Computational speech segregation based on monaural cues

Tobias May

One of the most impressive abilities of the human auditory system is that it is able to focus on a desired target source and to segregate it from competing talkers and interfering background noise. The main goal of this project is to develop a machine-based segregation system that is able to segregate speech in noisy environments by combining knowledge about auditory processing with modern signal processing strategies. The computational segregation system first decomposes the noisy speech mixture into individual time-frequency units. In a second step, the system attempts to classify these units into either speech-dominated or noise-dominated units by exploiting a variety of monaural cues, among them amplitude modulation and periodicity.

Supported by the European Marie Curie Initial Training Network “ICanHear” and by the European Commission and International consutium "Two!Ears" 

DTU Orbit                     

Contact

Tobias May
Assistant Professor
DTU Health Tech
+45 45 25 39 59