文献类型：期刊文献

中文题名：鸟声标注技术及其在被动声学监测中的应用

英文题名：Advances in bird sound annotation methods for passive acoustic monitoring

作者：郭倩茸[1];段淑斐[1];谢捷[2];董雪燕[3];肖治术[4]

第一作者：郭倩茸

机构：[1]太原理工大学电子信息工程学院,太原030600;[2]南京师范大学计算机与电子信息学院/人工智能学院,南京210023;[3]北京联合大学特殊教育学院,北京100075;[4]中国科学院动物研究所农业虫害鼠害综合治理研究国家重点实验室,北京100101

第一机构：太原理工大学电子信息工程学院,太原030600

年份：2024

卷号：32

期号：10

起止页码：72-93

中文期刊名：生物多样性

外文期刊名：Biodiversity Science

收录：;北大核心:【北大核心2023】；CSCD:【CSCD2023_2024】；

基金：国家自然科学基金(32371556,12004275);山西省自然科学基金(202403021211098);山西省回国留学人员科研教研资助项目(2024-060)。

语种：中文

中文关键词：鸟声数据集;人工标注;半自动标注;自动标注;鸟声识别;被动声学监测

外文关键词：bird sound dataset;manual annotation;semi-automatic annotation;automatic annotation;bird sound recognition;passive acoustic monitoring

摘要：鸟声标注用于标记声音中的鸟类信息,如种类、声音结构等,是鸟类被动声学监测及相关声学数据分析、物种自动识别分类的重要基础。本文以鸟声标注为重点,比较了人工标注、自动标注和半自动标注等常用方法的优势,点明了各自在数据质量、标注一致性和标注效率等方面面临的挑战,同时探讨了这些标注方法在被动声学监测中的应用进展,提出了自动标注模型优化、跨地区数据集建立和半自动标注系统完善等未来发展方向。尽管目前自动标注方法取得了显著进展,但鸟声标注仍面临冷启动问题,亟需更大规模的跨地区数据集和高效的质量检测半自动标注系统,以满足标注数量和质量的双重要求。本综述有助于帮助鸟声数据集创建者和标注者更好地理解现有标注技术及其潜在的发展趋势,为大规模鸟类声学监测数据的高效物种自动识别提供技术支撑。
Background&Aim:Bird sound annotation is essential for marking bird-related information in audio data,such as species identification and sound structure.It serves as a crucial foundation for passive acoustic monitoring,birds acoustic data analysis,as well as automatic species identification and classification.The purpose of this review is to help bird sound dataset creators and annotators better understand the existing labeling technologies and their potential development trends.It also provides technical support for improving the efficiency of automatic species identification in large-scale avian acoustic monitoring data.Summary:This paper compares the advantages of various common methods such as manual annotation,automatic annotation,and semi-automatic annotation.It highlights the challenges each method faces in terms of data quality,annotation consistency and annotation efficiency.The review also discusses recent applications of these methods in passive acoustic monitoring annotation models,establishing cross-regional datasets,and enhancing semi-automatic annotation systems.Perspectives:Despite significant progress in automatic annotation methods,challenges such as cold start remain.The field urgently needs larger-scale cross-regional datasets and efficient semi-automatic annotation systems to ensure quality control to meet the increasing demands for both annotation volume and accuracy.