 |
Description  |
|
|
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to a stereo audio encoding apparatus that is
effective for encoding digital audio signal data for digital transmission
or storage to a digital data storage medium, and also to a method
therefor.
2. Description of the Prior Art
While many digital compression audio coding methods have existed for the
last two decades, standardization efforts of digital compression source
coding methods for wideband audio signals of 15 kHz or 20 kHz bandwidth
have only taken place recently. The Near Instantaneous Companding Audio
Multiplex (NICAM) has been adopted as a broadcast standard in the mid
1980s by various countries to produce sounds with quality comparable to FM
stereo broadcast. In 1991, a subband coding (SBC) using feedforward
quantization scheme, used in conjunction with psychoacoustic modelling,
formed the core method of the audio coding standard to be adopted by the
ISO/WG11/MPEG (Moving Picture Experts Group). The subband coding scheme
would be the audio coding algorithm for coded representation of moving
picture information and associated audio at a total data rate of 1.5Mbps
(Megabits per second). The bit rates at which the audio coding algorithm
must work ranges from 64 kbps (kilobits per second) to 192 kbps per single
audio channel.
Description of the subband coding scheme using quadrature mirror filter for
the subband filterbank and using psychoacoustics for the dynamic bit
allocation can be found in United States Patent Application of Publication
No. 4972484 dated Nov. 20, 1990. Detailed description of a similar subband
coding method can be found in the document "Second Draft of Proposed
Standard of Information Technology--Coding of Moving Pictures and
Associated Audio for Digital Storage Media up to about 1.5 Mbps", Part 3:
Audio Coding Standard ISO/IEC JTC1/SC2/WG11 N0043 MPEG 90/001, September
1990. In the latter document, the subband coding is implemented using a
polyphase filterbank. In the stereo coding mode of this prior art, the
subband encoder involves partitioning of the audio samples of each audio
channel into 32 subbands via a polyphase filterbank, FFT analysis to
determine psychoacoustic parameters, use of these parameters for adaptive
bit allocation to subbands, mid-tread quantization of subband samples and
transmission of essential side information. The essential side information
includes bit allocation and scale factor data. This is illustrated in FIG.
5. At the decoder, the side information is used for the dequantization.
Output samples are reconstructed after passing through an inverse
filterbank.
In order to obtain better quality sounds at lower bit rates, it has been
proposed in the ISO/MPEG audio algorithm an option of joint stereo coding.
Joint stereo coding exploits the interchannel irrelevancy in a stereo pair
of audio channels for bitrate reduction. The joint stereo coding used in
ISO/MPEG is termed as intensity stereo coding. The purpose of this
technique is to increase the sound quality of that obtain at a higher bit
rate and/or reduce the bitrate for stereophonic signals. The intensity
stereo technique makes use of psychoacoustical results which show that at
frequencies above 2 kHz, the localization of the stereophonic image within
a critical band is determined by temporal envelope and not by the temporal
fine structure of the audio signal. This technique involves the
transmission of the summed signals instead of the individual left and
right signals for subbands that are to be in the stereo irrelevancy mode.
Stereophonic image is preserved by transmitting the scale factors of both
the channels. Quantization of the common summed samples, coding of these
summed samples and coding of common bit allocation are performed in the
same manner as in independent coding of each audio signal.
The intensity stereo scheme suggested in the MPEG document MPEG 90/011
recommends that the left and right subband samples be added. These added
values, serving as common subband samples, are scaled in the normal way.
The originally determined scale factors of the left and right channel
subband signals are transmitted according to the bitstream syntax.
Quantization of common subband samples, and coding of common bit
allocation are performed in the same way as independent coding. For a very
high positive correlation between two channels, this scheme will work.
However, for channels with negative correlation, the reproduced sound
quality would deteriorate tremendously.
An illustration is provided below using opposite phase left and right
signals.
If the magnitude of the original or broadcasted left and right signals L
and R in one frame are as follows:
L={10, 9, 8, 9, 6, -7, 5, -6, 8, 5}
R={-10, -9, -7, -7, -6, 8, -5, 6, -10, -5}
the maximums SF.sub.l and SF.sub.r of the absolute number in each frame of
sampled signals can be expressed as follows:
SF.sub.l =10
SF.sub.r =10
These values SF.sub.l and SF.sub.2 are referred to as left and right scale
factors.
Power P.sub.1 in left channel is as follows:
##EQU1##
wherein l.sub.i is a sampled data in signal L and n is the total number of
sampled data (which is 10 in this example) Power P.sub.r in right channel
is as follows:
##EQU2##
wherein r.sub.i is a sampled data in signal R.
According to the prior art, the left and right sampled signals L and R are
reproduced, using the left and right scale factors SF.sub.l and SF.sub.r,
to signals L' and R' as explained as follows.
