MedleyVox
收藏资源简介:
MedleyVox是由Gaudio Lab, Inc.和首尔国立大学合作创建的一个评估数据集,专门用于多重歌唱声音分离的研究。该数据集包含了23首歌曲的声乐轨道,分为四种不同的分离类别:齐唱、二重唱、主唱与和声以及其他多重歌唱分离。数据集的创建过程中,使用了多种单声歌唱数据集来构建多重歌唱混合物,并提出了改进的超分辨率网络(iSRNet)以提高分离网络的初步估计。MedleyVox旨在为音乐源分离研究提供一个基准数据集,特别是在解决流行音乐中的多重歌唱声音分离问题。
MedleyVox is an evaluation dataset co-developed by Gaudio Lab, Inc. and Seoul National University, specifically dedicated to research on multiple singing voice separation. The dataset includes vocal tracks from 23 songs, categorized into four distinct separation types: unison singing, duet singing, lead vocal with harmony, and other multiple singing voice separation tasks. During its construction, multiple monophonic singing voice datasets were utilized to generate multiple singing voice mixtures, and an improved super-resolution network (iSRNet) was proposed to refine the preliminary estimations of separation models. MedleyVox aims to provide a benchmark dataset for music source separation research, particularly for addressing the multiple singing voice separation problem in popular music.

- 1MedleyVox: An Evaluation Dataset for Multiple Singing Voices SeparationGaudio Lab, Inc., Seoul, South Korea 2Department of Intelligence and Information, Seoul National University 3 Interdisciplinary Program in Artificial Intelligence, Seoul National University 4 AI Institute, Seoul National University, Seoul, South Korea · 2023年



