Abstract
Future teleconferencing systems will use wide-band speech and will include acoustic echo control devices to achieve a sufficient quality. The classical echo cancellation algorithms involve an important amount of computations, due to the large adaptive filters identifying the impulse response of the echo path. The sub-band processing approach appears to be a reasonable way to reduce this computational burden, and it affords several other advantages. Two variants of sub-band echo cancellers havebeen implemented and tested: one is based on the two-band splitting of the speech spectrum by a QMF (Quadrature Mirror Filters) pair; the other uses pseudo-QMF banks to achieve the best reduction of the global computations. Performances obtained with both systems are reported. Main degradations due to critical sub-sampling in the pseudo-QMF banks are pointed out and illustrated. Finally, some solutions offering compromises between complexity and performances are proposed.

This publication has 5 references indexed in Scilit: