Hasso-Plattner-Institut
  
 

Vision of the tele-TASK Project

In the center of the research work of Prof. Dr. Christoph Meinel and his team in the field of Knowledge Engineering and Web University is the tele-TASK project (tele-Teaching Anywhere Solution Kit). It was started more than 15 years ago when we began to research how Internet and web technologies can be used for supporting teaching (> tele-teaching) and learning (> e-learning). Our vision was to design an easy to use mobile system for recording and broadcasting university lectures and presentations over the Internet in order to develop and test on the one hand new tele-teaching and e-learning concepts and on the other hand to innovative portal and navigation techniques.

Demonstration des tele-TASK Systems

Our Research Directions

Over the years tele-TASK turned out to be a very fruitful project which helped us, on one side, to gain valuable experiences and a deeper understanding of e-learning and tele-teaching. On the other side, it inspires us to try our upcoming techniques in the are of Web3.0 - semantic, social, service Web -, and to make them accessible for Web-university (details). 

Some Links to the tele-TASK Portal

Here are some links to the tele-TASK portal which provides meanwhile more than 5.000 recorded telelectures: 

Buy, lease or rent tele-TASK

If you like to record and transmit your presentations over the Internet online and offline you can work with our tele-TASK recording system. Simply buy or rent tele-TASK ... 

The tele-TASK Team

  • Prof. Dr. Christoph Meinel (Head)
  • Dipl-Inf. Matthias Bauer
  • Dipl-Ing. Haojin Yang
  • Franka Grünewald, MSc.
  • Frank Priester (Technical Support)
  • Former Members: Volker Schillings, Tongbo Chen, Mingchao Ma, Mathias Kutzner, Bert Baumann, Long Wang, Andreas Groß, Maria Siebert, ...

tele-TASK Symposia

Scientific Publication about tele-TASK

Sentence Boundary Detection Based on Parallel Lexical and Acoustic Models

Xiaoyin Che, Sheng Luo, Haojin Yang, Christoph Meinel
In Proceedings of Interspeech 2016, pages 257-261, San Francisco, CA, USA, 9 2016

DOI: 10.21437

Abstract:

In this paper we propose a solution that detects sentence boundary from speech transcript. First we train a pure lexical model with deep neural network, which takes word vectors as the only input feature. Then a simple acoustic model is also prepared. Because the models work independently, they can be trained with different data. In next step, the posterior probabilities of both lexical and acoustic models will be involved in a heuristic 2-stage joint decision scheme to classify the sentence boundary positions. This approach ensures that the models can be updated or switched freely in actual use. Evaluation on TED Talks shows that the proposed lexical model can achieve good results: 75.5% accuracy on error-involved ASR transcripts and 82.4% on error-free manual references. The joint decision scheme can further improve the accuracy by 3~10% when acoustic data is available.

Keywords:

Sentence Boundary Detection, Parallel Models, Deep Neural Network, Word Vector

BibTeX file

@inproceedings{2016_Che_Interspeech,
author = { Xiaoyin Che, Sheng Luo, Haojin Yang, Christoph Meinel },
title = { Sentence Boundary Detection Based on Parallel Lexical and Acoustic Models },
year = { 2016 },
pages = { 257-261 },
month = { 9 },
abstract = { In this paper we propose a solution that detects sentence boundary from speech transcript. First we train a pure lexical model with deep neural network, which takes word vectors as the only input feature. Then a simple acoustic model is also prepared. Because the models work independently, they can be trained with different data. In next step, the posterior probabilities of both lexical and acoustic models will be involved in a heuristic 2-stage joint decision scheme to classify the sentence boundary positions. This approach ensures that the models can be updated or switched freely in actual use. Evaluation on TED Talks shows that the proposed lexical model can achieve good results: 75.5% accuracy on error-involved ASR transcripts and 82.4% on error-free manual references. The joint decision scheme can further improve the accuracy by 3~10% when acoustic data is available. },
keywords = { Sentence Boundary Detection, Parallel Models, Deep Neural Network, Word Vector },
address = { San Francisco, CA, USA },
booktitle = { Proceedings of Interspeech 2016 },
language = { English },
priority = { 0 }
}

Copyright Notice

last change: Fri, 14 Oct 2016 12:27:43 +0200

Other Links

... to our Research
              Security Engineering - Learning & Knowledge Tech - Design Thinking - former
... to our Teaching
              Tele-Lectures - MOOCs - Labs - Systems 
... to our Publications
              Books - Journals - Conference-Papers - Patents
... and to our Annual Reports.