Deep Evaluation of Audio Representations (DEAR)
收藏资源简介:
DEAR数据集由卢塞恩应用科学大学等机构发布,包含1158个各为30秒的音频轨道,通过将专有独白与商业高质量日常声学场景录音混合制作而成。该数据集旨在评估音频表示基础模型在捕捉助听器所需的关键声学属性方面的性能,包含八个评估任务,涵盖了一般背景、语音来源和技术声学特性。数据集应用于助听器和助听设备领域,旨在解决声学场景分析中的挑战。
The DEAR Dataset was released by Lucerne University of Applied Sciences and Arts and other institutions. It comprises 1,158 30-second audio tracks generated by mixing proprietary monologues with commercial high-quality recordings of daily acoustic scenes. This dataset is designed to evaluate the performance of audio representation foundation models in capturing the critical acoustic properties required for hearing aids, and it includes eight evaluation tasks covering general acoustic backgrounds, speech sources, and technical acoustic characteristics. The dataset is applied in the field of hearing aids and assistive listening devices, aiming to address the challenges in acoustic scene analysis.

- 1Evaluation of Deep Audio Representations for Hearables卢塞恩应用科学大学 · 2025年



