The DNLE dataset contains 1668 environmental background noise recordings (WAV files) labeled according to type and level of noise. The recordings are approximately equally balanced between three main categories, i.e., "mechanic", "melodic", and "quiet". These noise categories were selected as we found in a previous study that these are the background noises that can distract users in crowdsourcing when performing tasks. The recordings were collected through the audio-web API employing different Windows and Mac computers. This dataset is part of my PhD dissertation.