Kolloquium - Details zum Vortrag

Promotions-Vortrag: High-Definition Telephony over Heterogeneous Networks

Dipl.-Ing. Bernd Geiser
18. April 2012
14:00 Uhr
Hörsaal 4G IKS

As of today, (fixed and mobile) telephone connections are still restricted to audio frequencies below 4 kHz, leading to the familiar sound character of “telephone speech.” Meanwhile several coding standards for “High-Definition” Telephony HD voice (7 kHz) are available which offer a significantly better audio quality and speech intelligibility. However, the required costly and time-consuming modifications of the existing network equipment turned out to be a major obstacle for the introduction. The transition from today´s plain old telephony to the future HD voice networks will take a very long time period.
To account for this situation, in this thesis, concepts, methods and algorithms are investigated, evaluated and compared that facilitate a major audio quality upgrade of existing speech communication systems while maintaining backwards compatibility with the installed infrastructure.
The thesis makes major contributions to the following three principal scenarios:
Bandwidth Extension for Embedded Speech and Audio Coding
Two new bandwidth extension (BWE) algorithms are presented which have been developed in the context of recent ITU-T standardization projects for embedded speech and audio coding:
Artificial Bandwidth Extension without Auxiliary Information
Additional audio frequen­cies can be estimated from the received, band-limited signal alone. A consistent quality improvement over band-limited speech signals is obtained, but the quality does not reach the level of the embedded codec.
Bandwidth Extension with Steganographic Parameter Transmission
The availability of (even a small amount of) additional information can dra­mati­cally improve the quality of state-of-the-art speech codecs. For the common case that a modification of the digital bitstream format is not allowed, a new solution is proposed: Data hiding techniques are used to deliver the BWE information to the receiving terminal without altering the standard bitstream format. The inaudibility of the hidden information is ensured by a joint source encoding and data hiding procedure. As a practically relevant application, this concept is applied to ACELP (Algebraic Code Excited Linear Prediction) codecs as used in GSM/UMTS mobile telephony. The key advantage of the proposed solution is its full backwards compatibility with the standard narrowband codecs, i.e., the existing network infrastructure can be kept and used without any modifications.

zurück