A wavelet packet algorithm for 1D data with no block end effects
Ben Leslie, Mark B. Sandler · 2003
New algorithms are presented for carrying out the wavelet packet analysis and synthesis of a one-dimensional signal which has been segmented into blocks. This is achieved on each block by using samples and transform coefficients from the previous block to carry out the filtering operations in the analysis and synthesis processes. These techniques avoid reconstruction artifacts at block boundaries. The performance is compared in an audio codec with an alternative which treats each block in isolation as a finite-length signal. The SSNR for the new algorithm is shown to average 2.5 dB better than the alternative over a range of sound samples, and is perceived to be superior in informal listening tests.