November 15, 2016

MIVIA road audio events data set

The dataset

The MIVIA road audio events data set is composed of a total of 400 events for road surveillance applications, namely tire skidding and car crashes. The events are divided into 4 folds of 100 events each, in order to account for cross-validation experiments.


The sounds have been registered with an Axis P8221Audio Module and an Axis T83 omnidirectional microphone for audio surveillance applications , sampled at 32000 Hz and quantized at 16 bits per PCM sample. The audio clips are distributed as WAV files.
Each fold contains a number of audio file of about 1 minute duration, in which a series of hazardous events is superimposed to a typical road background sound. Each audio file has a different background sound, so that several different real situations are simulated.
The events of interest are organized in two classes (car crashes and tire skidding) and their total duration in the data set is reported in the following table

Training set
#Events Duration (s)
Background 2732
Car Crashes 200 326,38
Tire skidding 200 522,5


If you aim to use the data set, please refer to the following papers:

  • P. Foggia, N.Petkov, A. Saggese, N. Strisciuglio, M. Vento: Audio surveillance of roads: a system for detecting anomalous sounds In: IEEE, (Ed.): IEEE Transactions on Intelligent Transportation Systems, to appear
  • P. Foggia, A. Saggese, N. Strisciuglio, M. Vento (2014): Cascade Classifiers Trained on Gammatonegrams for Reliably Detecting Audio Events. In: IEEE, (Ed.): IEEE International Conference on Advanced Video and Signal-Based Surveillance (AVSS 2014), 2014, ISBN: 978-1-4799-4871-0/14


In order to download the data set click here.