Note: Descriptions are shown in the official language in which they were submitted.
CA 0,2664466 2009-03-24
Encoding an information signal
Description
The present invention relates to information signal
encoding such as audio encoding, and, in that context, in
particular to SBR (spectral band replication) encoding.
In applications having a very small bit rate available, it
is known, in the context of encoding audio signals, to use
an SBR technique for encoding. Only the low-frequency
portion is encoded fully, i.e. at an adequate temporal and
spectral resolution. For the high-frequency portion, only
the spectral envelope, or the envelope of the spectral
temporal curve of the audio signal, is detected and
encoded. On the decoder side, the low-frequency portion is
retrieved from the encoded signal and is subsequently used
to reconstruct, or "replicate", the high-frequency portion
therefrom. However, to adapt the energy of the high-
frequency portion, which has thus been preliminarily
reconstructed, to the actual energy within the high-
frequency portion of the original audio signal, the
spectral envelope transmitted is used, on the decoder side,
for spectral weighting of the high-frequency portion
reconstructed preliminarily.
For the above effort to be worthwhile, it is important, of
course, that the number of bits used for transmitting the
spectral envelopes be as small as possible. It is therefore
desirable for the temporal grid within which the spectral
envelope is encoded to be as coarse as possible. On the
other hand, however, too coarse a grid leads to audible
artefacts, which is notable, in particular, with
transients, i.e. at locations where the high-frequency
portions will predominate rather than, as usual, the low-
frequency portions, or where there is at least a rapid
increase in the amplitude of the high-frequency portions.
CA 02664466 2009-03-24
- 2
In audio signals, such transients correspond, for example,
to the beginnings of a note, such as actuation of a piano
string or the like. If the grid is too coarse over the time
period of a transient, this may lead to audible artefacts
in the decoder-side reconstruction of the entire audio
signal. For, as one knows, on the decoder side, the high-
frequency signal is reconstructed from the low-frequency
portion in that, within the grid area, the spectral energy
of the decoded low-frequency portion is normalized and then
adapted to the spectral envelope transmitted by means of
weighting. In other words, spectral weighting is simply
performed within the grid area so as to reproduce the high-
frequency portion from the low-frequency portion. However,
if the grid area around the transient is too large, a lot
of energy will be located, within this grid area, in
addition to the energy of the transient, in the background
and/or chord portion in the low-frequency portion which is
used for reproducing the high-frequency portion. Said low-
frequency portion is co-amplified by the weighting factor,
even though this does not result in a good estimation of
the high-frequency portion. Across the entire grid area,
this will lead to an audible artefact which, in addition,
will set in even before the actual transient. This problem
may also be referred to as "pre-echo".
The problem could be solved when the grid area around the
transient is fine enough so that the transient/background
ratio of the part of the low-frequency portion within this
grid area is improved. Small grid areas or small grid
boundary distances, however, are obstacles on the way to
the above-outlined desire for a low bit consumption for
encoding the spectral envelopes.
In the ISO/IEC 14496-3 standard - simply referred to as
"the standard" below - an SBR encoding is described in the
context of the AAC encoder. The AAC encoder encodes the
low-frequency portion in a frame-by-frame manner. For each
such SBR frame, the above-specified time and frequency
CA 02664466 2009-03-24
- 3 -
resolution is defined at which the spectral envelope of the
high-frequency portion is encoded in this frame. To address
the problem that transients may also fall on SBR frame
boundaries, the standard allows that the temporal grid may
temporarily be defined such that the grid boundaries do not
necessarily coincide with the frame boundaries. Rather, in
this standard, the encoder transmits, per frame, a syntax
element bs_frame_class to the decoder, said syntax element
indicating per frame whether the temporal grid of the
spectral envelope gridding for the respective frame is
defined precisely between the two frame boundaries or
between boundaries which are offset from the frame
boundaries, specifically at the front and/or at the back.
Overall, there are four different classes of SBR frames,
i.e. FIXFIX, FIXVAR, VARFIX and VARVAR. The syntax used by
the encoder in the standard to define the grid per SBR
frame is depicted in a pseudo code representation in Fig.
12. In particular, in the representation of Fig. 12, those
syntax elements which are actually encoded and/or
transmitted by the encoder are printed in bold type in Fig.
12, the number of the bits used for transmission and/or
encoding being indicated in the second column from the
right in the respective row. As may be seen, the syntax
element bs_frame_class which has just been mentioned is
initially transmitted for each SBR frame. As a function
thereof, further syntax elements will follow which, as will
be illustrated, define the temporal resolution and/or
gridding. If, for example, the 2-bits syntax element
bs_frame_class indicates that the SBR frame in question is
a FIXFIX SBR frame, the syntax element tmp which defines
the number of grid areas in this SBR frame, and/or which
defines the number of envelopes, as 2tmP will be transmitted
as the second syntax element. The syntax element
bs_amp_res, which is used for the quantization step size
for encoding the spectral envelope in the current SBR
frame, is automatically adjusted as a function of
bs_num_env, and is not encoded or transmitted. Finally, for
a FIXFIX frame, a bit is transmitted for determining the
CA 02664466 2009-03-24
- 4 -
2
frequency resolution of the grid bs_freq_res. FIXFIX frames
are defined precisely for one frame, i.e. the grid
boundaries coincide with the frame boundaries as defined by
the AAC encoder.
This is different for the other three classes. For FIXVAR,
VARFIX and VARVAR frames, syntax elements bs_var_bord_l
and/or bs_bar_bod_O are transmitted to indicate the number
of time slots, i.e. the time units wherein the filter bank
for spectral decomposition of the audio signal operates, by
which are offset relative to the normal frame boundaries.
As a function thereof, syntax elements bs_num_rel_l and an
associated tmp and/or bs_num_rel_O and an associated tmp
are also transmitted so as to define a number of grid
areas, or envelopes, and the size thereof from the offset
frame boundary. Finally, a syntax element bs_pointer is
also transmitted within the variable SBR frames, said
syntax element pointing to one of the defined envelopes and
serving to define one or two noise envelopes for
deteLmining the noise portion within the frame as a
function of the spectral envelope gridding, which, however,
shall not be explained in detail below in order to simplify
the representation. Finally, the respective frequency
resolution is determined, namely by a respective one-bit
syntax element bs_freq_res per envelope, for all grid areas
and/or envelopes in the respective variable frames.
Fig. 13a represents, by way of example, a FIXFIX frame
wherein the syntax element tmp is 1, so that the number of
envelopes is bs_num_env 21=2. In Fig. 13a it shall be
assumed that the time axis extends from the left to the
right in a horizontal manner. An SBR frame, i.e. one of the
frames in which the AAC encoder encodes the low-frequency
portion, is indicated by reference numerals 902 in Fig.
13a. As can be seen, the SBR frame 902 has a length of 16
QMF slots, the QMF slots being, as has been mentioned, the
time slots in which units the analysis filter bank
operates, the QMF slots being indicated by box 904 in Fig.
CA 02664466 2009-03-24
- 5
13a. In FIXFIX frames, the envelopes, or grid areas, 906a
and 906b, i.e. two in number here, have the same length
within the SBR frames 902, so that a time grid and/or
envelope boundary 908 is defined precisely in the center of
the SBR frame 902. In this manner the exemplary FIXFIX
frame of Fig. 13a defines that a spectral distribution for
the grid area, or the envelope, 906a, and a further one for
envelope 906, is temporally determined from the spectral
values of the analysis filter bank. The envelopes, or grid
areas, 906a and 906b thus specify the grid in which the
spectral envelope is encoded and/or transmitted.
By comparison, Fig. 13b shows a VARVAR frame. SBR frame 902
and associated QMF slots 904 are indicated again. For this
SBR frame, however, syntax elements bs_var_bord_O and/or
bs_var_bord_l have defined that the envelopes 906a', 906b'
and 906c' associated therewith are not to start at the SBR
frame start 902a and/or to end at the SBR frame end 902b.
Rather, one may see from Fig. 13b that the previous SBR
frame (not to be seen in Fig. 13b) has already been
extended two QMF slots beyond the SBR frame start 902a of
the current SBR frame, so that the last envelope 910 of the
preceding SBR frame still extends into the current SBR
frame 902. The last envelope 906c' of the current frame
also extends beyond the SBR frame end of the current SBR
frame 902, namely, by way of example, also by two QMF slots
here. In addition, one can also see here, by way of
example, that the syntax elements of the VARVAR frame
bs_num_rel_O and bs_num_rel_l are adjusted to 1,
respectively, with the additional information that the
envelopes thus defined have a length of four QMF slots at
the start and at the end of the SBR frame 902, i.e. 906a'
and 906b' in accordance with tmp = 1, so as to extend from
the frame boundaries into the SBR frame 902 by this number
of slots. The remaining space of the SBR frame 902 will
then be occupied by the remaining envelope, in this case
the third envelope 906b'.
CA 02664466 2009-03-24
- 6 -
By having T in one of the QMF slots 904, Fig. 13b
indicates, by way of example, the reason why a VARVAR frame
has been defined here, namely because the transient
position T is located close to the SBR frame end 902b, and
because there probably was a transient (not to be seen)
also in the SBR frame preceding the current one.
The standardized version in accordance with ISO/ICE 14496-3
thus involves overlapping of two successive SBR frames.
This enables setting the envelope boundaries in a variable
manner, irrespective of the actual SBR frame boundaries in
accordance with the waveform. Transients may thus be
enveloped by envelopes of their own, and their energy may
be cut off from the remaining signal. However, an overlap
also involves an additional system delay, as was
illustrated above. In particular, four frame classes are
used for signaling in the standard. In the FIXFIX class,
the boundaries of the SBR envelopes coincide with the
boundaries of the core frame, as is shown in Fig. 13a. The
FIXFIX class is used when no transient is present in this
frame. The number of envelopes specifies their equidistant
distribution within the frame. The FIXVAR class is provided
when there is a transient in the current frame. Here, the
respective set of envelopes thus starts at the SBR frame
boundary and ends, in a variable manner, in the SBR
transmission area. The VARFIX class is provided for the
event that a transient is not located in the current, but
in the previous frame. The sequence of envelopes from the
last frame here is continued by a new set of envelopes
which ends at the SBR frame boundary. The VARVAR class is
provided for the case that a transient is present both in
the last frame and in the current frame. Here, a variable
sequence of envelopes is continued by a further variable
sequence. As has been described above, the boundaries of
the variable envelopes are transmitted in relation to one
another.
