Mel Spectrograms of Trains' Transit for Railway Monitoring

This dataset consists of a collection of mel spectrograms derived from audio recordings of train passages on a railway line in Bologna. The recordings were acquired using a custom-built Internet of Things (IoT) device, specifically designed and assembled to facilitate high-fidelity acoustic monitoring. The dataset encapsulates time-frequency representations of train-generated acoustic signals, providing a structured resource for tasks such as train classification, carriage counting and train speed estimation. The mel spectrograms have been generated through a pre-processing pipeline to ensure consistency and robustness in feature extraction. Detailed information on the data acquisition process, including hardware specifications, acquisition conditions and pre-processing methods, is fully documented in the accompanying research paper: "RATTLE: a Framework for Train Characterisation and Identification through Audio Fingerprinting". This dataset serves as a valuable resource for advancing research in railway monitoring, environmental acoustics, and deep learning applications. The dataset is organized as follow: labels.csv Mel-spectograms.zip Lables.csv is the file which contains all the mel-spectograms images' names labelled with train model and provides the number of carriages associated to the trains, manually annotated, and the speed estimate calculated as an approximation derived from our custom algorithm. Mel-spectograms.zip contains the PNG files.