In 2014 NIST organized a competition in speaker recognition as part of the Odyssey- Langauge and Speaker Recognition Workshop, which is the one of the most established scientific communication channels in the field of speaker recognition. More than 120 researchers entered the competition and together contributed more than 9000 distinct systems for evaluation. The system developed by the University of Ljubljana and our colleagues from the SME Alpineon incorporated duration information into the processing pipeline and ranked among the top ten performers of the competition. Our result in the competition contributed to the visibility of the university of Ljubljana and Alpineon in the speaker-recognition community. The system that was used in the competition is described in [COBISS.SI-ID 10818644].
E.02 International awards
COBISS.SI-ID: 10818644In 2014 the University of Colorado, the University of Notre Dame and NIST organized an international face recognition competition as part of the international joint conference on biometrics (IJCB'14). The goal of the competition was to assess the performance of state-of-the-art face recognition systems and to identify future research direction in the field of face-based biometrics. Several established international research institutions participated in the competition, including the Stevens Institute of Technology, The University of Campinas, the Advanced Digital Science Center in Singapore and Slovenia’s University of Ljubljana among others. The competition was conducted on the challenging PaSC database. Our system (i.e., the system of the University of Ljubljana) resulted in the best performance on one of the two experimental protocols of the competition. When implementing the prototype face recognition system for the competition, we gathered novel knowledge and important insights into the problems and challenges met when developing fully functional (deployable) face recognition systems.
E.02 International awards
COBISS.SI-ID: 10932564We present the configuration and the development of the Slovenian emotional speech databases developed for purposes of speech synthesis and automatic emotion recognition. The main focus is about the development of methodology and software used to label the paralinguistic information from speech. The design of the database and development of the software for crowd-sourcing was produced and developed at the Laboratory of Artificial Perception, Systems and Cybernetics at Faculty of Electrical Engineering of Ljubljana. The currently annotated database consists of speech signals extracted from 17 radio dramas, with the academic licence for processing and annotating the audio signals authorized by from RTV Slovenia. For purposes of testing the developed crowd-sourcing software we focused in labelling emotional speakers states of one male and one female speaker. The emotional labels were annotated using the developed web based application with five volunteers. In this article we present the implementation of web based application for crowd-sourcing based on CMS Plone and annotating procedure which results in emotional speech database consisting of 1110 recordings. We additionally focus in the problems of annotating the speech corpora in the crowd-sourcing environment for annotating the paralinguistic informations from speech and on the example of the annotated database we report about the obtained annotations based on annotators majority vote.
F.15 Development of a new information system/databases
COBISS.SI-ID: 10810708The proceedings contain papers presented at IS-JT 2014, The Ninth Language Technologies Conference held on October 9th, 10th 2014 in Ljubljana. The proceedings contain 31 contributions, which present a wide variety of research topics. Three papers, of which two were invited contributions, present the CLARIN research infrastructure, which aims to facilitate research in the humanities, social sciences by enabling access to language resources, services. Several papers present results or plans for European on national research projects. A special mention should be given to the numerous papers by our Croatian colleagues, where they report on the compilation of new language resources, machine learning methods applied to a wide spectrum of linguistic annotation tasks. The proceedings also contain descriptions of research on speech technologies, corpus linguistic research, overview papers, and presentations of applications.
B.01 Organiser of a scientific meeting
COBISS.SI-ID: 275927552The invention describes a novel method and device for capturing depth or 3D images, that uses pattern projection onto the observed scene. Prior work includes depth imaging methods that make use of a scene that has been illuminated by narrow beams of infra-red light (WO 2007/043036 A1, US 8.150.142 B2). Our research goal was set to find a method and device for capturing depth images od scenes, whereby the depth of all points on the scene should be determined in a simple manner without the necessity to determine the correspondence between the projected pattern and the image of this pattern. The major advantage of the proposed method and device for capturing depth images is that the depth of any point in the scene can be determined just by determining the distance of its image from the sensor spot that has been foreseen in advance for this particular point.
F.33 Slovenian patent
COBISS.SI-ID: 10948948