VocalSet (VocalSet: A Singing Voice Dataset)

VocalSet 是一个歌声数据集，由 10.1 小时的专业歌手单声道录音组成，展示了所有 5 个元音的标准和扩展声乐技巧。现有的歌声数据集旨在捕获歌声特征的一个集中子集，通常只包含几个歌手。 VocalSet 包含来自 20 位不同歌手（9 位男性，11 位女性）和一系列声音类型的录音。 VocalSet 旨在改善现有歌声数据集和歌声研究的状态，不仅捕捉一系列元音，而且捕捉许多不同声乐技巧的不同声音，在音阶、琶音、长音和摘录的背景下演唱.VocalSet is a singing voice dataset comprising 10.1 hours of monophonic recordings from professional singers, showcasing standard and extended vocal techniques for all five vowels. Existing singing voice datasets typically aim to capture a concentrated subset of vocal characteristics and usually only include a small number of singers. VocalSet contains recordings from 20 distinct singers (9 male, 11 female) spanning a range of voice types. Designed to advance the state of the art in existing singing voice datasets and singing voice research, VocalSet not only captures a diverse set of vowels but also distinct renditions of many different vocal techniques, performed in the contexts of scales, arpeggios, sustained notes, and excerpted segments.

搜集汇总

背景与挑战

背景概述

VocalSet是一个包含20位专业歌手(9男11女)10.1小时录音的歌声数据集，特色是覆盖5个元音的标准和扩展声乐技巧，填补了现有歌声数据集在声乐技巧多样性方面的不足。该数据集由西北大学于2018年发布，旨在促进歌声研究发展。