Audio subband coding with improved representation of
transient signal segments

by J. Kliewer and A. Mertins

Proc. European Signal Processing Conf. (EUSIPCO '98), pages 2345-2348,
Rhodes, Greece, September 1998


In this paper, we present a subband audio coding scheme with an
attack-sensitive framing of the input audio signal, where the frame
boundaries closely match both ends of the transient.  Since each
transient frame is processed as a symmetrically extended
finite-length signal with a support-preservative MDFT filter bank,
pre-echos can be almost completely avoided.  Furthermore, the
bit-allocation, which is calculated on a frame-by-frame basis, can
be determined more accurately, because the calculation of the
masking thresholds is carried out on signal segments with almost
"stationary" energy distribution.


The following audio samples are outputs of the experimental codec described in
the above paper using a target source rate of 75 kBit/s (per channel). The original
audio samples are taken from the EBU SQAM (Sound Quality Assessment Material) CD
and are in single channel format.
Original castagnet  excerpt: casta_orig.wav  (750 kB)
Reconstructed signal at 75kBit/s: casta_demo.wav (750 kB)

Original triangle excerpt: tri_orig.wav (400 kB)
Reconstructed signal at 75kBit/s: tri_demo.wav  (400 kB)


(Download  postscript version of full paper)