CA 02664466 2012-07-11
- 7 -
Even though the number of QMF slots by which the boundaries may be
offset relative to the fixed frame boundaries by means of the
syntax elements bs_var_bord 0 and bs var bord 1, this possibility
results in a delay on the decoder side due to the occurrence of
envelopes which extend beyond SBR frame boundaries and thus
necessitate the formation and/or averaging of spectral signal
energies across SBR frame boundaries. However, this time delay is
not tolerable in some applications, such as in applications in the
field of telephony or other live applications which rely on the
time delay caused by the encoding and decoding to be small. Even
though the occurrence of pre-echoes is thus prevented, the
solution is not suitable for applications requiring a short delay
time. In addition, the number of bits required for transmitting
the SBR frames in the above-described standard is relatively high.
It is an advantage of the present invention to provide an encoding
scheme which enables, with sufficient addressing of the transient
and/or pre-echo problem, shorter delay times at a moderate or even
lower bit rate, or, with sufficient addressing of the transient
and/or pre-echo problem, a reduced delay time at moderate bit-rate
losses.
A finding of the present invention is that the transient problem
may be sufficiently addressed, and for this purpose, a further
delay on the decoding side may be reduced, if a new SBR frame
class is employed wherein the frame boundaries are not offset,
i.e. the grid boundaries are still synchronized with the frame
boundaries, but wherein a transient position indication is
additionally used as a syntax element so as to be used, on the
encoder and/or decoder sides, within the frames of this new frame
CA 02664466 2009-03-24
- 8
class for determining the grid boundaries within these
frames.
In accordance with one embodiment of the present invention,
the transient position indication is used such that a
relatively short grid area, referred to as transient
envelope below, will be defined around the transient
position, whereas only one envelope will extend, in the
remaining part before and/or behind it, in the frame, from
the transient envelope to the start and/or the end of the
frame. The number of bits to be transmitted and/or to be
encoded for the new class of frames is thus also very
small. On the other hand, transients and/or pre-echo
problems associated therewith may be sufficiently
addressed. Variable SBR frames, such as FIXVAR, VARFIX and
VARVAR, will then no longer be required, so that delays for
compensating envelopes which extend beyond SBR frame
boundaries will no longer be necessary. In accordance with
an embodiment of the present invention, only two frame
classes thus will now be admissible, namely a FIXFIX class
and this class which has just been described and which will
be referred to as LD_TRAN class below.
In accordance with a further embodiment of the present
invention, it is not always the case that one or several
spectral envelopes and/or spectral energy values are
transmitted and/or inserted into the encoded information
signal for each grid area within the frames of the LD_TRAN
class. Specifically, this is not even done when the
transient envelope specified in its position within the
frame by the transient position indication is located close
to the frame boundary which is leading in terms of time, so
that the envelope of this LD_TRAN frame, said envelope
being located between the frame boundary which is leading
in terms of time and the transient envelope, will extend
only over a short time period, which is not justified from
the point of view of encoding efficiency, since, as one
knows, the brevity of this envelope is not due to a
CA 02664466 2009-03-24
- 9 -
transient, but rather to the accidental temporal proximity
of the frame boundary and the transient. In accordance with
this alternative embodiment, the spectral energy value(s)
and the respective frequency resolution of the previous
envelope are taken over, therefore, for this envelope
concerned, just like the noise portion, for example. Thus,
transmission may be omitted, which is why the compression
rate is increased. Conversely, losses in terms of
audibility are only small, since there is not transient
problem at this point. In addition, no delay will occur on
the decoder side, since utilization for high-frequency
reconstruction is directly possible for all envelopes
involved, i.e. envelopes from a previous frame, transient
envelope and intervening envelope.
In accordance with a further embodiment, the problems of an
unintentionally large amount of data in the occurrence of a
transient at the end of an LD_TRAN frame are addressed in
that an agreement is reached between the encoder and the
decoder as to how far the transient envelope which is
located at the trailing frame boundary of the current
LD_TRAN frame is to virtually project into the subsequent
frame. The decision is made, for example, by means of
accessing the tables in the encoder and the decoder alike.
In accordance with the agreement, the first envelope of the
subsequent frame, such as the single envelope of a FIXFIX
frame, is shortened so as to begin only at the end of the
virtual extended envelope. The encoder calculates the
spectral energy value(s) for the virtual envelope over the
entire time period of this virtual envelope, but transmits
the result, as it seems, only for the transient envelope,
possibly in a manner which is reduced as a function of the
ratio of the temporal portion of the virtual envelope in
the leading and trailing frames. On the decoder side, the
spectral energy value(s) of the transient envelope located
at the end are used both for high-frequency reconstruction
in this transient envelope and, separate therefrom, for
high-frequency reconstruction in the initial extension area
CA 02664466 2009-03-24
- 10 -
in the subsequent frames, in that one and/or several
spectral energy value(s) for this area are derived from
that, or those, of the transient envelope. "Oversampling"
of transients located at frame boundaries is thereby
avoided.
In accordance with a further aspect of the present
invention, a finding of the present invention is that the
transient problems described in the introduction to the
description may be sufficiently addressed, and a delay on
the decoder side may be reduced, if an envelope and/or grid
area division is indeed used, according to which envelopes
may indeed extend across frame boundaries so as to overlap
with two adjacent frames, but if these envelopes are again
subdivided by the decoder at the frame boundary, and the
high-frequency reconstruction is performed at the grid
which is subdivided in this manner and coincides with the
frame boundaries. For the partial grid areas, thus
obtained, of the overlap grid areas a spectral energy
value, or a plurality of spectral energy values, is/are
obtained, respectively, on the decoder side, from the one
or the plurality of spectral energy value(s) as have been
transmitted for the envelope extending across the frame
boundary.
In accordance with a further aspect of the present
invention, a finding of the present invention is that a
delay on the decoding side may be obtained by reducing the
frame size and/or the number of the samples contained
therein, and that the effect of the increased bit rate
associated therewith may be reduced if a new flag is
introduced, and/or a transient absence indication is
introduced, for frames having reconstruction modes
according to which the grid boundaries coincide with the
frame boundaries of these frames, such as FIXFIX frames,
and/or for the respective reconstruction mode.
Specifically, if there is no transient present in such a
shorter frame, and if no other transient is present in the
CA 02664466 2009-03-24
- 11 -
vicinity of the frame, so that the information signal is
stationary at this point, the transient absence indication
may be used not to introduce, for the first grid area of
such a frame, any value describing the spectral envelope
into the encoded infolmation signal, but to derive, or
obtain, same on the decoder side, rather from the value(s)
representing the spectral envelope, said values being
provided in the encoded information signal for the last
grid area and/or the last envelope of the temporally
preceding frame. In this manner, shortening of the frames
with a reduced effect on the bit rate is possible, which
shortening enables shorter delay time, on the one hand, and
enables the transient problems because of the smaller frame
units, on the other hand.
Preferred embodiments of the present invention will be
explained below in more detail with reference to the
accompanying figures, wherein:
Fig. 1 shows a block diagram of an encoder in accordance
with an embodiment of the present invention;
Fig. 2 shows a pseudo code for describing the syntax of
the syntax elements used by the encoder of Fig. 1
for defining the SBR frame grid division;
Fig. 3 shows a table which may be defined, on the
encoder and decoder sides, to obtain, from the
syntax element bs_transient_position in Fig. 2,
the information on the number of envelopes and/or
grid areas and the positions of the grid area
boundaries within an LD_TRAN frame;
Fig. 4a shows a schematic representation for illustrating
an LD_TRAN frame;
CA 02664466 2009-03-24
- 12 -
Fig. 4b shows a schematic representation for illustrating
the interplay of the analysis filter bank and the
envelope data calculator in Fig. 1;
Fig. 5 shows a block diagram of a decoder in accordance
with an embodiment of the present invention;
Fig. 6a shows a schematic representation for illustrating
an LD_TRAN frame with a transient envelope
located far toward the leading end for
illustrating the problems arising in this case;
Fig. 6b shows a schematic representation for illustrating
a case wherein a transient is located between two
frames, for illustrating the respective problems
with regard to the high encoding expenditure in
this case;
Fig. 7a shows a schematic representation for illustrating
an envelope encoding in accordance with an
embodiment for overcoming the problems of Fig.
6a;
Fig. 7a shows a schematic representation for illustrating
an envelope encoding in accordance with an
embodiment for overcoming the problems of Fig.
6b;
Fig. 8 shows a schematic representation for illustrating
an LD_TRAN frame with a transient position
TranPos = 1 in accordance with the table of Fig.
3;
Fig. 9 shows a table which may be defined, on the
encoder and decoder sides, to obtain, from the
syntax element bs_transient_position in Fig. 2,
the information on the number of envelopes and/or
grid areas and the positions of the grid area
CA 02664466 2009-03-24
- 13 -
boundary (boundaries) within an LD_TRAN frame as
well as the information on the data acceptance
from the previous frame in accordance with Fig.
7a and the data extension into the subsequent
frame in accordance with Fig. 7b;
Fig. 10 shows a schematic representation of a FIXVAR-
VARFIX sequence for illustrating an envelope
signaling with envelopes extending across frame
boundaries;
Fig. 11 shows a schematic representation of a decoding
which enables a shorter delay time despite
envelope signaling in accordance with Fig. 10, in
accordance with a further embodiment of the
present invention;
Fig. 12 shows a pseudo code of the syntax for SBR frame
envelope division in accordance with the ISO/IEC
14496-3 standard; and
Figs. 13a and 13b show schematic representations of a
FIXFIX and/or VARVAR frame.
