The key idea underlying project GLOCAL is to use events as the primary means for organizing and indexing media. Events have a local and a global dimension. The local dimension involves the assignment of tags (conceptualizations) to media (personal experiences). The global dimension involves the sharing of general event structures and specific event descriptions, which enables social sharing and networking of events, tags and media. Within networked communities, common (global) descriptions of the world can be built and continuously enriched by a continuous flow of individual (local) descriptions. One of the project enabling technologies deals with the analysis of multimedia content. The running GLOCAL project focuses mainly on text, image and video analysis. The recent study of user requirements showed that the project would benefit from advanced speech recognition components. The proposed GLOCAL project extension will bring additional functionality in the area of context-aware semantic annotation of multimedia files based on sophisticated audio and video processing. The search on the multimedia files will overcome limits of the current "one-best" voice transcription systems by employing graph-based representation of all probable alternative speech transcripts. A special attention will be paid to the detection of OOV (out-of-vocabulary) words. In contrast to the existing state-of-the-art speech processing systems, the GLOCAL-EEU will also provide sufficient accuracy when searching for new names of persons, places, companies etc. To address this, the project will combine the vocabulary-based speech recognizer with a keyword spotting module implementing the functionality of phonetic search. The addition of BUT to the consortium will help to lessen the impact of the semantic gap by exploiting the local and global (GLOCAL) knowledge about events and related contents to locally index media and experiences and later to globally share them within networked communities. This will improve the end-user experience with the GLOCAL-enhanced technologies and hence make the project more effective and more efficient.

Description in Czech
The key idea underlying project GLOCAL is to use events as the primary means for organizing and indexing media. Events have a local and a  global dimension. The local dimension involves the assignment of tags (conceptualizations) to media (personal experiences). The global dimension involves the sharing of general event structures and specific event descriptions, which enables social sharing and networking of events, tags and media. Within networked communities, common (global) descriptions of the world can be built and continuously enriched by a  continuous flow of individual (local) descriptions. One of the project enabling technologies deals with the analysis of multimedia content. The running GLOCAL project focuses mainly on text, image and video analysis. The recent study of user requirements showed that the project would benefit from advanced speech recognition components. The proposed GLOCAL project extension will bring additional functionality in the area of context-aware semantic annotation of multimedia files based on sophisticated audio and video processing. The search on the multimedia files will overcome limits of the current "one-best" voice transcription systems by employing graph-based representation of all probable alternative speech transcripts. A special attention will be paid to the detection of OOV (out-of-vocabulary) words. In contrast to the existing state-of-the-art speech processing systems, the GLOCAL-EEU will also provide sufficient accuracy when searching for new names of persons, places, companies etc. To address this, the project will combine the vocabulary-based speech recognizer with a keyword spotting module implementing the functionality of phonetic search. The addition of BUT to the consortium will help to lessen the impact of the semantic gap by exploiting the local and global (GLOCAL) knowledge about events and related contents to locally index media and experiences and later to globally share them within networked communities. This will improve the end-user experience with the GLOCAL-enhanced technologies and hence make the project more effective and more efficient.

Keywords
event analysis, speech processing

7E12055

English

Smrž Pavel, doc. RNDr., Ph.D. - principal person responsible

Department of Computer Graphics and Multimedia
- responsible department (6.6.2011 - not assigned)
Computer Graphics Research Group
- internal (6.6.2011 - 30.11.2012)
Knowledge Technology Research Group
- internal (6.6.2011 - 30.11.2012)
Speech Data Mining Research Group BUT Speech@FIT
- internal (6.6.2011 - 30.11.2012)
Department of Computer Graphics and Multimedia
- beneficiary (6.6.2011 - 30.11.2012)

KRÁL, J.; HRADIŠ, M. Restricted Boltzman Machines for Image Tag Suggestion. Proceedings of the 19th Conference STUDENT EEICT 2012. Brno: Brno University of Technology, 2012. p. 1-5.
Detail

HRADIŠ, M.; KOLÁŘ, M.; KRÁL, J.; LÁNÍK, A.; ZEMČÍK, P.; SMRŽ, P. Annotating images with suggestions - user study of a tagging system. Advanced Concepts for Intelligent Vision Systems. Lecture Notes in Computer Science. Lecture Notes in Computer Science. Brno: Springer Verlag, 2012. p. 155-166. ISBN: 978-3-642-33139-8. ISSN: 0302-9743.
Detail

Responsibility: Smrž Pavel, doc. RNDr., Ph.D.