DNS Challenge

该数据集包含了用于训练的清洁语音子集，其中包含了噪声和房间脉冲响应信息。此外，数据集还包含了如VCTK和AISHELL-1等不同子集，特别关注多样化的失真情况。这是一个大规模的数据集，涵盖了多种失真类型，其任务主要针对语音恢复。This dataset includes a clean speech subset for training purposes, which incorporates noise and room impulse response information. Additionally, it encompasses diverse subsets such as VCTK and AISHELL-1, with a particular focus on varied distortion conditions. As a large-scale dataset, it covers a wide range of distortion types, and its primary application task centers on speech restoration.

搜集汇总

背景与挑战

背景概述

DNS Challenge是一个旨在促进单通道语音增强研究的全球性挑战，专注于提升语音的感知质量和可懂度。挑战分为实时和非实时两个轨道，提供大规模训练和测试数据集，并允许使用外部数据，最终通过盲测集进行主观评价。