Fig. 1 shows the architecture of an encoder in accordance
with an embodiment of the present invention. The encoder of
Fig. 1 is, by way of example, an audio encoder generally
indicated by reference numeral 100. It includes an input
102 for the audio signal to be encoded, and an output 104
for the encoded audio signal. It shall be assumed below
that the audio signal in input 102 is a sampled audio
signal, such as a PCM-encoded signal. However, the encoder
of Fig. 1 may also be implemented differently.
The encoder of Fig. 1 further includes a down-sampler 104
and an audio encoder 106 which are connected, in the order
mentioned, between the input 102 and a first input of a
formatter 108, the output of which, in turn, is connected
CA 02664466 2009-03-24
- 14 -
to the output 104 of the encoder 100. Due to the connection
of the portions 104 and 106, an encoding of the down-
sampled audio signal 102 results at the output of the audio
encoder 106, said encoding, in turn, corresponding to an
encoding of the low-frequency portion of the audio signal
102. The audio encoder 106 is an encoder which operates in
a frame-by-frame manner in the sense that the encoder
result present at the output of the audio encoder 106 can
only be decoded in units of these frames. By way of
example, it shall be assumed below that the audio encoder
106 is an encoder in conformity with AAC-LD in accordance
with the standard of ISO/IEC 14496-3.
An analysis filter bank 110, an envelope data calculator
112 as well as an envelope data encoder 114 are connected,
in the order mentioned, between the input 102 and a further
input of the formatter 108. In addition, the encoder 100
includes an SBR frame controller 116 which has a transient
detector 118 connected between its input and the input 102.
Outputs of the SBR frame controller 116 are connected both
to an input of the envelope data calculator 112 and to a
further input of the formatter 108.
Now that the architecture of the encoder of Fig. 1 has been
described above, its mode of operation will be described
below. As has already been mentioned, an encoded version of
the low-frequency portion of the audio signal 102 arrives
at the first input of formatter 108 in that the audio
encoder 106 encodes the down-sampled version of the audio
signal 102, wherein, e.g., only every other sample of the
original audio signal is forwarded. The analysis filter
bank 110 generates a spectral decomposition of the audio
signal 102 with a certain temporal resolution. It shall be
assumed, by way of example, that the analysis filter bank
110 is a QMF filter bank (QMF = quadrature mirror filter).
The analysis filter bank 110 generates M subband values per
QMF time slot, the QMF time slots each including 64 audio
samples, for example. To reduce the data rate, the envelope
CA 02664466 2009-03-24
=
- 15 -
data calculator 112 forms, from the spectral information of
the analysis filter bank 110 which has high temporal and
spectral resolutions, a representation of the spectral
envelope of audio signal 102 with a suitably lower
resolution, i.e. within a suitable time and frequency grid.
In this context, the time and frequency grid is set by the
SBR frame controller 116 per frame, i.e. per frame of the
frames as are defined by the audio encoder 106. Again, the
SBR frame controller 116 performs this control as a
function of detected and/or localized transients as are
detected and/or localized by the transient detector 118.
For detection transients and/or note commencement times,
the transient detector 118 performs a suitable statistical
analysis of the audio signal 102. The analysis may be
performed in the time domain or in the spectral domain. The
transient detector 118 may evaluate, for example, the
temporal envelope curve of the audio signal, such as the
evaluation of the increase in the temporal envelope curve.
As will be described in more detail below, the SBR frame
controller 116 associates each frame and/or SBR frame to
one of two possible SBR frame classes, namely either to the
FIXFIX class or to the LD_TRAN class. In particular, the
SBR frame controller 116 associates the FIXFIX class with
each frame which contains no transient, whereas the frame
controller associates the LD_TRAN class with each frame
having a transient located therein. The envelope data
calculator 112 sets the temporal grid in accordance with
the SBR frame classes as have been associated with the
frames by the SBR frame controller 116. Irrespective of the
precise association, all frame boundaries will always
coincide with grid boundaries. Only the grid boundaries
within the frames are influenced by the class association.
As will be explained below in more detail, the SBR frame
controller sets further syntax elements as a function of
the frame class associated, and outputs these to the
formatter 108. Even though not explicitly depicted in Fig.
1, the syntax elements may naturally also be subjected to
an encoding operation.
CA 02664466 2009-03-24
- 16 -
-
Thus, the envelope data calculator 112 outputs a
representation of the spectral envelopes in a resolution
which corresponds to the temporal and spectral grid
predefined by the SBR frame controller 116, namely by one
spectral value per grid area. These spectral values are
encoded by the envelope data encoder 114 and forwarded to
the formatter 108. The envelope data encoder 114 may
possibly also be omitted. The formatter 108 combines the
information received into the encoded audio data stream 104
and/or to the encoded audio signal, and outputs same at the
output 104.
The mode of operation of the encoder of Fig. 1 will be
described in a little more detail below using Figs. 2 to 4b
with regard to temporal grid division which is set by the
SBR frame controller 116 and used by the envelope data
calculator 112 to detelmine, from the analysis filter bank
output signal, the signal envelope in the predefined grid
division.
Fig. 2 initially shows, by means of a pseudo code, the
syntax elements by means of which the SBR frame controller
116 predefines the grid division which is to be used by the
envelope data calculator 112. Just like in the case of Fig.
12, those syntax elements which are actually forwarded from
the SBR frame controller 116 to the formatter 108 for
encoding and/or for transmission are depicted in bold print
in Fig. 2, the respective row in the column 202 indicating
the number of bits used for representing the respective
syntax element. As may be seen, a determination is
initially made, by the syntax element bs_frame_class, for
the SBR frame, whether the SBR frame is a FIXFIX frame or
an LD_TRAN frame. Depending on the determination (204),
different syntax elements are then transmitted. In the case
of the FIXFIX class (206), the syntax element
bs_num_env[ch] of the current SBR frame ch is initially set
to 2tm' by the 2-bit syntax element tmp (208). Depending on
the number bs_num_env[ch] the syntax element bs_amp_res is
CA 02664466 2009-03-24
- 17
left at a value of 1 which has been preset by default, or
is set to zero (210), the syntax element bs_amp_res
indicating the quantization accuracy with which the
spectrally enveloping values which are obtained by the
calculator 112 in the predefined gridding are forwarded to
the formatter 108 in a state in which they are encoded by
the encoder 114. The grid areas and/or envelopes predefined
in their numbers by bs_num_env[ch] are set - with regard to
their frequency resolution, which is to be used in same by
the envelope data calculator 112 to determine the spectral
envelope within them - by a common (211) syntax element
bs_freq_res[ch] which is forwarded (212) to the formatter
108 with a bit from the SBR frame controller 116.
The mode of operation of the envelope data calculator 112
is to be described again below with reference to Fig. 13a
when the SBR frame controller 116 specifies that the
current SBR frame 902 is a FIXFIXFIX frame. In this case,
the envelope data calculator 112 equally subdivides the
current frame 902, which consists - here by way of example
- of N = 16 analysis filter bank time slots 904, into grid
areas and/or envelopes 906a and 906b, so that here both
grid areas and/or both envelopes 906a, 906b have a length
of N/bs_num_inv[ch] time slots 904 and take up as many time
slots between the SBR frame boundaries 902a and 902b. In
other words, with FIXFIX frames, the envelope data
calculator 112 arranges the grid boundaries 908 uniformly
between the SBR frame boundaries 902a, 902b such that they
are equidistantly distributed within these SBR frames. As
has already been mentioned, the analysis filter bank 110
outputs subband spectral values per time slot 904. The
envelope data calculator 112 temporally combines the
subband values in an envelope-by-envelope manner and adds
their square sums in order to obtain the subband energies
in an envelope resolution. Depending on the syntax element
bs_freq_res[ch], the envelope data calculator 112 also
combines, in a spectral direction, several subbands to
reduce the frequency resolution. In this manner, the
CA 02664466 2009-03-24
- 18 -
envelope data calculator 112 outputs, per envelope 906a,
906b, a spectrally enveloping energy sampling at a
frequency resolution which depends on bs_freq_res[ch].
These values are then encoded by the encoder 114 with a
quantization which in turn depends on bs_amp_res.
So far, the preceding description related to the case where
the SBR frame controller 116 associated a specific frame
with the FIXFIX class, which is the case if there are no
transients in this frame, as was described above. The
following description, however, relates to the other class,
i.e. the LDN-TRAN class, which is associated with a frame
if it has a transient located in it, as is indicated by the
detector 118. Thus, if the syntax element bs_frame_class
indicates that this frame is an LDN-TRAN frame (214), the
SBR frame controller 116 will determine and transmit, with
four bits, a syntax element bs_transient_position so as to
indicate - in units of the time slots 904, for example
relative to the frame start 902a or, alternatively,
relative to the frame end 902b - the position of the
transient as has been localized by the transient detector
118 (216). At present, four bits are sufficient for this
purpose. An exemplary case is depicted in Fig. 4a. Fig. 4a,
in turn, shows the SBR frame 902 including the 16 time
slots 904. The sixth time slot 904 from the SBR frame start
902a has a transient T located therein, which would
correspond to bs_transient_position = 5 (the first time
slot is the time slot zero). As is indicated at 218 in Fig.
