南京大学学报（自然科学版）2025，Vol.61Issue(5)：752-762,11.DOI:10.13232/j.cnki.jnju.2025.05.004
面向深度学习的个性化HRTF数据库的构建与分析
Establishment and analysis of personalized HRTF database for deep learning
摘要
Abstract
The personalized Head-Related Transfer Function(HRTF)plays a crucial role in implementing immersive audio experiences.However,existing public HRTF databases often lack sufficient scale and diversity to effectively support the training of deep learning models,thereby hindering the advancement of personalized HRTF prediction.To address this challenge,this study constructed a large-scale HRTF database tailored for deep learning applications,integrating real HRTF measurements obtained in anechoic chambers and simulated HRTFs derived from high-precision head scans.Using both objective acoustic metrics and subjective listening experiments,the study systematically analyzed the significant differences between generic and personalized HRTFs across key frequency bands,particularly in the high-frequency range(>10 kHz).These differences were found to have a substantial impact on auditory localization accuracy,with the high-improvement group achieving an average enhancement of 5.9°.The database and its analysis not only provide a comprehensive data foundation and validation resource for deeplearning-driven HRTF prediction,but also offer empirical insights into the significance of personalization and the optimization of virtual auditory experiences.关键词
个性化头相关传递函数/深度学习/边界元仿真/声源定位Key words
HRTF individualization/deep learning/boundary element method simulation/sound source localization分类
数理科学引用本文复制引用
刘嘉伟,吕朋博,林志斌..面向深度学习的个性化HRTF数据库的构建与分析[J].南京大学学报（自然科学版）,2025,61(5):752-762,11.基金项目
国家自然科学基金(12274221) （12274221）
