M-Group – Privacy-Aware Machine Listening for the Detection of Sound Events in Smart Cities

To ensure sustainable development in public outdoor places, such as urban areas and cities, it is of paramount importance to guarantee the quality of life of its inhabitants. One of the main concerns is environmental noise. When, for example, considering the reduction in road traffic during the COVID19 lockdown, many Europeans have started to realize that their cities and parks can be much more breathable, peaceful and quiet. According to the European environment agency more than 100 million people in Europe are exposed to long-term noise levels that are harmful to their wellbeing health . It is explained that inappropriate levels of exposure have proven effects such as sleep disruption and long-term effects such as annoyance, hypertension, sleep disturbance, heart disease.

However, sound analysis using machine learning and artificial intelligence requires a large amount of data. Privacy regulations forbid the recording of speech, which may pose difficulties in analyzing the data for detecting sound events. To analyse the data, recordings need to be done. This poses a risk of recording speech, even if the end user is not interested in speech. Privacy-aware methodologies take into account the privacy aspect. There are several ways of performing this. For example, edge computing analyses the data near the sensor device and only outputs the results. Other possibilities use a two-step approach by detecting speech fragments first and discarding these fragments in further processing.

Bibliography
– European Environmental Agency. Environmental Topics, Environment and Health, Noise. Available online: https://www.eea.europa.eu/publications/environmental-noise-in-europe
– Wei Wang, et al.. 2019. Privacy-aware environmental sound classification for indoor human activity recognition. In Proceedings of the 12th ACM International Conference on PErvasive Technologies Related to Assistive Environments (PETRA ’19). Association for Computing Machinery, New York, NY, USA, 36–44. DOI: https://doi.org/10.1145/3316782.3321521
– Cory Cornelius et al. . 2008. Anonysense: privacy-aware people-centric sensing. In Proceedings of the 6th international conference on Mobile systems, applications, and services (MobiSys ’08). Association for Computing Machinery, New York, NY, USA, 211–224. DOI:https://doi.org/10.1145/1378600.1378624
– F. Chen, J. Adcock, and S. Krishnagiri, “Audio Privacy: Reducing Speech Intelligibility While Preserving Environmental Sounds,” in Proceedings of the 16th ACM International Conference on Multimedia, ser. MM ’08. ACM, 2008, pp. 733–736. [Online]. Available: http://doi.acm.org/10.1145/1459359.1459472
– D. Liaqat, E. Nemati, M. Rahman and J. Kuang, “A method for preserving privacy during audio recordings by filtering speech,” 2017 IEEE Life Sciences Conference (LSC), Sydney, NSW, 2017, pp. 79-82, doi: 10.1109/LSC.2017.8268148.
– S. Kumar, L. T. Nguyen, M. Zeng, K. Liu, and J. Zhang, “Sound shredding: Privacy preserved audio sensing,” in Proceedings of the 16th International Workshop on Mobile Computing Systems and Applications, ser. HotMobile ’15. New York, NY, USA: ACM, 2015, pp. 135–140. [Online]. Available: http://doi.acm.org/10.1145/2699343.2699366

The overall goal of this thesis is to examine and develop a privacy-aware methodology or algorithm to detect sounds taking into account the privacy nature of speech. The overall goal can be subdivided in to different subgoals:

• Identify and perform a literature study on privacy-aware data collection involving persons.
• The student has to be able to list the advantages and disadvantages.
• The student will also study the dataset SONYC-UST which can be used to design and develop a proof-of-concept
• Guided by your supervisors, the student will choose a methodology which can demonstrate a classification of a sound event in a privacy aware manner. The methodology or algorithm needs to run on an embedded device. Therefore, the methodology needs to be designed and developed with constrained resources in mind.
• This methodology needs to be tested and evaluated with respect to accuracy and throughput.

This project is a collaboration between the campus of Brugge and Geel and with the IT department of Stad Brugge.