Listening Tests for Extrapolation Algorithm

During Internet audio streaming, packet loss results in missing audio data.  This gap of audio data could be estimated by interpolation only if the gap is small and valid data (packets) follow the gap.  When there is no valid audio data (packets have not arrived yet) following the gap, the gap tends to be filled with silence or noise.  The gap could also be filled with an extrapolation.  The following audio clips have gaps of varying lengths.  These gaps are filled with either silence, noise, or the output of the extrapolation system.  The extrapolation system is Minimum Weighted Norm Extrapolation with Frequency-Domain Blocking.  The input to the extrapolation system is the preceding audio data of the same length as the gap.   Overlap percentage is set at 6.25%, and block length is set at one-fourth the number of samples in the gap.

The following table contains audio clips with varying gaps as specified by the left column.  The gaps are filled by either silence, white noise, or extrapolation system output as specified by the header on each column.  The three examples (classical, jazz, and rock) repeat each experiment.

Continue

Back to Table of Contents