Neighbour’s drilling more irritating than traffic noise

Context plays an important role not only in the recognition of sound, but also how we perceive it. Information about the environment can therefore greatly improve the quality of sound-recognition computer software, cognitive scientist Maria Niessen has concluded. She will be awarded a PhD on 22 October 2010 for her research on automatic sound recognition.

Niessen developed a computer model for sound recognition that combines acoustic information with contextual information. Without information about the environment, noises can sound very similar, making it difficult to identify them. Without context, a cat’s purring can barely be distinguished from the sound of an engine.

Contextual information

Niessen added information about context to her model by training the computer with sound recordings. She did this by using the way people call upon their memories to process sounds. In humans, previous experiences and memories play an essential part in the interpretation of sounds in new situations. By calculating how probable it is that a sound occurs in a certain situation, Niessen’s model can determine whether a sound belongs to that environment or whether it is just background noise.

Neighbour

The context of a sound also influences how noise nuisance is perceived. The same sound can be experienced as a nuisance or not depending on the situation. For example, people interpret the noise of a scooter in a park in very different ways. The scooter rider is hardly bothered at all by the sound of the engine and may even enjoy listening to it, whereas someone who is sitting reading quietly on a bench will experience the scooter as disturbance as it races by. The same principle applies to neighbours: when a neighbour starts drilling this is much more likely to be perceived as noise nuisance than cars driving past that actually produce just as much noise.

Measuring nuisance

Niessen’s model greatly improves how noise nuisance is assessed. Current methods rely on noise measurements alone – decibels – to assess noise nuisance. Niessen believes this is insufficient: ‘It is only with extremely loud noises, such as a low-flying fighter jet passing overhead, that nearly everyone experiences the same noise as a nuisance. But up to about 70 dB it is the sound context that is defining. A model that also considers context makes it much easier to gain insight into the question of whether people actually experience a particular noise as a nuisance.’

Robots

The model can also be applied to robots that move independently through an unknown environment, such as robots that help the residents of a residential care home with household tasks. Niessen: ‘In research on such robots a lot of attention is paid to sight but not to sound, even though sound is particularly important for perceiving the environment and reacting to it.’

Curriculum vitae

Maria E. Niessen (Apeldoorn, 1980) studied Artificial Intelligence at the University of Groningen. Her research was partly funded by SenterNovem. Niessen will be awarded her PhD by the Faculty of Mathematics and Natural Sciences. Her supervisors were Prof. L. R. B. Schomaker and Dr T. C. Andringa. Since February, Niessen has been working as a postdoctoral researcher at INCAS3, where she (in collaboration with L'équipe ‘Lutheries - Acoustique – Musique’ in Paris) is continuing her research into noise nuisance. The title of her thesis is ‘Context-Based Sound Event Recognition’.

Note to the press

Further information: Maria Niessen, tel. +33 (0)1 53954321, e-mail: marianiessen incas3.eu