Sven Behnke
 Multimodal Communication and Speech Processing

Multimodal communication via speech, mimics, eye gazes, and body language makes intuitive interaction between humans and machines possible.
In the project "Learning Humanoid Robots", I participate in the development of humanoid communication robots that will be used as museum guide.

I am interested in computational auditory scene analysis, the discovery of speech features, and noise robustness.
My goal is to transfer the hierarchical approach with local recurrent connectivity that proved to be useful for visual perception to the auditory domain.

Since a common fundamental is one of the main cues for auditory grouping, I am interested in finding and tracking the fundamental frequency of the speaker of interest in order to use the harmonic structure of voiced speech to separate the speaker from the background.

Publications