FluidInference/musan

MUSAN是一个包含音乐、语音和噪声录音的语料库，设计用于训练用于语音活动检测和音乐/语音区分的模型。这个综合性的收集适用于各种音频处理任务。数据集总时长约为109小时，包含音乐、语音和噪声三个类别，音频格式为WAV。音乐类别包括古典、流行/摇滚、爵士等子类别；语音类别主要是英语，包含不同演讲者朗读书籍和美国政府记录；噪声类别包括环境声音和技术噪音等。所有文件都遵循Creative Commons授权或美国公有领域（无商业使用限制）。

MUSAN is a corpus of music, speech, and noise recordings designed for training models for voice activity detection and music/speech discrimination. The dataset comprises approximately 109 hours of audio across three categories: Music, Speech, and Noise, in WAV format. The Music category includes subcategories such as Classical, Pop/Rock, Jazz, etc., sourced from the Free Music Archive, Jamendo, and others. The Speech category is primarily in English, featuring various speakers reading books and government proceedings from sources like LibriVox and US Government recordings. The Noise category consists of environmental sounds and technical noises from Free Sound and other sound effect collections. All files in this corpus are available under Creative Commons licenses or in the USA Public Domain, with no restrictions on non-commercial use.