2, the subsequent syntax for setting the grid of an LD_TRAN
frame is dependent on bs_transient_position, which must be
taken into account, on the decoder side, in the parsing
performed by a respective demultiplexer. However, at 218,
the mode of operation of the envelope data calculator 112
upon obtaining the syntax element bs_transient_position
from the SBR frame controller 116 may be illustrated, which
is as follows. By means of the transient position
indication, the calculator 112 looks up
bs_transient_position in a table, an example of which is
CA 02664466 2009-03-24
- 19 -
shown in Fig. 3. As will be explained in more detail below
with reference to the table of Fig. 3, the calculator 112
will set, by means of the table, an envelope subdivision
within the SBR frame in such a manner that a short
transient envelope is arranged around transient position T,
whereas one or two envelopes 222a and 222b occupy the
remaining part of the SBR frame 902, namely the part from
the transient envelope 220 to the SBR frame start 902a,
and/or the part from the transient envelope 220 to the SBR
frame end 902b.
The table shown in Fig. 3 and used by the calculator 112
now includes five columns. The possible transient positions
which, in the present example, extend from zero to 15 have
been entered into the first column. The second column
indicates the number of envelopes and/or grid areas 220,
222a and/or 222b which result at the respective transient
position. As may be seen, the possible numbers are 2 or 3,
depending on whether the transient position is located
close to the SBR frame start or the SBR frame end 902a,
902b, only two envelopes being present in the latter case.
The third column indicates the position of the first
envelope boundary within the frame, i.e. the boundary of
the first two adjacent envelopes in units of time slots
904, specifically the position of the start of the second
envelope, the position = zero indicating the first time
slot in the SBR frame. The fourth column accordingly
indicates the position of the second envelope boundary,
i.e. the boundary between the second and third envelopes,
this indication naturally being defined only for those
transient positions for which three envelopes are provided.
Otherwise, the values entered are negligible in this
column, which is indicated by "-" in Fig. 3. As may be seen
by way of example in the table of Fig. 3, there is, for
example, only the transient envelope 220 and the subsequent
envelope 222b in the event that the transient position T is
located in one of the first two time slots 904 from the SBR
frame start 902a. It is not until the transient position is
CA 02664466 2009-03-24
- 20 -
located in the third time slot from the SBR frame start
902a that there are three envelopes 222a, 220, 222b,
envelope 222a including the first two time slots, transient
envelope 220 including the third and fourth time slots, and
envelope 222b including the remaining time slots, i.e. from
the fifth one onwards. The last column in the table of Fig.
3 indicates, for each transient position possibility, which
of the two or three envelopes corresponds to that which has
the transient and/or the transient position located
therein, this information obviously being redundant and
thus not necessarily having to be set forth in a table.
However, the information in the last column serves to
specify - in a manner which will be described in more
detail below - the boundary between two noise envelopes,
within which the calculator 112 determines a value which
indicates the magnitude of the noisy portion within these
noise envelopes. The manner in which the boundary between
these noise envelopes and/or grid areas is determined by
the calculator 112 is known on the decoder side, and is
performed in the same manner on the decoder side, just like
the table of Fig. 3 is also present on the decoder side,
namely for parsing and for grid division.
Referring back to Fig. 2, the calculator 112 may thus
determine the number of envelopes and/or grid areas in the
LD_TRAN frames from Table 2 of Fig. 3, the SBR frame
controller (116) indicating, for each one of these two or
three envelopes, the frequency resolution by a respective
1-bit syntax element bs_freq_res[ch] per envelope (220).
The controller 116 also transmits the syntax values
bs_freq_res[ch], which set the frequency resolution, to the
formatter 108 (220).
Thus, the calculator 112 calculates, for all LD_TRAN
frames, spectral envelope energy values as temporal means
over the duration of the individual envelopes 222a, 220,
222b, the calculator combining, in the frequency
CA 02664466 2012-07-11
- 21 -
resolution, different numbers of subbands as a function of
bs freq_res of the respective envelope.
The above description mainly dealt with the mode of operation of
the encoder with regard to calculating the signal energies for
representing the spectral envelopes in the time/frequency grid as
is specified by the SBR frame controller. Additionally, however,
the encoder of Fig. 1 also transmits, for each grid area of a
noise grid, a noise value which indicates, for this temporal noise
grid area, the magnitude of the noisy portion in the high-
frequency portion of the audio signal. Using these noise values,
an even better reproduction of the high-frequency portion from the
decoded low-frequency portion may be performed on the decoder
side, as will be described below. As may be seen from Fig. 2
(224), the number bs_num noise of the noise envelopes for LD TRAN
frames is always two, whereas the number for FIXFIX frames with
bs num env - 1 may also be one.
_ _
The subdivision of the LD TRANS SBR frames into the two noise
_
envelopes, but also of the FIXFIX frames into the one or two noise
envelopes, may be performed, for example, in the same manner as is
described in chapter 4.6.18.3.3 in the above-mentioned standard,
to which reference shall be made in this context, and which
passage shall be included, in this respect, by reference in the
description of the present application. In particular, for
example, the boundary between the two noise envelopes is
positioned, by the envelope data calculator 112 for LD TRAN
frames, onto the same boundary as - if the envelope 220a exists -
the envelope boundary between the envelope 220a and the transient
envelope 220 and as - if the envelope 222 does not exist - the
envelope boundary between the transient envelope 220 and the
envelope 222b.
Before continuing with the description of a decoder which is able
to decode the encoded audio signal at output 104 of encoder 100 of
Fig. 1, the interplay between the analysis
CA 02664466 2009-03-24
- 22 -
filter bank 110 and the envelope data calculator 112 shall
be dealt with in more detail. By the box 250, Fig. 4b
depicts, by way of example, the individual subband values
which are output by the analysis filter bank 110. In Fig.
4b it is assumed that the time axis t again extends from
the left to the right in a horizontal manner. A column of
boxes in a vertical direction thus corresponds to the
subband values as obtained by the analysis filter bank 110
at a certain time slot, an axis f being intended to
indicate that the frequency is to increase in the upward
direction. Fig. 4b shows, by way of example, 16 successive
time slots belonging to an SBR frame 902. It is assumed, in
Fig. 4b, that the present frame is an LD_TRAN frame and
that the transient position is the same as was indicated,
by way of example, in Fig. 4. The resulting grid
classification within the frame 902 and/or the resulting
envelopes are also illustrated in Fig. 4b. Fig. 4b also
indicates the noise envelopes, specifically by 252 and 254.
Using the folmation of the sum of squares, the envelope
data calculator 112 now determines mean signal energies in
the temporal and spectral grid, as is depicted in Fig. 4b
by the dashed lines 260. In the embodiment of Fig. 4b, the
envelope data calculator 112 thus determines, for the
envelope 222a and the envelope 222b, only half as many
spectral energy values for representing the spectral
envelope as for the transient envelope 220. However, as may
also be seen, the spectral energy values for the
representation of the spectral envelopes are formed only by
means of the subband values 250 located in the higher-
frequency subbands 1 to 32, whereas the low-frequency
subbands 33 to 64 are ignored, since the low-frequency
portion is encoded, as is known, by the audio encoder 106.
In this context, it shall be noted, as a precaution, that
the number of the subbands here is only by way of example,
of course, as is the bundling of the subbands within the
individual envelopes to form groups of four or two,
respectively, as is indicated in Fig. 4b. To remain with
the example of Fig. 4b, a total of 32 spectral energy
CA 02664466 2009-03-24
- 23
values are calculated by the envelope data calculator 112
in the example of Fig. 4b for representing the spectral
envelopes, the quantization accuracy of which is performed
for encoding, again as a function of bs_amp_res, as was
described above. In addition, the envelope data calculator
112 determines a noise value for the noise envelopes 252
and 254, respectively, on the basis of the subband values
of the subbands 1 to 32 within the respective envelope 252
or 254, respectively.
Now that the encoder has been described above, the
following will provide a description of a decoder in
accordance with an embodiment of the present invention
which is suited to decode the encoded audio signal at the
output 103, said description below also addressing the
advantages entailed by the LD_TRAN class described with
regard to bit rate and delay.
The decoder of Fig. 5, which is generally indicated at 300,
comprises a data input 302 for receiving the encoded audio
signal, and an output 304 for outputting a decoded audio
signal. The input of a demultiplexer 306, which possesses
three outputs, is adjacent to the input 302. An audio
decoder 308, an analysis filter bank 310, a subband adapter
312, a synthesis filter bank 314 as well as an adder 316
are connected, in the order mentioned, between a first one
of these outputs and the output 304. The output of the
audio decoder 308 is also connected to a further input of
the adder 316. As will be described below, a connection of
the output of the analysis filter bank 310 to a further
input of the synthesis filter bank 314 may be provided
instead of the adder 316 with its additional input. The
output of the analysis filter bank 310, however, is also
connected to an input of a gain value calculator 318, the
output of which is connected to a further input of the
subband adapter 312, and which also comprises second and
third inputs, the second of which is connected to a further
output of the demultiplexer, and the third input of which
CA 02664466 2009-03-24
- 24 -
,
is connected, via an envelope data decoder 320, to the
third output of the multiplexer 306.
The mode of operation of the decoder 300 is as follows. The
demultiplexer 306 splits up the arriving encoded audio
signal at the input 302 by means of parsing. Specifically,
the demultiplexer 306 outputs the encoded signal relating
to the low-frequency portion, as has been generated by the
audio encoder 106, to the audio decoder 308 configured such
that it is able to obtain, from the information obtained, a
decoded version of the low-frequency portion of the audio
signal and to output it at its output. The decoder 300 thus
already has knowledge of the low-frequency portion of the
audio signal to be decoded. However, the decoder 300 does
not obtain any direct information on the high-frequency
portion. Rather, the output signal of the decoder 308 also
serves, at the same time, as a preliminary high-frequency
portion signal or at least as a master, or basis, for the
reproduction of the high-frequency portion of the audio
signal in the decoder 300. Portions 310, 312, 314, 318, and
320 from the decoder 300 serve to utilize this master to
reproduce, or to reconstruct, the final high-frequency
portion therefrom, this high-frequency portion thus
reconstructed being combined, by the adder 316, again with
the decoded low-frequency portion so to eventually obtain
the decoded audio signal 304. In this context it shall be
noted, for completeness' sake, that the decoded low-
frequency signal from the decoder 308 could also be subject
to further preparatory treatments before it is input into
the analysis filter bank 310, this not being shown,
however, in Fig. 5.
