Institute of Information Technology
- 1:
About the Institute. - 2: Staff.
- 2.1:
Information Transmission. - 2.2: Dialogue Systems.
- 2.3: Alumni.
- 2.1:
- 3: Teaching.
- 4: Information Transmission.
- 5: Dialogue Systems.
- 6: Intranet.
- 7: Links.
Reconstruction of Speech Signals
Author: Bernd Iser
Status: completed
Description:
The subject of this thesis is the development and study of several methods for the reconstruction of speech signals. Reconstruction in this context means the extraction of speech model based parameters from a speech signal that is distorted or band limited in some way and the following synthesis of an artificially supplemented speech signal. The methods that are to be compared in terms of performance include existing techniques as well as enhanced versions that are being developed.
The performance of these methods is primarily be measured regarding the subjective impression of listeners. Other measures include recognition tests and several spectral distortion measures. For the training as well as for the verification large speech corpora are used (e.g. TIMIT). The most promising method is implemented in a real-time environment using appropriate hardware. As a first approach the real-time implementation could be designed for the specific case of speech signals that have been subject to band limitation due to transmission over a telephone network.
The goal of this thesis is to achieve quality enhancement of clean speech signals as well as in presence of noise. This assessment depends on whether the enhanced signal is subject to further processing and the respective demands or emitted directly. In addition it is analyzed in what extent the obtained algorithm is able to perform as a noise reduction system under certain conditions.
PhD_Description.pdf