An average between the left and right channel signals can be given as
follows:
{(l.sub.i +r.sub.i)/2}={0, 0, 0.5, 1, 0, 0.5, 0, 0, -1, 0}
Let SF.sub.m, which is the maximum absolute magnitude of the signal
obtained from averaging between the left and right channel signals, be
termed as the combined scale factor. In this example, SF.sub.m =1. The
left and right signals are reproduced according to the following equations
:
L'=SF.sub.l *{(L+R)/2}/SF.sub.m
R'=SF.sub.r *{(L+R)/2}/SF.sub.m
Thus,
L'={0, 0, 5, 10, 0, 5, 0, 0, -10, 0}
R'={0, 0, 5, 10, 0, 5, 0, 0, -10, 0}
are obtained and are used for audio signals supplied to left and right
speakers.
Reconstructed powers P.sub.l ', and P.sub.l ', for left and right channels
are as follows.
##EQU3##
When the signals L' and R' are used, about 50% of the power is reduced when
the reconstruction system of the prior art is used.
SUMMARY OF THE INVENTION
Therefore, an object of the present invention is to provide a stereo audio
encoding apparatus and method for suppressing the loss of quality in the
reproduced audio signal.
To achieve this object, according to the present invention, a stereo audio
encoding method for encoding left and right original signals, each defined
by a train of frames and each frame containing a plurality of sampled
data, to a left and right reproduced signals, comprising the steps of:
(a) calculating a correlation between said left and right original signals
to determine whether the pair of said left and right original signals have
an opposite phase characteristics and left and right original signals
having a same phase characteristics;
(b) processing said left and right original signals having the opposite
phase characteristics according to a power equalization method to obtain
said left and right reproduced signals; and
(c) processing said left and right original signals having the same phase
characteristics according to an error minimization method to obtain said
left and right reproduced signals.
Because the signals are encoded using a scale factor modified according to
the phase coefficient between plural audio signals, loss of audio signal
quality can be prevented at low bit rates.
BRIEF DESCRIPTION OF THE DRAWINGS
The present invention will become more fully understood from the detailed
description given below and the accompanying diagrams wherein:
FIG. 1 is a block diagram of the encoder used in the stereo audio encoding
apparatus according to the preferred embodiment of the present invention,
FIGS. 2a, 2b and 2c taken together as shown in FIG. 2 show a flow chart
describing the operation of the encoder used in the preferred embodiment,
FIG. 3 is a flow chart showing detailed of steps for calculating the scale
factor modifier in the preferred embodiment,
FIG. 4a is a graph showing the original audio signal applied to the encoder
of FIG. 1,
FIGS. 4b and 4c are graphs showing the right and left channel results of
processing the audio signal shown in FIG. 4a by the encoder of FIG. 1, and
FIG. 5 is a block diagram of the encoder according to the prior art.
FIGS. 6a and 6b are flowcharts which are useful for illustrating the
operation of the exemplary embodiment of the present invention.
DESCRIPTION OF PREFERRED EMBODIMENTS
The preferred embodiments of the present invention are described below with
reference to the accompanying figures.
Referring to FIG. 1, a stereo audio encoding apparatus according to the
present invention has: filter banks 2 and 9 for receiving left and right
channel audio signals along lines 1 and 8, respectively; left and right
scale factor detectors 3 and 10 for detecting left and right scale factors
SF.sub.l and SF.sub.r, respectively; transcoding unit 4 for producing
scaled left and right samples L, R, and scaled average M of the left and
right samples; left and right scale factor modifiers 5 and 12 for
modifying the left and right scale factors SF.sub.l and SF.sub.r to
SF'.sub.l and SF'.sub.r (or to SF".sub.l and SF".sub.r); a psychoacoustics
model generator 13; and a multiplexer 6 for multiplexing the scale
factors, the scaled left, right and average samples L, R and M.
Multiplexer 6 produces left and right scale factors SF.sub.l and SF.sub.r
to SF'.sub.l and SF'.sub.r (or to SF".sub.l and SF".sub.r), and the scaled
left, right and average samples L, R and M in a time sharing manner. The
signal produced from the multiplexer 6 may be transmitted through a cable
or some other data transmission means to a receiver which includes
demultiplexer 14, inverse transcoding unit 15, and inverse filter banks 16
and 17.
Filter Bank 2 receives digital audio signal and divides, in frequency
domain, the audio signal covering a frequency band of 0-24 kHz into a
number of, such as 32, frequency ranges 0-749.999 Hz, 750-1499.999 Hz,
1500-2249.999 Hz, each range covering 750 Hz. Thus, there are 32 outputs
from the filter bank 2, but only one output is shown for the sake of
brevity. Each output from the filter bank 2 produces digital data sampled
at a predetermined sampling rate, such as 0.67 ms. Furthermore, filter
bank 2 divides each output, in time domain, into frames, each frame being
8 ms long. Thus, each frame has a train of twelve sampled data.
It is to be noted that, of the 32 outputs, the outputs in the high
frequency ranges, such as above 2 kHz are processed according to the
intensity stereo method and the low frequency ranges are processed
according to the individual stereo method, the details of which are
disclosed in (i) U.S. Pat. No. 4,972,484 to Theile et al. issued Nov. 20,
1990, and (ii) "Second Draft of Proposed Standard of Information
Technology--Coding of Moving Pictures and Associated Audio for Digital
Storage Media up to about 1.5Mbps", Part 3: Audio Coding Standard ISO/IEC
JTC1/SC2/WG11 N0043 MPEG 90/001, September 1990, both are understood as
being taken herein by reference.