In the analysis filter bank 310, the decoded low-frequency
signal is again subject to a spectral dispersion with a
fixed time resolution and a frequency resolution which
essentially corresponds to that of the analysis filter bank
of the encoder 110. Remaining with the example of Fig. 4b,
the analysis filter bank 310 would output 32 subband values
CA 02664466.2009-03-24
- 25 -
,
per time slot, for example, said subband values
corresponding to the 32 low-frequency subbands (33-64 in
Fig. 4b). It is possible that the subband values as are
output by analysis filter bank 310 are reinterpreted, as
early as at the output of this filter bank, or before the
input of the subband adapter 312, as the subband values of
the high-frequency portion, i.e. are copied into the high-
frequency portion, as it were. However, it is also possible
that in the subband adapter 312, the low-frequency subband
values obtained from the analysis filter bank 310 initially
have high-frequency subband values added to them in that
all or some of the low-frequency subband values are copied
into the higher-frequency portion, such as the subband
values of subbands 33 to 64, as are obtained from the
analysis filter bank 310, into subbands 1 to 32.
In order to perform the adaptation to the spectral envelope
as has been encoded, on the encoder side, into the encoded
audio signal 104, the demultiplexer 306 will initially
forward that part of the encoded audio signal 302 which
relates to the encoding of the representation of the
spectral envelope, as has been generated by the encoder 114
on the encoder side, to the envelope data decoder 320,
which, in turn, will forward the decoded representation of
this spectral envelope to the gain values calculator 318.
In addition, the demultiplexer 306 outputs that part of the
encoded audio signal which relates to the syntax elements
for grid division, as have been introduced into the encoded
audio signal by the SBR frame controller 116, to the gain
values calculator 318. The gain values calculator 318 now
associates the syntax elements of Fig. 2 with the frames of
the audio decoder 308 in a manner which is as synchronized
as that of the SBR frame controller 116 on the encoder
side. For the exemplary frame contemplated in Fig. 4b, for
example, the gain values calculator 318 obtains, for each
time/frequency domain of the dashed grid 260, an energy
value from the envelope data decoder 320, which energy
values together represent the spectral envelope.
CA 02664466 2009-03-24
- 26 -
In the same grid 260, the gain values calculator 318 also
calculates the energy in the preliminarily reproduced high-
frequency portion so as to be able to normalize the
reproduced high-frequency portion in this grid and to
weight it with the respective energy values it has obtained
from the envelope data decoder 320, whereby the
preliminarily reproduced high-frequency portion is
spectrally adjusted to the spectral envelope of the
original audio signal. Here, the gain values calculator
takes into account the noise values which also have been
obtained from the envelope data decoder 320 per noise
envelope, so as to correct the weighting values for the
individual subband values within this noise frame. Thus,
what is forwarded at the output of the subband adapter 312
are subbands comprising subband values which are adapted
with corrected weighting values to the spectral envelope of
the original signal in the high-frequency portion. The
synthesis filter bank 314 puts together the high-frequency
portion thus reproduced in the time domain using these
spectral values, whereupon the adder 316 combines this
high-frequency portion with the low-frequency portion from
the audio decoder 308 into the final decoded audio signal
at the output 304. As is indicated by the dashed line in
Fig. 5, it is also possible, alternatively, for the
synthesis filter bank 314 to use, for synthesis, not only
the high-frequency subbands as have been adapted by subband
adapter 312, but to also use the low-frequency subbands as
directly correspond to the output of the analysis filter
bank 310. In this manner, the result of the synthesis
filter bank 314 would directly correspond to the decoded
output signal which could then be output at the output 304.
The above embodiments had in common that the SBR frames
comprised an overlap region. In other words, the time
division of the envelopes was adapted to the time division
of the frames, so that no envelope overlaps two adjacent
frames, for which purpose a respective signaling of the
CA 02664466 2012-07-11
- 27 -
envelope time grid was conducted, specifically by means of LD TRAN
and FIXFIX classes. However, problems will arise if transients
occur at the edges of the blocks or frames. In this case, a
disproportionately large number of envelopes is required to encode
the spectral data including the spectral energy values, or the
spectral envelope values, and the frequency resolution values. In
other words, more bits are consumed than would be required by the
location of the transients. In principle, two such "unfavorable"
cases may be distinguished, which are illustrated in Figs. 6a and
6b.
The first unfavorable situation will occur when the transient,
which is established by the transient detector 118, is located
almost at a frame start of a frame 404, as is illustrated in Fig.
6a. Fig. 6a shows an exemplary case wherein a frame 406 of the
FIXFIX class, which comprises a single envelope 408 which extends
over all 16 QMF slots, precedes the frame 404, at the start of
which a transient has been detected by the transient detector 118,
which is why the frame 404 has been associated, by the SBR frame
controller 116, with an LD TRAN class, with a transient position
pointing to the third QMF slot of the frame 404, so that the frame
404 is subdivided into three envelopes 410, 412, and 414, of which
envelope 412 represents the transient envelope, and the other
envelopes 410 and 414 surround same and extend to the frame
boundaries 416b and 416c of the respective frame 404 (with 416a
denoting the leading frame boundary of frame 406). Merely to avoid
confusion, it shall be pointed out that Fig. 6a is based on the
assumption that a different table than in Fig. 3 has been used.
As is now indicated by the arrow 418 which points to the first
envelope 410 in the LD_TRAN frame 404, the transmission of
spectral energy values, or the frequency resolution value and
noise value, specifically for the respective time domain, i.e. QMF
slots 0 and 1, is actually not justified, since the domain does
obviously not
CA 02664466 2009-03-24
- 28 -
correspond to any transient, but, conversely, is very small
in terms of time. This "expensive" envelope is therefore
highlighted in a hatched manner in Fig. 6a.
A similar problem will arise if a transient exists between
two frames, or is detected by the transient detector 118.
This case is represented in Fig. 6b. Fig. 6b shows two
successive frames 502 and 504, each having a length of 16
QMF slots, a transient having been detected by the
transient detector 118 between the two frames 502 and 504,
or in the vicinity of the frame boundary between these two
SBR frames 502 and 504, so that both frames 502 and 504
have been associated with an LD_TRAN class by the SBR frame
controller 116, both with only two envelopes 502a, 502b,
and 504a and 504b, respectively, such that the transient
envelopes 502b of the leading frame 502 and the transient
envelope 504b of the subsequent frame 504 will border on
the SBR frame boundary. As may be seen, the transient
envelope 502b of the first frame 502 is extremely short and
extends only over one QMF slot. Even for the presence of a
transient, this represents a disproportionately large
amount of expenditure for envelope encoding, since spectral
data are again encoded for the subsequent transient
envelope 504b, as was described above. Therefore, the two
transient envelopes 502b and 504b are highlighted in a
hatched manner.
Both cases which have been outlined above with reference to
Figs. 6a and 6b have in common, therefore, that in each
case envelopes (hatched area) are required which describe a
relatively short period and accordingly cost too many, or a
relatively large number of, bits. These envelopes contain a
spectral data set which might as well describe a complete
frame. However, the precise time division is necessary to
encapsulate the energy around the transients, since
otherwise pre-echoes will arise, as has been described in
the introduction to the description of the present
application.
CA 02664466 2009-03-24
- 29 -
Therefore, a description will be given below of an
alternative mode of operation of an encoder and/or a
decoder, by means of which the above problems in Figs. 6a
and 6b are addressed, or data sets which describe too short
a time period need not be transmitted on the encoder side.
If one considers, for example, the case of Fig. 6a, wherein
the transient detector 118 indicates the presence of a
transient in the vicinity of the start of the frame 404,
the SBR frame controller 116 will still associate, in the
embodiment described, the LD_TRAN class comprising the same
transient position indication with this frame, but no scale
factors and/or spectral energy values, and no noise portion
are generated by the envelope data calculator 112 and the
envelope data encoder 114 for the envelope 410, and no
frequency resolution indication is forwarded to the
formatter 108 for this envelope 410 by the SBR frame
controller 116, which is indicated in Fig. 7a, which
corresponds to the situation of Fig. 6a, in that the line
of the envelope 410 is depicted as a dashed line and that
the respective QMF slots are hatched to indicate that for
this purpose, the data stream output by the formatter 108
in the output 104 actually contains no data for high-
frequency reconstruction. On the decoder side, this "data
void" 418 is filled in that all necessary data, such as
scale factors, noise portion and frequency resolution, is
obtained from the respective data of the preceding envelope
408. More specifically, and as will be explained below in
more detail with reference to Fig. 9, the envelope data
decoder 320 concludes from the transient position
indication for the frame 404 that the case at hand is a
case in accordance with Fig. 6a, so that it does not expect
any envelope data for the first envelope in the frame 404.
To symbolize this alternative mode of operation, Fig. 5
indicates, by means of a dashed arrow, that in terms of its
mode of operation, or syntactical analysis, the envelope
data decoder 320 also depends on the syntax elements which
CA 02664466 2009-03-24
- 30 -
. '
are printed in bold in Fig. 2, in this case particularly on
the syntax element bs_transient_position. Now the envelope
data decoder 320 fills the data void 418 in that it copies
the respective data from the preceding envelope 408 for the
envelope 410. In this manner, the data set of the envelope
408 is extended from the preceding frame 406 to the first
(hatched) QMF slots of the second frame 404, as it were.
Thus, the time grid of the missing envelope 410 in the
decoder 300 is reconstructed again, and the respective data
sets are copied. Thus, the time grid of Fig. 7a again
corresponds to that of Fig. 6a with regard to the frame
404.
