ETRI-Knowledge Sharing Plaform

KOREAN
논문 검색
Type SCI
Year ~ Keyword

Detail

Conference Paper SubSpectral Normalization를 적용한 음향 이벤트/장면 인식 알고리즘 성능 분석
Cited - time in scopus Share share facebook twitter linkedin kakaostory
Authors
정영호, 박수영, 이태진
Issue Date
2022-06
Citation
대한전자공학회 학술 대회 (하계) 2022, pp.1346-1349
Publisher
대한전자공학회
Language
Korean
Type
Conference Paper
Abstract
In this paper, we propose a novel acoustic event/scene recognition algorithm applied with SubSpectral Normalization that can maintain the importance of the unique characteristics of each frequency sub-band. Through the optimization of the existing algorithm, the model complexity is improved by 61% for Advanced GCRNN and 73.9% for Advanced Trident-ResCNN compared to the existing algorithm. Despite the improvement in model complexity, Advanced GCRNN improves accuracy by 0.8% and Advanced Trident-ResCNN improves F1-score by 1.9% compared to the existing algorithm in evaluation tests for 10 types of acoustic event/scene data.
KSP Keywords
Acoustic event, F1-score, Scene Recognition, model complexity, recognition algorithm, sub-band, unique characteristics