0x3/audioset-nonspeech

Audioset-Nonspeech 是一个处理过的知名数据集 agkphysics/AudioSet 的版本。处理过程移除了所有可能包含清晰可辨人声的音频录音，仅保留非语音音频录音。该数据集不仅可用于音频事件分类，还可在训练语音识别模型时，用于以指定信噪比与语音录音混合进行增强。每个示例提供音频数据链接和以分号分隔的类别标签列表。

Audioset-Nonspeech is a processed version of the well-known agkphysics/AudioSet dataset. The processing was performed to remove all audio recordings that may contain clearly distinguishable human speech, leaving only non-speech audio recordings. The resulting dataset can be used not only for audio event classification but also for augmentation (mixing with a specified signal-to-noise ratio) of speech recordings when training speech recognition models. For each example, a link to the audio data and a list of class labels separated by semicolons are provided.