Signal segmentation for live speech network assessment

A digital adaptive filter (DAF) configuration is presented that models speech in the short term and allows accurate characterisation of echo paths. With the development of telecommunications networks it is imperative that network operators have the facility to continually assess the quality of service they are providing their customers. This is particularly important in international circuits as the near end network operator invariably has no control over the far end network and therefore requires an accurate estimate of far end network parameters. It has previously been shown that echo path characterisation can yield important information in estimating call quality degradations. Simulations show that the proposed configuration results in significant improvements in convergence rate, exceeds adaptive linear predictive techniques and compares well with noise driven adaptation while incurring minimal hardware overheads.<
>