Optimisation techniques for low bit rate speech coding

Abstract

This thesis extends the background theory of speech and major speech coding schemes used in existing networks to an implementation of GSM full-rate speech compression on a RISC DSP and a multirate application for speech coding. Speech coding is the field concerned with obtaining compact digital representations of speech signals for the purpose of efficient transmission. In this thesis, the background of speech compression, characteristics of speech signals and the DSP algorithms used have been examined. The current speech coding schemes and requirements have been studied. The Global System for Mobile communication (GSM) is a digital mobile radio system which is extensively used throughout Europe, and also in many other parts of the world. The algorithm is standardised by the European Telecommunications Standardisation histitute (ETSI). The full-rate and half-rate speech compression of GSM have been analysed. A real time implementation of the full-rate algorithm has been carried out on a RISC processor GEPARD by Austria Mikro Systeme International (AMS). The GEPARD code has been tested with all of the test sequences provided by ETSI and the results are bit-exact. The transcoding delay is lower than the ETSI requirement. A comparison of the half-rate and full-rate compression algorithms is discussed. Both algorithms offer near toll speech quality comparable or better than analogue cellular networks. The half-rate compression requires more computationally intensive operations and therefore a more powerful processor will be needed due to the complexity of the code. Hence the cost of the implementation of half-rate codec will be considerably higher than full-rate. A description of multirate signal processing and its application on speech (SBC) and speech/audio (MPEG) has been given. An investigation into the possibility of combining multirate filtering and GSM fill-rate speech algorithm. The results showed that multirate signal processing cannot be directly applied GSM full-rate speech compression since this method requires more processing power, causing longer coding delay but did not appreciably improve the bit rate. In order to achieve a lower bit rate, the GSM full-rate mathematical algorithm can be used instead of the standardised ETSI recommendation. Some changes including the number of quantisation bits has to be made before the application of multirate signal processing and a new standard will be required.