Novel Digital Speech Compression Techniques
Project Award Date: 0000-00-00
The objective of this project is to develop novel medium bit rate speech coding algorithms which are compatible with existing low bit rate coders, and are robust to harsh transmission environments. This project addresses issues associated with the compression of digital speech data for a variety of applications related to the storage and transmission of speech. These applications include storage for intelligence analysis, storage for voice messaging systems, and congestion control in communications systems. The techniques developed here can best be described as "enabling technologies", in that they provide the basis for a substantial expansion of system capabilities, flexibility, and robustness. In particular, the methods discussed here allow speech to be coded and compressed in multiple stages, without interaction with the initial data source. This capability implies that data can initially be received and stored at high bit rates in order to provide high quality, and can later be efficiently converted to lower bit rates with minimal loss of quality. Preliminary work on this multi-stage approach suggests that quality comparable to that of single step coders can be attained. This project extends that work by using more advanced interpolation techniques, and by determining the proper parameters for best performance. In order to determine the performance of these algorithms in actual service, a voice mail application system suitable for wide-spread has been created. This application has text and window based user interfaces, and is portable to a variety of workstation architectures. As part of the development of the voice mail application, protocols for communication between computers running voice mail systems have been developed.
Primary Sponsor(s): Rome Laboratories (U.S. Air Force)