Dataset of everyday sounds helps develop devices that hearRobots are already able to “hear”, but soon they may also understand what they are hearing. Researchers of Tampere University of Technology have published an audio database of everyday sounds to further develop comprehension. The material is also openly available to other researchers.
The Audio Research Group at TUT has released a dataset of audio recordings from everyday environments, such as parks, streets, homes, offices, stores and restaurants. The recordings contain sounds of cars passing by, people walking or talking, birds singing, children yelling, and other sounds common in our everyday life.
The release will help improve automatic sound recognition, which is a prerequisite for creating devices that hear – whether they are applications on mobile phones, autonomous cars, or full-fledged robots.
“Sound recognition capabilities will enable devices to understand what they hear and present this information to their human user. Possible applications include a wide range of daily scenarios, such as home alarm systems for the hearing impaired or monitoring noise pollution sources or remote bird populations,” Postdoctoral Researcher Annamaria Mesaros from TUT’s Department of Signal Processing explains.
The dataset that has been made available contains nearly ten hours of audio recorded in Tampere and Helsinki. The database is openly available to anyone who wishes to use it in their work.
“This database is a great asset for the scientific community. This is the most extensive collection publicly available, as collecting and labelling the material requires a lot of time and effort.”
“Some of the recordings included in the database have been annotated – in other words, labelled – to include information about individual sound sources, marking the exact times where their occurrences have been identified. Detailed annotations of this kind facilitate the development of precise automatic recognition methods,” Mesaros continues.
Ongoing international competition
Through the database release, TUT supports open access to data and fosters collaborative relationships with other institutions in the field. To further support innovation and advancing the state-of-the-art solutions in the field, the Audio Research Group is organizing an international competition. A related workshop will be organized in Budapest in early September. The challenge and workshop will be arranged together with Queen Mary University of London (UK), University of Surrey (UK) and Institut de Recherche en Communications et Cybernétique de Nantes (France).
The challenge has attracted over 150 participants from 79 international universities and companies, competing to develop the best algorithm for sound recognition.
“I am excited to see what kind of recognition methods they developed using our dataset”, Mesaros says.
Further information: Postdoctoral Researcher Annamaria Mesaros, email@example.com and Academy Research Fellow Tuomas Virtanen, firstname.lastname@example.org
Further information on the research: http://www.cs.tut.fi/sgn/arg/dcase2016/