|
Identifier |
uch.csd.msc//2002agiomyrgiannakis |
Title |
Επέκταση Φάσματος Ομιλίας |
Alternative Title |
Speech Spectrum Expansion |
Creator |
Agiomyrgiannakis, Ioannis
|
Abstract |
Speech signal has spectral content that extends up to 20 KHz. Since current analogue telephone lines carry the lowest 0.3-3.7 KHz of the speech signal, and mobile telephony standard speech codecs encode only the 0-4 KHz signal, a significant portion of the speech signal is not included. This leads to a degradation of perceived speech quality. The purpose of this research is to use the statistical behaviour of the transmitted signal to estimate the untransmitted signal, and restore the wideband 0-8 KHz speech signal, perceptually better than the 0-4 KHz signal, and as close as possible to the original 0-8 KHz. For this reason the lost 4-8 KHz of the speech signal were modelled with two parameters; an LSF description of the 4-8 KHz spectral envelope and an expression of its energy. This model produced speech of high quality. The spectral envelope estimation was done with several estimators. A novel pitch based multiple estimator system is introduced. The errors of the spectral envelope estimator are successfully corrected with post-processing. The spectral expansion system was trained with a big database. The energy estimation remains an open issue. The result of this research is a Speech Spectrum Expansion system that requires the transmission of only one parameter, the energy ratio between 0-4 KHz and 4-8 KHz. The SSE system produces reconstructed wideband speech of quality, as subjective tests indicate. The SSE system can improve a speech codec of 0-4 KHz with minimal bandwidth requirements.
|
Issue date |
2002-03-01 |
Date available |
2002-04-12 |
Collection
|
School/Department--School of Sciences and Engineering--Department of Computer Science--Post-graduate theses
|
|
Type of Work--Post-graduate theses
|
Views |
436 |