Using learning vector quantizers for network bandwidth optimization in the QCELP speech coder

摘要

We attempt to show how average bandwidth usage during network transmissions can be reduced using the QCELP speech coder in conjunction with learning vector quantizers (LVQ). We identify three types of noise which can occur during the process of transmission. We then identify various techniques, some natural and some mathematical, to estimate the speech content of the signal. We then use LVQ to construct decision boundaries. Our simulation results show significant reductions in average bandwidth usage without large degradation in speech quality.