In this paper, we present a frame based approach for transient detection and encoding of audio signals. The transient detection procedure, as presented here, uses linear prediction within a signal frame followed by an envelope estimation to build an adaptive threshold. Detected transients will automatically be separated and the gaps left by the removed transient are filled with samples from forward and backward extrapolation. To encode detected transients, dyadic approximation approaches are discussed. Results of the application to different audio signals are also presented.