The approach in accordance with Fig. 7a offers a further
advantage over the approach described above with reference
to Fig. 3, since in this manner it is possible to always
accurately signal the transient start on the QMF slot. The
transients detected by the transient detector 118 may be
mapped more sharply as a result. To illustrate this
further, Fig. 8 depicts the case where, in accordance with
Fig. 3, a FIXFIX frame 602 comprising an envelope 604 is
followed by an LD_TRAN frame 606 comprising two envelopes,
namely a transient envelope 608 and a final envelope 610,
the transient position indication pointing to the second
QMF slot. As may be seen from Fig. 8, the transient
envelope 608 comprising the first QMF slot of the frame 606
starts in the same manner as it would have done in the case
of a transition position indication pointing to the first
QMF slot, as may be seen from Fig. 3. The reason for this
approach is that it is less worthwhile, for reasons of
encoding efficiency, to provide a third envelope at the
start of the frame 606 in the shifting of the transient
position indication from TRANS-POS = 0 to TRANS-POS = 1,
since, to this end, envelope data would specifically have
to be transmitted again. In accordance with the approach of
Fig. 7a, this does not present a problem, since it is
obvious that no envelope data at all need to be transmitted
for the start envelope 410. For this reason, an alignment -
CA 02664466 2009-03-24
- 31 -
. '
in units of QMF slots - of the transient envelope as a
function of the transient position indication in LD_TRAN
classes is possible in an effective manner in accordance
with the approach of Fig. 7a, for which purpose a possible
embodiment is represented in the table of Fig. 9. The table
of Fig. 9 represents a possible table as may be used in the
encoder of Fig. 1 and the decoder of Fig. 5, as an
alternative to the table of Fig. 3, in the context of the
alternative approach of Fig. 7a. The table includes seven
columns, wherein the categories of the first five
correspond to the first five columns in Fig. 3, i.e.
wherein from the first to the fifth columns the transient
position indication and, for this transient position
indication, the number of the envelopes provided in the
frame, the location of the first envelope boundary, the
location of the second envelope boundary, and the transient
index pointing to the envelope within which the transient
is located, are listed. The sixth column indicates the
transient position indication for which a data void 418 is
provided in accordance with Fig. 7a. As is indicated by a
one, this is the case for transient position indications
located between one and five (inclusively, in each case).
For the remaining transient position indications, a zero
has been entered in this column. The last column will be
dealt with below with reference to Fig. 7b.
Considering the case of Fig. 6b, in accordance with an
approach which is provided as an alternative or in addition
to the modification in accordance with Fig. 7a, an
unfavorable division of the transient area into the
transient envelopes 502b and 504b is prevented in that
virtually an envelope 502 is used which extends over the
QMF slots of both transient envelopes 502b and 504b, that
the scale factors which are obtained across this envelope
402 are transmitted along with the noise portion and the
frequency resolution, but only for the transient envelope
502b of the frame 502, and are simply used, on the decoder
side, also for the QMF slots at the start of the following
CA 02664466 2009-03-24
- 32 -
frame, as is indicated in Fig. 7b, which otherwise
corresponds to Fig. 6b, by the single hatching of the
envelope 502b, the indication of the transient envelope
504b by a dashed line, and the hatching of the QMF slot at
the start of the second frame 504.
Put more specifically, in the event of the occurrence of a
transient between the frames 502 and 504 in accordance with
Fig. 7b, the encoder 100 will act in the following manner.
The transient detector 118 indicates the occurrence of the
transient. Thereupon, the SBR frame controller 116 selects,
for the frame 502, as in the case of Fig. 6b, the LD_TRAN
class comprising a transient position indication pointing
to the last QMF slot. However, due to the fact that the
transient position indication points to the end of the
frame 502, the envelope data calculator 112 forms, from the
QMF output values, the scale factors or spectral energy
values, but not only across the QMF slot of the transient
envelope 502b, but rather across all QMF slots of the
virtual envelope 702, which additionally comprises the
three QMF slots immediately following the following frame
504. As a result, a delay is not connected at the output
104 of the encoder 100, since the audio encoder 106n can
forward the frame 504 to the formatter 108 only at the
frame end. In other words, the envelope data calculator 112
forms the scale factors by averaging across the QMF values
of the QMF slots of the virtual envelope 702 in a
predetermined frequency resolution, the resulting scale
factors being encoded by the envelope encoder 114 for the
transient envelope 502b of the first frame 502 and being
output to the formatter 108, the SBR frame controller 116
forwarding the respective frequency resolution value for
this transient envelope 502b. Irrespective of the decision
regarding the class of the frame 502, the SBR frame
controller 116 makes the decision on the class membership
of the frame 504. In the present case, by way of example,
no transient is now located in the vicinity of the frame
504 or within the frame 504, so that the SBR frame
CA 02664466 2009-03-24
- 33
controller 116 selects, in this exemplary case of Fig. 7b,
a FIXFIX class for the frame 504 with only one envelope
504a'. The SR frame controller 116 outputs the respective
decision to the formatter 108 and to the envelope data
calculator 112. However, the decision is interpreted in a
different way than usual. The envelope data calculator 112
namely has "remembered" that the virtual envelope 702 has
extended into the current frame 504, and it therefore
shortens the immediately adjacent envelope 504a' of the
frame 504 by the respective number of QMF slots in order to
determine the respective scale values only across this
smaller number of QMF slots and output same to the envelope
data encoder 114. Thus, a data void 704 arises, in the data
stream at the output 104, across the first three QMF slots.
In other words, in accordance with the approach of Fig. 7b,
the complete data set is initially calculated, on the
encoder side, for the envelope 702, for which purpose one
also uses data from the future QMF slots, from the point of
view of the frame 502, at the start of the frame 504, by
means of which the spectral envelope is calculated at the
virtual envelope. This data set is then transmitted to the
decoder as belonging to the envelope 502b.
At the decoder, the envelope data decoder 320 generates the
scale factors for the virtual envelope 702 from its input
data, as a result of which the gain values calculator 318
possesses all necessary information, for the last QMF slot
of the frame 502, or the last envelope 502b, to perform the
reconstruction still within this frame. The envelope data
decoder 320 also obtains scale factors for the envelope(s)
of the following frame 504 and forwards them to the gain
values calculator 318. From the fact that the transient
position input of the preceding LD_TRAN frame points to the
end of this frame 502, said gain values calculator 318
knows, however, that the envelope data which has been
transmitted for the final transient envelope 502b of this
frame 502 also relates to the QMF slots at the start of the
frame 504, which data belongs to the virtual envelope 702,
= CA 02664466 2009-03-24
- 34
which is why it introduces, or establishes, a specific
envelope 504b' for these QMF slots, and assumes, for this
envelope 504b' established, scale factors, a noise portion
and a frequency resolution obtained by the envelope data
calculator 112 from the respective envelope data of the
preceding envelope 502b so as to calculate, for this
envelope 504b', the spectral weighting values for the
reconstruction within the module 312. The gain values
calculator 318 only then applies the envelope data obtained
from the envelope data decoder 320 for the actual
subsequent envelope 504a' to the subsequent QMF slots
following the virtual envelope 702, and forwards gain
and/or weighting values which have been calculated
accordingly to the subband adapter 312 for high-frequency
reconstruction. In other words, on the decoder side, the
data set for the virtual envelope 702 is initially applied
only to the last QMF slot(s) of the current frame 502, and
the current frame 502 is thus reconstructed without any
delay. The data set of the second, subsequent frame 504
includes a data void 704, i.e. the new envelope data
transmitted is valid only as from the following QMF slot,
which is the third QMF slot in the exemplary example of
Fig. 7b. Thus, only one single envelope is transmitted in
the case of Fig. 7b. As in the first case, the missing
envelope 504b' is again reconstructed and filled with the
data of the previous envelope 502b. The data void 704 is
thus closed, and the frame 504 may be reproduced.
In the exemplary case of Fig. 7b, the second frame 504 has
been signaled with a FIXFIX class, wherein the envelope(s)
actually span(s) the entire frame. However, as has just
been described, on account of the preceding frame 502, or
its LD_TRAN class membership comprising a high transient
position indication, the envelope 504a' in the decoder is
restricted, and the validity of the data set does not
start, in terms of time, until several QMF slots later. In
this context, Fig. 7b addressed the case where the
transient rate is thin. However, if transients occur, in
CA 02664466 2009-03-24
- 35
several successive frames, at the edges in each case, the
transit position will be transmitted with the LDN-TRAN
class in each case and will be expanded accordingly in the
following frame, as has been described above with reference
to Fig. 7b. The first envelope, respectively, is reduced in
size, or restricted at its start, in accordance with the
expansion, as was described by way of example above with
reference to the envelope 504a' with reference to a FIXFIX
class.
As was described above, it is known, among encoders and
decoders, how far a transient envelope is expanded, at the
end of an LD_TRAN frame, into the subsequent frame, a
possible agreement on this also being depicted in the
embodiment of Fig. 9, or in the table depicted there, which
thus presents an example combining both modified approaches
in accordance with Figs. 7a and 7b. In this embodiment,
Table 9 is used by the encoder and the decoder. For
signaling the time grid of the envelopes, again, only
transient index bs_transient_position is used. In the case
of transient positions at the start of the frame, a
transmission of an envelope is prevented (Fig. 7a), as was
described above and may be seen from the second but last
column of the table of Fig. 9. What is also established, in
the last column of Fig. 9, in this connection is the
expansion factor with which - or the number of QMF slots
across which - a transient envelope at the end of the frame
is to be expanded into the subsequent frame (cf. Fig. 7b).
A difference in the signaling in accordance with Fig. 9
with regard to the first case (Fig. 7a) and the second case
(Fig. 7b) consists in the point of time of the signaling.
In case 1, the signaling takes place in the current frame,
i.e. there is no dependence regarding the preceding frame.
It is only the transient position that is crucial. The
cases in which the first envelope of a frame is not
transmitted may be seen, accordingly, on the decoder side,
from a table as in Fig. 9 comprising entries for all
transient positions.
