Information

Abstract

Speech transmission is the central service in many telecommunication infrastructures. The encoding of many channels according to modern standards requires a fair amount of processing capacity. With the recent GPU product lines, powerful platforms have become available as supplement to desktop CPUs. This thesis tries to leverage these developments and examines the possibilities of general purpose GPU employment in the context of speech coding. The speech codec used in the TETRA mobile radio system is implemented using the CUDA programming model. The main question is, how many channels can be encoded in real time on current GPUs. Results show that through careful implementation and with some effort, a substial number of channels can be processed. It seems however that modern multicore CPUs are much better qualified for the task. The presented optimizations are far from complete and further research directions are suggested.

Additional Files and Images

Weblinks

No further information available.

BibTeX

@mastersthesis{Goldmann_2011_GPU,
  title =      "Towards GPU Speech Coding",
  author =     "Axel Goldmann",
  year =       "2011",
  abstract =   "Speech transmission is the central service in many
               telecommunication infrastructures. The encoding of many
               channels according to modern standards requires a fair
               amount of processing capacity. With the recent GPU product
               lines, powerful platforms have become available as
               supplement to desktop CPUs. This thesis tries to leverage
               these developments and examines the possibilities of general
               purpose GPU employment in the context of speech coding. The
               speech codec used in the TETRA mobile radio system is
               implemented using the CUDA programming model. The main
               question is, how many channels can be encoded in real time
               on current GPUs. Results show that through careful
               implementation and with some effort, a substial number of
               channels can be processed. It seems however that modern
               multicore CPUs are much better qualified for the task. The
               presented optimizations are far from complete and further
               research directions are suggested.",
  month =      may,
  address =    "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
  school =     "Institute of Computer Graphics and Algorithms, Vienna
               University of Technology ",
  URL =        "https://www.cg.tuwien.ac.at/research/publications/2011/Goldmann_2011_GPU/",
}