ReverbFX
收藏资源简介:
ReverbFX是一个新型的房间脉冲响应(RIR)数据集,专为歌唱声音去混响研究而设计。与基于实际录制的RIRs的现有数据集不同,ReverbFX拥有一系列从音乐制作中常用的各种混响音频效果插件中捕获的RIRs。我们使用该数据集进行了全面的实验,以基准测试受人工混响影响的歌唱录音的去混响挑战。我们使用ReverbFX训练了两种最先进的生成模型,并证明在人工混响场景中,使用插件派生的RIRs训练的模型优于仅在真实RIRs上训练的模型。
ReverbFX is a novel room impulse response (RIR) dataset specifically designed for singing voice dereverberation research. Unlike existing datasets based on real-recorded RIRs, ReverbFX features a collection of RIRs captured from various reverberation audio effect plugins commonly used in music production. We conducted comprehensive experiments using this dataset to benchmark the dereverberation challenge for singing recordings contaminated by artificial reverberation. We trained two state-of-the-art generative models with ReverbFX, and demonstrate that models trained on plugin-derived RIRs outperform those trained solely on real RIRs in artificial reverberation scenarios.




