Abstract
The past decade has seen a very fast growth of the telecommunications industry. Mobile telephony has evolved from a specialist application to being commonplace and affordable, and is now a mass-market industry. A similar evolution is expected from multimedia communications, where voice, video and data are all to be integrated into one device. These services require a large amount of bandwidth, which is a relatively cheap and expandable resource in wire based fixed networks. However it is at a premium in satellite or cellular radio systems. In order to cope with the growing demand and the increasing number of subscribers, it is necessary to make optimal use of the bandwidth available. This implies using efficient source coding technologies, including speech compression algorithms. Many of the recent cellular radio communication systems have used speech coders based upon the Code Excited Linear Prediction (CELP) model. These provide high speech quality at bit rates of 8 kb/s and above, however this reduces significantly when the bit rate is lowered. Vocoders on the other hand have been used for very low bit rate applications, where they provide low quality speech. This usually restricts their use to specialised applications such as private radio or military use. The aim of the research presented here is to improve the speech quality produced by low bit rate vocoders, ideally bringing it close to that of higher bit rates CELP coders while retaining a low bit rate. In order to achieve this it has been necessary to introduce new and refined parameter estimation and quantisation techniques, which were integrated in an improved vocoder model. The resulting coder was then adapted to a range of low and very low bit rate applications, and submitted as candidates to three major standardisation efforts.