METHOD AND APPARATUS OF SOUND EVENT DETECTING ROBUST FOR FREQUENCY CHANGE

Disclosed is a method for detecting a sound event robust to a frequency change. The method comprises the steps of: receiving an audio signal; converting the audio signal into a two-dimensional signal; learning the two-dimensional signal with a convolutional neural network (CNN) to extract a feature...

Full description

Saved in:
Bibliographic Details
Main Author LIM WOO TAEK
Format Patent
LanguageEnglish
Korean
Published 25.09.2019
Subjects
ACOUSTICS
MUSICAL INSTRUMENTS
PHYSICS
SPEECH ANALYSIS OR SYNTHESIS
SPEECH OR AUDIO CODING OR DECODING
SPEECH OR VOICE PROCESSING
SPEECH RECOGNITION
Online AccessGet full text

Cover

More Information
Summary:Disclosed is a method for detecting a sound event robust to a frequency change. The method comprises the steps of: receiving an audio signal; converting the audio signal into a two-dimensional signal; learning the two-dimensional signal with a convolutional neural network (CNN) to extract a feature map; pooling the feature map based on a frequency; and determining whether a sound event for each of at least one time interval occurs according to a pooling result. 오디오 신호를 수신하는 단계, 상기 오디오 신호를 2차원 신호로 변환하는 단계, 컨볼루션 뉴럴 네트워크(Convolutional Neural Network; CNN)로 상기 2차원 신호를 학습하여 특징 맵(feature map)을 추출하는 단계, 상기 특징 맵을 주파수(frequency)를 기준으로 풀링(pooling)하는 단계 및 상기 풀링 결과에 따라 적어도 하나의 타임 인터벌(time interval) 각각에 대한 음향 이벤트 발생 여부를 판단하는 단계를 포함하는 음향 이벤트 검출 방법이 개시된다.
Bibliography:Application Number: KR20180030329