Databases Reference
In-Depth Information
get a new spectral envelope and hence a new bit allocation. The spectral envelope is coded
differentially. The first exponent is sent as is. The difference between exponents is encoded
using one of five values
. Three differences are encoded using a 7-bit word. Note
that three differences can take on 125 different combinations. Therefore, using 7 bits, which
can represent 128 different values, is highly efficient.
{
0
,
±
1
,
±
2
}
The D25 and D45 Methods
If the audio is not stationary, the spectral envelope is sent more often. To keep the bit rate
down, the Dolby AC-3 algorithm uses one of two strategies. In the D25 strategy, which is used
for moderate spectral activity, every other coefficient is encoded. In the D45 strategy, used
during transients, every fourth coefficient is encoded. These strategies make use of the fact
that during a transient the fine structure of the spectral envelope is not that important, allowing
for a more crude representation.
17.6 Other Standards
We have described a number of audio compression approaches that make use of the limitations
of human audio perception. These are by nomeans the only ones. Competitors toDolbyDigital
include Digital Theater Systems (DTS) and Sony Dynamic Digital Sound (SDDS). Both of
these proprietary schemes use psychoacoustic modeling. The Adaptive TRansform Acoustic
Coding (ATRAC) algorithm [
223
] was developed for the MiniDisc by Sony in the early 1990s,
followed by enhancements inATRAC3 andATRAC3plus. As with the other schemes described
in this chapter, the ATRAC approach uses MDCT for frequency decomposition, though the
audio signal is first decomposed into three bands using a two-stage decomposition. As in the
case of the other schemes, the ATRAC algorithm recommends the use of the limitations of
human audio perception in order to discard information that is not perceptible.
Another algorithm that also uses MDCT and a psychoacoustic model is the open source
encoder Vorbis. The Vorbis algorithm also uses vector quantization and Huffman coding to
reduce the bit rate.
17.7 Summary
The audio coding algorithms described in this chapter take, in some sense, the opposite tack
from the speech coding algorithms described previously. Instead of focusing on the source
of information, as is the case with a speech coding algorithm, the focus in an audio coding
algorithm is on the sink, or user, of the information. By identifying the components of the
source signal that are not perceptible, the algorithms reduce the amount of data that needs to
be transmitted.
