Start ReadSpeaker XT

Contact




Seminar at Speech, Music and Hearing:

Intelligent hearing - blind speech extraction

Nedelko Grbic, Blekinge Tekniska Högskola

Abstract

This presentation deals with the problems involved in creating an acoustic system that has the ability to, completely automated, discern acoustic speech sources from non-speech sources. Additionally, the system needs to counteract the impact of the acoustic propagation channel, i.e. to perform de-reverberation. The system assumes no prior knowledge of the physical environment, in which it operates. The main direction of acoustic signal processing has recently changed from a traditional view founded in the physical interpretation of a filtering problem into new blind algorithmic formulations. Most of these new formulations include optimization criteria based on operability in only a subset of the existing physical domains. While many of these algorithms are optimal in the sense that they successfully locate the optimum of their respective criterion, they lead to sub-optimal solutions for the physical problem at hand. Attempts to combine different algorithms have shown to be a challenge since different optimization criteria leads to different physical realizations and therefore the problem formulation becomes suboptimal. This presentation will present a historical overview together with a description of “state of the art” solutions in the field. The aim is to describe the common ground for all the existing methods, namely the physical domains in which they all operate. A description of a fundamentally new blind solution (“Blind Signal Extraction”), proposed by the author and currently accepted as a worldwide patent (PCT), will also be given. This method combines operability in all the existing physical domains and it thereby provides an optimal solution to otherwise unsolved problems. While the method is formulated to extract many different types of source signals, the main focus will be speech extraction in this presentation. The presentation includes real time demonstrations where the main goal is to provide the audience a hands-on feeling of what can be achieved with the new method. The aim is to incorporate the solution into a variety of intelligent human hearing devices, such as, hearing aids, hearing protection, assistive listening devices, teleconferencing systems, mobile phones etc.

15:15 - 17:00
Tuesday December 8, 2009

The seminar is held in Fantum.

| Show complete seminar list







Published by: TMH, Speech, Music and Hearing
Webmaster, webmaster@speech.kth.se

Last updated: Wednesday, 23-Jun-2010 09:22:46 MEST