High quality low bit rate transform sub-band coding of speech

Describes a hybrid speech coding system [vector transform coder (VTC)] based on three of the most efficient coding techniques available. These are transform coding, subband coding and vector quantisation. Transform coding and sub-band coding are both frequency-domain techniques. Vector quantisation is a block coding method and the major motivation for using it is a fundamental result of Shannon´s rate-distortion theory which indicates that better performance is always achieved by coding data in blocks (vectors) rather than as scalars. This system is capable of producing very good quality coded speech at 16 and 9.6 kbits/s. In recent work comb filtering and artificial neural networks (ANN) methods have been employed in order to improve the efficiency of this system at 4.8 kbits/s. New methods have been developed for the design of vector quantiser codebooks as well as for the continuous updating of these quantisers. Due to the adaptive nature of these methods, considerable improvement has been achieved in the quality of the coded speech and a total increase of about 0.9 dB has been obtained in the signal to noise ratio at 4.8 kbits/s. The coder has recently been implemented in real-time by using an AT&T DSP32C digital signal processor chip