Method and apparatus for sound event detection robust to frequency change

Patent metadata

Field	Value
Publication number	US-10540988-B2
Application number	US-201816196356-A
Country	US
Kind code	B2
Filing date	Nov 20, 2018
Priority date	Mar 15, 2018
Publication date	Jan 21, 2020
Grant date	Jan 21, 2020

Abstract

Disclosed is a sound event detecting method including receiving an audio signal, transforming the audio signal into a two-dimensional (2D) signal, extracting a feature map by training a convolutional neural network (CNN) using the 2D signal, pooling the feature map based on a frequency, and determining whether a sound event occurs with respect to each of at least one time interval based on a result of the pooling.

First claim

What is claimed is:

1. A sound event detecting method performed by a sound event detecting apparatus, the sound event detecting method comprising:
receiving an audio signal;
transforming the audio signal into a two-dimensional (2D) time-frequency signal;
extracting a feature map from the 2D signal using a trained convolutional neural network (CNN);
pooling the feature map based on a frequency; and
determining whether a sound event occurs with respect to each of one or more time intervals based on a result of the pooling.

2. The sound event detecting method of claim 1 , wherein the determining comprises:
calculating a probability value of a sound event occurring with respect to each of the one or more time intervals based on the result of the pooling; and
determining whether a sound event occurs with respect to each of the one or more time intervals based on the probability value.

3. The sound event detecting method of claim 2 , wherein the determining of whether a sound event occurs with respect to each of the one or more time intervals based on the probability value comprises determining that a sound event occurs at a time interval if a probability value corresponding to the time interval is greater than or equal to a predetermined value.

4. The sound event detecting method of claim 1 , further comprising:
classifying a sound event occurring at each time interval based on predefined sound event information.

5. The sound event detecting method of claim 1 , wherein the audio signal is transformed into the 2D signal using one of fast Fourier transform (FFT), constant Q transform (CQT), and Wavelet.

6. A non-transitory computer-readable medium storing instructions that when executed by one or more processors, cause the one or more processors to perform the method of claim 1 .

7. A sound event detecting apparatus, comprising:
a memory configured to store a control program;
one or more processors configured to operate based on the control program; and
a receiver configured to receive an audio signal from an outside,
wherein the control program is configured to perform:
receiving an audio signal from an outside,
transforming the audio signal into a two-dimensional (2D) time-frequency signal,
extracting a feature map from the 2D signal using a trained neural network (CNN),
pooling the feature map based on a frequency, and
determining whether a sound event occurs with respect to each of one or more time intervals based on a result of the pooling.

8. The sound event detecting apparatus of claim 7 , wherein the determining comprises:
calculating a probability value of a sound event occurring with respect to each of the one or more time intervals based on the result of the pooling; and
determining whether a sound event occurs with respect to each of the one or more time intervals based on the probability value.

9. The sound event detecting apparatus of claim 8 , wherein the determining of whether a sound event occurs with respect to each of the one or more time intervals based on the probability value comprises determining that a sound event occurs at a time interval if a probability value corresponding to the time interval is greater than or equal to a predetermined value.

10. The sound event detecting apparatus of claim 7 , wherein the control program is further configured to perform classifying a sound event occurring at each time, interval based on predefined sound event information.

11. The sound event detecting apparatus of claim 7 , wherein the audio signal is transformed into the 2D signal using one of fast Fourier transform (FFT), constant transform (CQT), and Wavelet.

Assignees

• Electronics & Telecommunications Res Inst

Inventors

• Lim Woo-Taek

Classifications

• G06N3/08

  Learning methods · CPC title

• G10L25/51

  for comparison or discrimination · CPC title

• G10L25/18

  the extracted parameters being spectral information of each sub-band · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title