On October 26 Mohamed Abou-Zleikha will give a talk at 13:00 in room A6-306.

Title: Exemplar Theory Meets Speech-based Technology

Abstract: Exemplar Theory was first introduced in psychology as a model of perception and categorization of concepts in the memory. It has emerged in several domains, AI reasoning (case-based reasoning) linguistics (phonetics, phonology, morphology, syntax and semantics), machine learning (example based classification and categorization) and speech processing (recognition, production, and synthesis). The presentation will give an overview of my work on implementing the exemplar theory for speech recognition and synthesis. It will address the use of an exemplar-based technique combined with a model-based technique for speech recognition. It will also give an overview of different exemplar-based techniques for the generation of prosody parameters, which underpin my PhD thesis.

Bio: Mohamed Abou-Zleikha is a final stage doctoral student with a specific research interest in speech technology.  He received his B.A. from Damascus University in Syria, and his M.Sc from Katholieke Universiteit Leuven in Belgium. His work focuses on the construction of computational models of speech. He is interested in cognitively inspired models and the use of these models for speech recognition and synthesis.  His Ph.D. work focuses on prosody parameters generation using an exemplar-based technique for speech synthesis.

On September 26, Jesper Rindom Jensen successfully defended his Ph.D. thesis. During the defense, Jesper gave a presentation of his work after which the thesis committee asked questions about the research. The thesis committe was comprised of Prof. Alle-Jan van der Veen (Delft University of Technology), Prof. Jingdong Chen (Northwestern Polytechnical University) and Prof. Bernard Fleury (Aalborg University). Jesper’s Ph.D. studies were carried out at the Dept. of Electronic Systems, where Jesper was employed before joining the Audio Analysis Lab, and he was supervised by Prof. Søren Holdt Jensen and Assoc. Prof. Mads Græsbøll Christensen. Congratulations to Jesper.

The Audio Analysis Lab is part of the new CoSound project. CoSound is a joint effort headed by Assoc. Prof. Jan Larsen at DTU, funded by the Danish Council for Strategic Research. The project is about a cognitive systems approach to enriched and actionable information from audio streams. The funding covers a three year postdoc position for a joint position with us and the Dept. of Electronic Systems, which is currently open. We had a nice kickoff workshop earlier this year and will be ramping up the activities during the fall.

As of August 1, the Villum Foundation project entitled Spatio-Temporal Filtering Methods for Enhancement and Separation of Speech Signals has officially started. The project aims at exploring fundamentally new ways of solving the cocktail party problem by generalizing novel optimal temporal filtering methods to multiple microphones, resulting in so-called spatio-temporal filtering methods. The filters are optimal in the sense that they let the signal of interest pass undistorted while everything else is attenuated. The Audio Analysis Lab has received ~7,000,000 DKK from the Villum Foundation’s Young Investigator Programme for the project. Jesper Rindom Jensen will be employed as postdoc as of August 15, and two Ph.D.-students join us during the fall. In January 2013, Prof. Jacob Benesty (University of Quebec) will come here as Guest Professor to work on the project for half a year. Prof. Andreas Jakobsson (Lund University) will also be part of the project. Nordjyske wrote a nice article about the project, which can be accessed here. You can also read more about the funded project on the Villum Foundation homepage.

Mads Græsbøll Christensen of the Audio Analysis Lab gave a tutorial at the Sound and Music Computing Conference entitled Parametric Pitch Estimators for Music Signals on June 11 2012. The slides can be downloaded by clicking here, and you can read more about the conference here.

The homepage for the newly founded Audio Analysis Lab at Aalborg University is now up and running! On this page, you can read news and background information about ongoing research projects as well as the people who are working on them. In time, you will also be able to access demos, source code (e.g., MATLAB) and shared experimental data.