EFFICIENT VERY-LOW-RATE SPEECH TRANSMISSI ON USING INTERBLOCK CODING

The article presents a new data compression scheme for very-low-rate s peech coding at 1000 b/s and below. The method which is called matrix product quantization (MPQ) encodes matrices containing successive LPC parameter vectors by a product formulation of two matrices which descr ibe the average parameter vector and the temporal parameter contour. M PQ can be applied in fixed rate as well as in variable rate coding sys tems. At a fixed bit rate MPQ yields a very high coding efficiency at relatively low coding delay and medium complexity. Applied in a variab le rate segment vocoder MPQ achieves even higher compression gains at the cost of increased coding delay and complexity. In general, it is s hown that interblock coding, which is made by MPQ, considerably outper forms memoryless coding. In a particular comparison to interpolative c oding methods which is often used for very-low-rate speech coding, MPQ is clearly superior since only MPQ is able to exactly model the tempo ral parameter contour.