FedAudio
收藏资源简介:
FedAudio是由南加州大学电气与计算机工程系开发的一个联邦学习音频任务基准。该数据集包含四个代表性的音频数据集,涵盖关键词识别、语音情感识别和声音事件分类三个重要任务。FedAudio特别引入了数据噪声和标签错误,以模拟真实世界中部署联邦学习系统时的挑战。此数据集不仅包括基准测试结果,还提供了一个PyTorch库,旨在帮助研究人员公平比较他们的算法。FedAudio有望成为音频任务的参考联邦学习基准,推动声学和语音研究领域的发展。
FedAudio is a federated learning audio task benchmark developed by the Department of Electrical and Computer Engineering at the University of Southern California. This dataset includes four representative audio datasets, covering three critical tasks: keyword spotting, speech emotion recognition, and sound event classification. FedAudio specifically introduces data noise and label errors to simulate the challenges encountered when deploying federated learning systems in real-world scenarios. In addition to benchmark test results, this dataset also provides a PyTorch library designed to help researchers fairly compare their algorithms. FedAudio is expected to become a reference federated learning benchmark for audio tasks, promoting the advancement of the acoustic and speech research field.




