model,
wherein the dividing comprises dividing the acoustic feature into frequency band intervals of different lengths based on a frequency axis of the acoustic feature.

### Claim 2 (depends on 1)

The method of claim 1, wherein the determining of the acoustic feature comprises:
converting the acoustic signal into a frequency domain; and
generating log Mel-filterbank energy based on the converted acoustic signal.

### Claim 3 (depends on 1)

The method of claim 1, wherein the trained learning model comprises a combinational model and a plurality of neural network models, wherein each of the neural network models is trained to determine an acoustic event or acoustic scene included in the divided acoustic feature, and the combinational model is configured to combine output results of the plurality of neural network models.

### Claim 4

A method of training a learning model for recognizing an acoustic signal, the method comprising:
identifying an acoustic signal comprising an acoustic event or acoustic scene and a label representing the acoustic event or acoustic scene included in the acoustic signal;
determining an acoustic feature of the acoustic signal;
dividing the determined acoustic feature for each of a plurality of frequency band intervals; and
training the learning model using the acoustic feature divided by each of the plurality of frequency band intervals and the label,
wherein the dividing comprises dividing the acoustic feature into frequency band intervals of different lengths based on a frequency axis of the acoustic feature.

### Claim 5 (depends on 4)

The method of claim 4, wherein the determining of the acoustic feature comprises:
converting the acoustic signal into a frequency domain; and
generating log Mel-filterbank energy based on the converted acoustic signal.

### Claim 6 (depends on 4)

The method of claim 4, wherein the trained learning model comprises a combinational model and a plurality of neural network models, wherein each of the neural network models is trained to determine an acoustic event or acoustic scene included in the divided acoustic feature, and the combinational model is configured to combine output results of the plurality of neural network models.

### Claim 7 (depends on 4)

The method of claim 4, wherein the training of the learning model comprises:
obtaining an output signal in which the acoustic event or acoustic scene is determined from the learning model by inputting the divided acoustic signal to the learning model; and
updating weights of the learning model according to a difference between the output signal and the label.

### Claim 8

A recognition device for performing a method of recognizing an acoustic signal, the recognition device comprising:
a processor,
wherein the processor is configured to:
identify an acoustic signal comprising an acoustic event or acoustic scene,
determine an acoustic feature of the acoustic signal,
divide the determined acoustic feature for each of a plurality of frequency band intervals, and
determine the acoustic event or acoustic scene included in the acoustic signal by inputting the divided acoustic features to a trained learning model,
wherein the processor is further configured to divide the acoustic feature into frequency band intervals of different lengths based on a frequency axis of the acoustic feature.

### Claim 9 (depends on 8)

The recognition device of claim 8, wherein the processor is further configured to convert the acoustic signal into a frequency domain, and generate log Mel-filterbank energy based on the converted acoustic signal.

### Claim 10 (depends on 8)

The recognition device of claim 8, wherein the trained learning model comprises a combinational model and a plurality of neural network models, wherein each of the neural network models is trained to determine an acoustic event or acoustic scene included in the divided acoustic feature, and the combinational model is configured to combine output results of the plurality of neural network models.