LOUD – Labeled nOise aUdio Dataset

The paper presents a new dataset on ambient noise. The dataset consists of 21 sound recordings reflecting indoor and outdoor noise. All recordings were made under consistent, controlled conditions to ensure high-quality and balanced data. A unique feature of the dataset is that it consists of long-duration recordings (up to 30 minutes). This makes it suitable for both continuous audio analysis and event classification tasks (by segmenting recordings into shorter recordings). We present an analysis of the acoustic parameters and a classification performance using spectrograms and a convolutional neural network (CNN), achieving an overall accuracy of 94.69%.