Importance

Loading Importance Indicators...

Abstract

Digital samples of a voice signal are greatly compressed by filtering the samples to separate the information contained in them into several different frequency bands. In each of a succession of time intervals, the dominant filter output value is identified and converted to floating point form including an exponent and a generally truncated mantissa. All other filter output values for that time interval are also converted to floating point form having the same exponent. The mantissas are then further compressed by grouping values associated with each filter and again converting the mantissas in each group to floating point form having a common local exponent and submantissas. The Walsh-Haddamard transfer is also employed.

Loading the Abstract Image...

First Claim