Libri2Vox
收藏资源简介:
Libri2Vox数据集是由日本国立信息学研究所和新加坡科技设计大学合作创建的,旨在解决目标说话者提取(TSE)任务中的数据多样性和鲁棒性问题。该数据集结合了LibriTTS的干净目标语音和VoxCeleb2的噪声干扰语音,提供了在真实噪声环境下的多样化说话者集合。数据集通过合成语音生成模型进一步增强了说话者的多样性,并采用了课程学习策略来逐步训练TSE模型。Libri2Vox数据集主要应用于语音处理领域,特别是在语音控制系统和远程会议等场景中,旨在提高语音信号提取的准确性和鲁棒性。
The Libri2Vox dataset was collaboratively developed by the National Institute of Informatics (NII) of Japan and the Singapore University of Technology and Design (SUTD), aiming to address the issues of data diversity and robustness in the Target Speaker Extraction (TSE) task. This dataset combines clean target speech from LibriTTS and noisy interfering speech from VoxCeleb2, providing a diverse set of speakers in real-world noisy environments. It further enhances speaker diversity through synthetic speech generation models and adopts a curriculum learning strategy for progressive training of TSE models. The Libri2Vox dataset is primarily applied in the field of speech processing, particularly in scenarios such as voice control systems and remote conferences, with the goal of improving the accuracy and robustness of speech signal extraction.




