CURE数据集

CURE数据集是由微软公司创建的，专门为听力损失人群设计的音频事件分类数据集。该数据集包含约6500个5秒的音频片段，涵盖13种与听力损失人群相关的特定音频事件。数据来源于Freesound在线资源库，经过人工转录生成准确的音频事件标签。CURE数据集旨在解决现实世界中听力损失人群面临的音频识别问题，特别是在智能手机等设备上的应用。The CURE dataset was developed by Microsoft Corporation, and is exclusively designed for audio event classification tasks targeting hearing-impaired populations. It comprises approximately 6,500 5-second audio clips, covering 13 specific audio events relevant to hearing-impaired individuals. The dataset's data is sourced from the Freesound online repository, with accurate audio event labels generated through manual transcription. The core goal of the CURE dataset is to address audio recognition challenges faced by hearing-impaired people in real-world scenarios, particularly for applications on devices such as smartphones.

搜集汇总

背景与挑战

背景概述

CURE数据集是一个针对听力障碍人士设计的音频事件分类数据集，包含精选的5秒音频片段，主要用于研究梯子网络在音频事件分类中的性能。该数据集通过比较不同分类器（如SVM、ELM和梯子网络）的性能，展示了梯子网络在分类准确性和鲁棒性方面的优势。