CA 02664466 2009-03-24
- 36 -
In the second case, however, the decision is made in the
preceding frame and transferred into the next one. Using
the last table column in Fig. 9, specifically, an expansion
factor is specified the transient position of the
predecessor frame at which the transient envelope of the
predecessor frame is to be expanded into the next frame,
and to what extent. This means that - if in a frame a
transition position is established at the end of the
current frame, in accordance with Fig. 9, at the last or
second but last QMF slot - the expansion factor indicated
in the last column of Fig. 9 will be stored for the next
frame, by which means the time grid for the next frame is
thereby established, or specified.
Before a next embodiment of the present invention will be
addressed below, it shall be mentioned before that,
similarly to the approach for generating the envelope data
for the virtual envelope in accordance with Fig. 7b, the
generation of the envelope data for the envelope 408, in
the example of Fig. 7a, could also be determined over an
extended time period, i.e. by the two QMF slots of the
"saved" envelope 410, so that the QMF output values of the
analysis filter bank 110 for these QMF slots will also be
included in the respective envelope data of the envelope
408. However, the alternative approach is also possible, in
accordance with which the envelope data for the envelope
408 is determined only via the QMF slots associated with
it.
The preceding embodiments avoided a large amount of delay
using an LD-TRAN class. What follows is a description of an
embodiment in accordance with which the avoidance is
achieved by means of a grid, or envelope, classification
wherein envelopes may also extend across frame boundaries.
In particular, it shall be assumed in the following that
the encoder of Fig. 1 generates, at its output 104, a data
stream wherein the frames are classified into four frame
classes, i.e. a FIXFIX, a FIXVAR, a VARFIX and a VARVAR
= CA 02664466 2009-03-24
- 37 -
,
class, as has been established in the above-mentioned
MPEG4-SBR standard.
As is described in the introduction to the description of
the present application, the SBR frame controller 116, too,
classifies the sequence of frames into envelopes which may
also extend across frame boundaries. To this end, syntax
elements bs_num_rel_# are provided which specify for frame
classes FIXVAR, VARFIX and VARVAR, among other things, the
position - in relation to the leading or trailing frame
boundary of the frame - at which the first envelopes starts
and/or the last envelope of this frame ends. The envelope
data calculator 112 calculates the spectral values, or
scale factors, for the grid specified by the envelopes with
the frequency resolution specified by the SBR frame
controller 116. As a consequence, envelope boundaries may
be arbitrarily spread, for the SBR frame controller 116,
across the frames and an overlap region by means of these
classes. The encoder of Fig. 1 may perfoLm the signaling
with the four different classes in such a manner that a
maximum overlap region from one frame results, which
corresponds to the delay of the CORE encoder 106 and, thus,
also to the time period which may be buffered without
causing an additional delay. Thus it is ensured that there
will always be sufficient "future" values available for the
envelope data calculator 112 for pre-calculating and
sending envelope data even though most of these data will
have validity only in later frames.
In accordance with the present embodiment, however, the
decoder of Fig. 5 now processes such a data stream with the
four SBR classes in a manner resulting in a low latency
with simultaneous compacting of the spectral data. This is
achieved by data voids in the bit stream. To this end,
reference shall initially be made to Fig. 10 which shows
two frames including their classification as results, in
accordance with the embodiment, from the encoder of Fig. 1,
the first frame being a FIXVAR frame and the second frame
CA 02664466 2009-03-24
- 38 -
being a VARFIX frame in this case, by way of example. In
the exemplary case of Fig. 10, the two successive frames
802 and 804 comprise two, or one, envelope(s), namely
envelopes 802a and 802b, and/or envelopes 804a,
respectively, the second envelope of the FIXVAR frame 802
extending into the frame 804 by three QMF slots, and the
start of the envelope 804a of the VARFIX frame 804 being
located at QFM slot 3 only. With regard to each envelope
802a, 802b and 804a, the data stream at the output 104
contains scale factor values determined by the envelope
data calculator 112 by averaging the QMF output signal of
the analysis filter bank 110 across the respective QMF
slots. For determining the envelope data for the envelope
802b, the calculator 112 resorts to "future" data of the
analysis filter bank 110, as was mentioned above, for which
purpose a virtual overlap region the size of a frame is
available, as is indicated in a hatched manner in Fig. 10.
To reconstruct the high-frequency portion for the envelope
802b, the decoder would have to wait until it receives the
reconstructed low-frequency portion from the analysis
filter band 310, which would cause a delay the size of a
frame, as was mentioned above. This delay may be prevented
if the decoder of Fig. 5 operates in the following manner.
The envelope data decoder 320 outputs the envelope data
and, in particular, the scale factors for the envelopes
802a, 802b and 804a to the gain values calculator 318.
However, the latter uses the envelope data for the envelope
802b, which extends into the subsequent frame 804, however
initially only for a first part of the QMF slots across
which this envelope 802b extends, namely that part going as
far as the SBR frame boundary between the two frames 802
and 804. Consequently, the gain values calculator 318 re-
interprets the envelope division in relation to the
division as provided by the encoder of Fig. 1 in the
encoding, and uses the envelope data initially only for
that part of the overlap envelope 802b which is located
within the current frame 802. This part is illustrated as
CA 02664466 2009-03-24
- 39 -
envelope 802b1 in Fig. 11, which corresponds to the
situation of Fig. 10. In this manner, the gain values
calculator 318 and the subband adapter 312 are able to
reconstruct the high-frequency portion for this envelope
802b1 without any delay.
Due to this re-interpretation, the data stream at the input
302 naturally lacks envelope data for the remaining part of
the overlap envelope 802b. The gain values calculator 318
overcomes this problem in a similar manner to the
embodiment of Fig. 7b, i.e. it uses envelope data derived
from that for the envelope 802b1 so as to reconstruct, on
the basis of same, along with the subband adapter 312, the
high-frequency portion at the envelope 802b2 extending over
the first QMF slots of the second frame 804 which
correspond to the remaining part of the overlap envelope
802b. In this manner, the data void 806 is filled.
Following the previous embodiments, wherein the transient
problem was addressed in different ways in a manner which
is effective in terms of bit rates, a description shall be
given below of an embodiment in accordance with which a
modified FIXFIX class as an example of a class with a frame
and grid boundary match is configured, in its syntax, in
such a manner that it comprises a flag, or a transient
absence indication, whereby it is possible to reduce the
frame size while incurring bit-rate losses, but at the same
time to reduce the quantity of the losses, since stationary
parts of the information and/or audio signal can be encoded
in a more bit rate-effective manner. In this context, this
embodiment may be employed both additionally in the above-
described embodiments and independently of the other
embodiments in the context of a frame class division with
FIXFIX, FIXVAR, VARFIX and VARVAR classes as was described
in the introduction to the description of the present
application, but while modifying the FIXFIX class, as will
be described below. Specifically, in accordance with this
embodiment, the syntax description of a FIXFIX class, as
CA 02664466 2009-03-24
- 40 -
was described above also with reference to Fig. 2, is
supplemented by a further syntax element, such as a one-bit
flag, the flag being set, on the encoder side, by the SBR
frame controller 116 as a function of the location of the
transients detected by the transient detector 118, to
indicate that the infolmation signal is or is not
stationary in the area of the respective FIXFIX frame. In
the folmer case, such as with a set transient absence flag,
in the event that the FIXFIX frame comprises several
envelopes, no envelope data signaling, or no transmission
of noise energy values and scale factors as well as
frequency resolution values, is performed in the encoded
data stream 104 for the envelope of the respective FIXFIX
frame or for the first envelope, in terms of time, in this
FIXFIX frame, but this missing information is obtained, on
the decoder side, from the respective envelope data for
that envelope of the preceding frame which is directly
preceding, in terms of time, it also being possible for
said frame to be a FIXFIX frame, for example, or any other
frame, said envelope data being contained in the encoded
information signal. In this manner, a bit rate reduction
may thus be achieved for a variant of the SBR encoding with
a smaller delay, or a combination of the bit rate increase
in such a low-delay variant may be achieved on account of
the increased, or doubled, repetition rate. In combination
with the above-described embodiments, such a signaling
provides a completion with regard to the bit rate
reduction, since it is not only transient signals that may
be transmitted and/or encoded in a bit rate-reduced manner,
but also stationary signals. With regard to obtaining or
deriving the missing envelope data information, reference
shall be made to the description with regard to the
previous embodiments, specifically with regard to Figs. 12
and 7b.
The following shall be noted with regard to the
illustrations concerning Figs. 6a to 11. Sometimes,
different tables from those of Fig. 3 have been used as the
CA 02664466 2009-03-24
- 41 -
,
basis for these figures. Naturally, such differences may
also apply to the definition of the noise envelopes. With
LD_TRAN classes, the noise envelopes may always extend
across the entire frame, for example. In the case of Figs.
7a and 7b, the noise values of the preceding frame or of
the preceding envelope would then be used for high-
frequency reconstruction on the part of the decoder, for
example for the first few QMF slots, which in this case are
2 or 3 in number, by way of example, and the actual noise
envelope would be shortened accordingly.
