Skip to ContentSkip to Navigation
About usNews and EventsNews articles

Signal-driven sound processing for uncontrolled environments

29 October 2010

PhD ceremony: Mr. J.D. Krijnders, 13.15 uur, Academiegebouw, Broerstraat 5, Groningen

Thesis: Signal-driven sound processing for uncontrolled environments

Promotor(s): prof. L.R.B. Schomaker

Faculty: Mathematics and Natural Sciences


Recent developments in soundscape research and systems for ambient awareness have shown a need for a new range of sound classication and recognition algoritms, because the results of current systems are rather limited. So, why is automatically extracting useful information from many sonic environments not yet successful? The applications of the recent developments require sound source recognition work in complex environments and with exible tasks. For some of these applications, for example acoustic aggression detection in the public space, the desired information is a single bit:“are there aggressive vocalizations or not?”, for other applications, like in soundscape research, a richer description is required.

Existing techniques for sound recognition are designed to function in closed, specialized domains. Speech recognition and music genre recognition, for example, work under the condition that the input is what they expect; speech from the speaker the system was trained on, or clean music recordings respectively. The idea that these closed domain techniques will generalize to open environments has so far not materialized. To operate in open environments we need to focus on the constancy and invariants in the signal: the physics that produced it.

In contrast to current “engineered”, specic systems, we aim to developsignal processing techniques that can handle sound in uncontrolled environments. Such environments are outside the range of the problems solved by current techniques, but are the normal environment for humans. These novel techniques are based on the research questions: “How to select sonic evidence that is likely to stem from a single source from a sound signal recorded in realistic acoustical circumstances?” and “How can the signal, instead of the system design, guide the processing of the signal, towards an optimal rendering of the information in the signal?”.


Last modified:15 September 2017 3.39 p.m.

More news