A city street grid seen from above, with acoustic ripples blooming over the busier junctions

Sound Events and Urban Soundscapes

A recording of a street carries more than noise. It carries what happened there, how often, and at what time of day — information that no decibel meter reports, because a loud street and an unpleasant street are not the same thing.

This is the line of work where our datasets, our models and a running public installation meet.

An acoustic map of a city

SoundscapeAI is an interactive map of how Athens actually sounds. Crowdsourced recordings are analysed by our audio models and aggregated into a spatiotemporal view of acoustic quality and acoustic events, by area and by time of day. It is built directly on our work on soundscape quality recognition, which we have approached both with hand-crafted and with deep audio features.

Open datasets

ATHUS is a dataset for urban soundscape quality recognition in Athens. ARCHEO covers sound event detection in areas of touristic interest, and is released together with code. Both exist because the question — what does this place sound like, and is that good or bad for the people in it — cannot be answered with the datasets the field had.

From a recording to an event

Underneath the soundscape work is a longer line on detecting events in continuous audio: fusing multiple audio sensors for acoustic event detection, recognising daily activities by meta-classification of low-level audio events, and packaging the whole thing as a ROS framework so it can run on a robot rather than in a notebook.

Where this runs

In MI-TRAP, an EU project on transport-related air pollution, we provide two real-time components for the project’s environmental monitoring stations: soundscape quality assessment from continuous audio, and vehicle counting from a single low-cost camera. Sound event recognition services built on our deepaudio-lab platform are also part of our contribution to the PHAROS AI Factory.

Selected publications

  • Giannakopoulos, T., Orfanidi, M., & Perantonis, S. Athens Urban Soundscape (ATHUS): A Dataset for Urban Soundscape Quality Recognition. In International Conference on Multimedia Modeling (pp. 338-348). Springer, Cham. (2019) doi
  • Giannakopoulos, T., & Perantonis, S. Recognizing the quality of urban sound recordings using hand-crafted and deep audio features Proceedings of the 12th ACM International Conference on PErvasive Technologies Related to Assistive Environments (2019) doi
  • Psallidas, T., Mitsou, A., Pikramenos, G., Spyrou, E., & Giannakopoulos, T. ARCHEO: A Dataset for Sound Event Detection in Areas of Touristic Interest 2020 15th International Workshop on Semantic and Social Media Adaptation and Personalization (SMA (pp. 1-6). IEEE. doi
  • Giannakopoulos, T., Siantikos, G., Perantonis, S., Votsi, N. E., & Pantis, J. Automatic soundscape quality estimation using audio analysis Proceedings of the 8th ACM International Conference on PErvasive Technologies Related to Assistive Environments (p. 19) (2015) doi
  • Siantikos, G., Sgouropoulos, D., Giannakopoulos, T., & Spyrou, E. Fusing multiple audio sensors for acoustic event detection Image and Signal Processing and Analysis (ISPA), 2015 9th International Symposium on (pp. 265-269). IEEE doi
  • Giannakopoulos, T., & Konstantopoulos, S. Daily Activity Recognition based on Meta-classification of Low-level Audio Events Proceedings of ICT4AWE2017, ISBN: 978-989-758-251-6 doi
  • Giannakopoulos, T., & Siantikos, G. A ROS Framework for Audio-Based Activity Recognition Proceedings of the 9th ACM International Conference on PErvasive Technologies Related to Assistive Environments (PETRA 2016) doi
  • Eleftheriadis, K., Gini, M., Manousakas, M., Diapouli, E., Vratolis, S., Papagiannis, S., Zografou, O., Giannakopoulos, T., Konstantopoulos, S., Mocnik, G., & Drinovec, L. MItigating Transport-Related Air Pollution in Europe: The MI-TRAP project The European Aerosol Conference (2024)