In addition, it shall be noted, with regard to the approach
of Figs. 7b and 11, that there are numerous possibilities
of how the envelope data or the scale factors for the
virtual envelopes 702 and 802b, respectively, may be
transmitted. As was described, scale factors are determined
for the virtual envelope via the QMF slots, which are four
in number, by way of example, in Fig. 7b, and six in
number, by way of example, in Fig. 11, specifically by
means of averaging, as was described above. In the data
stream, these scale factors, determined via the respective
QMF slots, for the transient envelope 502b or the envelope
502b1 may be transmitted. In this case, the calculator 318
might possibly take into account, on the decoder side, that
the scale factors, or the spectral energy values, have been
deteLmined, however, across the entire area to be four and
six QMF slots, respectively, and it would therefore
subdivide the magnitude of these values into the two
partial envelopes 502b and 504b', respectively, and 802b1
and 802b2, respectively, in a ratio which corresponds, for
example, to the ratio between the QMF slots associated with
the first frames 502 and 802, respectively, and the second
frames 504 and 804, respectively, so as to utilize the
portions, thus subdivided, of the scale factors transmitted
for controlling the spectral shaping in the subband adapter
312. However, it would also be possible that the encoder
directly transmits such scale factors which may initially
be directly applied, on the decoder side, for the first
CA 02664466 2009-03-24
- 42 -
partial envelopes 502b and 802b1, respectively, and which
are re-scaled accordingly for the following partial
envelopes 504b' or 804b' or 802b2, respectively, depending
on the overlap of the virtual envelopes 702 and 802b,
respectively, with the second frames 504 and 804,
respectively. The manner in which the energy is divided up
between the two partial envelopes may be arbitrarily
specified between the encoder and the decoder. In other
words, the encoder may directly transmit such scale factors
which may be directly applied, on the decoder side, for the
first partial envelopes 502b and 502b1, respectively,
because the scale factors have only been averaged over
these partial envelopes and/or the respective QMF slots.
This case may be illustrated, by way of example, as
follows. In the event of a more or less overlapping
envelope, wherein the first part consists of two time
units, or QMF slots, and the second consists of three time
units, what happens on the encoder side is that only the
first part is correctly calculated and/or the energy values
are averaged only in this part, and the respective scale
factors are output. In this manner, the envelope data
precisely matches the respective time portion in the first
part. However, the scale factors for the second part are
obtained from the first part and are scaled in accordance
with the dimensional proportions as compared to the first
part, i.e., in this case, 3/2 times scale factors of the
first part. This opportunity shall be taken to point out
that in the above the term 'energy' was used synonymously
with scale factor; energy, or scale factor, resulting from
the sum of all energy values of an SBR band along a time
period of an envelope. In the example which just been
illustrated, the auxiliary scale factors in each case
describe the sum of the energies of the two time units in
the first part of the more or less overlapping envelope for
the respective SBR band.
In addition, provision may also be made, of course, for the
spectral envelopes, or scale values, to always be
CA 02664466 2009-03-24
- 43
transmitted, in the above embodiments, in a manner which is
normalized to the number of QMF slots which are used for
determining the respective value, such as the square
average energy - i.e. the energy noLmalized to the number
of contributing QMF slots and the number of QMF spectral
bands - within each frequency/time grid area. In this case,
the measures which have just been described for splitting,
on the encoder side or decoder side, of the scale factors
for the virtual envelopes into the respective sub-portions
are not necessary.
With regard to the above description, several other points
shall also be noted. Even though a description has been
given, for example, in Fig. 1, that a spectral dispersion
is perfoimed, by means of the analysis filter bank 110,
with a fixed time resolution, which will then be adapted,
by the envelope data calculator 112, to the time/frequency
grid set by the controller 116, alternative approaches are
also feasible, in accordance with which - with regard to a
time/frequency resolution adapted to the specification
given by the controller 316 - the spectral envelope in this
resolution is calculated directly, without the two stages
as are shown in Fig. 1. The envelope data encoder 114 of
Fig. 1 may be missing. On the other hand, the type of the
encoding of the signal energies representing the spectral
envelopes could be performed, for example, by means of
differential encoding, it being possible for the
differential encoding to be implemented in a time or
frequency direction or in a hybrid form, such as in a
frame-wise or envelope-wise manner in the time and/or
frequency direction(s). It shall be noted, with reference
to Fig. 5, that the order in which the gain values
calculator performs the normalization with the signal
energies contained in the high-frequency portion which is
preliminarily reproduced, and the weighting with the signal
energies transmitted by the encoder for signaling the
spectral envelopes, are irrelevant. The same naturally also
applies to the correction for taking into account the noise
CA 02664466 2009-03-24
- 44 -
portion values per noise envelope. It shall also be noted
that the present invention is not boundaryed to spectral
dispersions by means of filter banks. Rather, a Fourier
transformation and/or inverse Fourier transformation or
similar time/frequency transformations could naturally also
be employed, wherein, for example, the respective
transformation window is shifted by the number of audio
values which is to correspond to a time slot. It shall also
be noted that there may be provisions that the encoder does
not perform the determination and the encoding of the
spectral envelope and the introduction of same into the
encoded audio signal with regard to all subbands in the
high-frequency portion in the time/frequency grid. Rather,
the encoder could also determine such portions of the high-
frequency portion for which it is not worthwhile to perform
a reproduction on the decoder side. In this case, the
encoder transmits, to the decoder, for example, the
portions of the high-frequency portion and/or the subband
areas in the high-frequency portion for which the
reproduction is to be performed. In addition, various
modifications are also possible with regard to setting the
grid in the frequency direction. For example, one may
provide that no setting of the frequency grid is performed,
wherein in this case the syntax elements bs_freq_res could
be missing and, for example, the full resolution would
always be used. In addition, an adjustability of the
quantization step width of the signal energies for
representing the spectral envelopes may be omitted, i.e.
the syntax element bs_amp_res could be missing. In
addition, a different down-sampling could be performed in
the down-sampler of Fig. 1 instead of a down-sampling by
every other audio value, so that high and low-frequency
portions would have different spectral extensions. In
addition, the table-assisted dependence of the grid
division of the LD_TRAN frames on bs_transient_position is
only exemplary, and an analytical dependence of the
envelope extensions and of the frequency resolution would
also be feasible.
CA 02664466 2009-03-24
- 45
At any rate, the above-described examples of an encoder and
a decoder allow the use of the SBR technology also for the
AAC-LD encoding scheme of the above-cited standard. The
large delay of AAC + SBR, which conflicts with the goal of
AAC-LD with a short algorithmic delay of about 20 ms at 48
kHz and a block length of 480, may be overcome using the
above embodiments. Here, the disadvantage of a linkage of
AAC-LD with the previous SBR defined in the standard, which
is due to the shorter frame length of the AAC-LD 480 or 512
as compared to 960 or 1024 for AAC-LD, which frame length
causes the data rate for an unchanged SBR element as
defined in the standard to double that of HE AAC, would be
overcome. Subsequently, the above embodiments enable the
reduction of the delay of AAC-LD + SBR and a simultaneous
reduction of the data rate for the side infoLmation.
In particular, in the above embodiments, the delays for an
LID variant of the SBR module the overlap region of the SBR
frames was removed in order to reduce the system. Thus, the
possibility of being able to place envelope boundaries
and/or grid boundaries irrespective of the SBR frame
boundary is dispensed with. The treatment of transients,
however, is then taken over by the new frame class LD_TRAN,
so that the above embodiments also require only one bit for
signaling so as to indicate whether the current SBR frame
is that of a FIXFIX class or of an LD_TRAN class.
In the above embodiments, the LD_TRAN class was defined
such that it has envelope boundaries, in a manner which is
always synchronized to the SBR frame, at the edges and
variable boundaries within the frame. The interior
distribution was determined by the position of the
transients within the QMF slot grid or time slot grid. A
small envelope which encapsulates the energy of the
transient was distributed around the position of the
transient. The remaining areas were filled up with
envelopes to the front and to the back up to the edges. To
CA 02664466 2009-03-24
- 46 -
this end, the table of Fig. 3 was used by the envelope data
calculator 312 on the encoder side, and by the gain values
calculator 318 on the decoder side, where a predefined
envelope grid is stored in accordance with the transient
position, the table of Fig. 3 naturally only being
exemplary, and, in individual cases, variations may
naturally also be made, depending on the case of
application.
In particular, the LD_TRAN class of the above embodiments
thus enables compact signaling and adjusting of the bit
requirement to an LD environment with a double frame rate,
which thus also requires a double data rate for the grid
infolmation. Thus, the above embodiments eliminate
disadvantages of previous SBR envelope signaling in
accordance with the standard, which disadvantages consisted
in that for VARVAR, VARFIX and FIXVAR classes the bit
requirements for transmitting the syntax elements and/or
side information were high-scale, and that for the FIXFIX
class a precise temporal adjustment of the envelopes to
transients within the block was not possible. By contrast,
the above embodiments enable conducting a delay
optimization on the decoder side, specifically a delay
optimization by six QMF time slots or 384 audio samples in
the audio signal original area, which roughly corresponds
to 8 ms at 48 kHz of audio signal sampling. In addition,
the elimination of the VARVAR, VARFIX and FIXVAR frame
classes enables savings in the data rate for the
transmission of the spectral envelopes, which results in
the possibility of higher data rates for low-frequency
encoding and/or the core and, thus, improved audio quality.
Effectively, the above embodiments provide the transients
to be enveloped within the LD_TRAN class frames which are
synchronous to the SBR frame boundaries.
It shall be noted, in particular, that, unlike the previous
exemplary table of Fig. 3, the transient envelope length
may also comprise more than only 2 QMF time slots, the
CA 02664466 2009-03-24
- 47 -
transient envelope length preferably being smaller than 1/3
of the frame length, however.
With regard to the above description it shall also be noted
that the present invention is not boundarved to audio
signals. Rather, the above embodiments could naturally also
be employed in video encoding.
It shall also be noted with regard to the above embodiments
that the individual blocks in Figs. 1 and 5 may be
implemented both in hardware and in software, for example,
e.g. as parts of an ASIC or as program routines of a
computer program.
This opportunity shall be taken to note that, depending on
the circumstances, the inventive scheme may also be
implemented in software. Implementation may be on a digital
storage medium, in particular a disk or CD with
electronically readable control signals which may interact
with a programmable computer system such that the
respective method is performed. Generally, the invention
thus also consists in a computer program product with a
program code, stored on a machine-readable carrier, for
performing the inventive method, when the computer program
product runs on a computer. In other words, the invention
may thus be realized as a computer program having a program
code for performing the method, when the computer program
runs on a computer. With regard to the embodiments
discussed above, it shall also be noted that the encoded
information signals generated there may be stored on, e.g.,
a storage medium, such as an electronic storage medium.
