论文信息 - CUNI at MediaEval 2014 Search and Hyperlinking Task: Visual and Prosodic Features in Hyperlinking

CUNI at MediaEval 2014 Search and Hyperlinking Task: Visual and Prosodic Features in Hyperlinking

In this report, we present our experiments performed for the Hyperlinking part of the Search and Hyperlinking Task in MediaEval Benchmark 2014. Our system successfully combines features from multiple modalities (textual, visual, and prosodic) and confirms the positive eect of our former method for segmentation based on Decision Trees.

[1] Thomas Seidl,et al. Signature Quadratic Form Distance , 2010, CIVR '10.

[2] Mark J. F. Gales,et al. Automatic Transcription of Multi-genre Media Archives , 2013, SLAM@INTERSPEECH.

[3] Jean-Luc Gauvain,et al. Speech Processing for Audio Indexing , 2008, GoTAL.

[4] Maria Eskevich,et al. Linking inside a video collection: what and how to measure? , 2013, WWW.

[5] Martin Krulis,et al. Combining CPU and GPU architectures for fast similarity search , 2012, Distributed and Parallel Databases.

[6] Pavel Pecina,et al. CUNI at MediaEval 2014 Search and Hyperlinking Task: Search Task Experiments , 2014, MediaEval.

[7] Martin Krulis,et al. Efficient Extraction of Feature Signatures Using Multi-GPU Architecture , 2013, MMM.

[8] Maria Eskevich,et al. Adapting Binary Information Retrieval Evaluation Metrics for Segment-based Retrieval Tasks , 2013, ArXiv.

[9] Paul Deléglise,et al. Enhancing the TED-LIUM Corpus with Selected Data for Language Modeling and More TED Talks , 2014, LREC.

[10] Maria Eskevich,et al. The Search and Hyperlinking Task at MediaEval 2013 , 2013, MediaEval.

[11] Björn W. Schuller,et al. Recent developments in openSMILE, the munich open-source multimedia feature extractor , 2013, ACM Multimedia.