Kommunikationsnetze

Research project: Studies on the Importance of Speech Frames

Funding

Eliteprogramme for Postdocs by the Landesstiftung Baden-Württemberg gGmbH, Dr. Christian Hoene

Duration

May 2006 till November 2010.

Employees

  • Mark Schmidt
  • Nihar Jhawar
  • Daniel Plomp

Description

If speech is transmitted over cellular or IP based networks, it is compressed in order to remove redundancies. The compression stream of speech frames has then a lower bit rate. Cellular network and VoIP system use compression codecs such as G.711, G.729, GSM-EFR, and AMR. Our previous research results have shown that about one quarter of compressed speech frames are hardly relevant even if they contain active speech. This is due to the fact that the loss concealment algorithms at the receiver work so well that the losses of many frames are hardly perceptible. Ideally, these unimportant frames should not be transmitted in the first place because they do not contribute to the final transmission quality. Then, existing voice transmission can be optimized because lesser number of speech frames need to be transmitted and the efficiency of the transmission system is improved.

The research project aims to develop algorithms that can determine the relevance of speech frame. One has to distinguish between off-line algorithms, which might have high complexity and high algorithm delays on the one side and real-time algorithms that can be used at real time in mobile or Internet phones. We developed off-line algorithms prior this project. Thus, we now invent and evaluate real-time algorithms.

If importance of speech frames would be known, it increases the efficiency of speech transmission significantly. For example, mobile phones would transmit fewer speech frames and the energy can be saved. Also, the bandwidth requirement of VoIP applications can be reduced – especially for the use during congestion periods.

Publications

  1. Sofiene Jelassi, Habib Youssef, Christian Hoene, and Guy Pujolle. Single-ended parametric voicing-aware models for live assessment of packetized voip conversations. Telecommunication Systems, pages 1-18, 2010. 10.1007/s11235-010-9350-y. [ .pdf ]
  2. Sofiene Jelassi, Habib Ben Youssef, Christian Hoene, and Guy Pujolle. Voicing-aware parametric speech quality models over VoIP networks. In Global Information Infrastructure Symposium (IEEE GIIS 2009), Hammamet, Tunisia, June 2009. [ .pdf ]
  3. Christian Hoene. Next generation Wi-Fi VoIP phones. talk on the IQPC Conference VoWLAN, July 2006. Berlin, Germany. [ http ]
  4. Christian Hoene, Ian Marsh, Günter Schäfer, and Adam Wolisz. Error propagation after concealing a lost speech frame. In Proceedings of Multicomm 2006, Istanbul, Turkey, June 2006. [ .pdf ]
  5. Christian Hoene. Next generation Wi-Fi voip phones: How to reduce the packet rate of speech codecs? talk on the ITG Fachtagung VoWLAN, May 2006. Würzburg, Germany. [ .pdf ]
  6. Christian Hoene and Till Wimmer. Classifying VoIP μ-law packets in real-time. In ITG-Fachtagung Sprachkommunikation 2006, Kiel, Germany, April 2006. [ .pdf ] [7] C. Mahlo, Christian Hoene, A. Rostami, and Adam Wolisz. Adaptive coding and packet rates for TCP-friendly VoIP flows. In Proc. of 3rd International Symposium on Telecommunications (IST2005), Shiraz, Iran, September 2005. [ http | .pdf ]