The present invention is particularly directed to the improvement in the
intensity stereo method, that is the processing in the high frequency
ranges. The improved intensity stereo method according to the present
invention has two modes of operations: the first mode is a power
equalization mode which is applied to a case when the left and right
signals have approximately opposite phase; and the second mode is an error
minimization mode which is applied to a case when the left and right
signals have relatively similar phase.
The description herein is directed only to the processing of one output
from the higher ranges of the 32 outputs, but other outputs from the
higher ranges are processed in the same manner.
One frame left signal L from the filter bank 2 is serially applied to
transcoding unit 4 which at the same time receives a corresponding one
frame right signal R from the filter bank 9. Two examples of left and
right signals L and R are shown below.
EXAMPLE 1
L={10, 9, 8, 9, 6, -7, 5, -6, 8, 5}
R={-10, -9, -7, -7, -6, 8, -5, 6, -10, -5}
EXAMPLE 2
L={10, 9, 8, 9, 6, -7, -5, -6, -8, -5},
R={12, 14, 12, 19, 16, -17, -20, -15, -10, -18},
As apparent from the above, Example 1 has almost an opposite phase and
therefore operates under the power equalization mode, and Example 2 has a
similar phase and therefore operates under the error minimization mode.
The detection between these two modes is done as follows.
All the corresponding sampled data between L and R are added, and the
absolute of the added sums are added to obtain an evaluation value. In the
case of Examples 1 and 2, the evaluation values E1 and E2 are as follows.
##EQU4##
wherein l.sub.i and r.sub.i are sampled data in left and right signals,
respectively.
Then, when the evaluation value is compared with a predetermined value,
such as 50, and is determined as the power equalization mode when the
evaluation value is less than the predetermined value, and is determined
as the error minimization mode when the evaluation value is equal to or
greater than the predetermined value. Any other method for detecting the
mode can be used.
First, the operation under the power equalization mode is shown in FIG. 6a
and will be described, using the above given Example 1.
When each frame is applied to the scale factor detector 3 (or 10), an
absolute value of each sampled data is taken, and the scale factor
detector 3 (or 10) produces a maximum absolute sampled data as a left
scale factor SF.sub.l (or a right scale factor SF.sub.r). In the above
Example 1, the left and right scale factors are as follows:
SF.sub.l =10,
and
SF.sub.r =10.
According to the present invention, under the power equalization mode, left
and right scale factor modifiers 5 and 12 modify the left and right scale
factors SF.sub.l and SF.sub.r to SF.sub.l ' and SF.sub.r ', respectively,
in the steps as described below.
First the power P.sub.l and P.sub.r of the left and right signals L and R
are obtained by the following equations:
##EQU5##
wherein l.sub.i is a sampled data in signal L and r.sub.i is a sampled
data in signal R.
Then, an average between the left and right channel signals are calculated
as follows:
{(l.sub.i +r.sub.i)/2}={0, 0, 0.5, 1, 0, 0.5, 0, 0, -1, 0}
Then, the scale factor of the combined samples, SF.sub.m, which is the
maximum absolute data, is calculated. Thus, SF.sub.m =1 is obtained.
Also a power of the average signal P.sub.m is obtained by the following
equation:
##EQU6##
wherein m is the data in the average signal.
Then, intervening signals L.sub.m and R.sub.m are calculated by the
following equations:
##EQU7##
The intervening signals L.sub.m and R.sub.m are equal to the reproduced
left and right signals used in the prior art.
Then, the power P.sub.lm and P.sub.rm of the intervening left and right
signals L.sub.m and R.sub.m are obtained by the following equations:
##EQU8##
which are not equal to prior art reconstructed powers P.sub.l, and
P.sub.r.
According to the present invention, scale factor modifier K.sub.l ' and
K.sub.r ' are calculated by the following equations:
##EQU9##
and the following relationships are given.
SF.sub.l '=K.sub.l '*SF.sub.l
SF.sub.r '=K.sub.r '*SF.sub.r
According to the present invention, and for the example given above,
modified left and right scale factors SF.sub.l ' and SF.sub.r ' are
obtained by the following equations.
##EQU10##
Then, by the use of the modified left and right scale factors SF.sub.l '
and SF.sub.r ', reproduced left and right signals L' and R' are calculated
by the following equations.
##EQU11##
To obtain equations (8a) and (8b), other approaches can be used.
Since the power P.sub.l ' and P.sub.r ' of the left and right signals L'
and R' can be calculated as follows,
##EQU12##
there is hardly any power change from the power P.sub.l and P.sub.r of the
original signal L and R, in the reproduced signals L' and R'. According to
the present invention, although it is inevitable that the temporal fine
structure of the sound is lost, the power within the temporal envelope is
maintained.
Next, the operation under the error minimization mode is shown in FIG. 6b
and will be described, using the above given Example 2.
In a similar manner to the above for the power equalization mode, the left
and right scale factors SF.sub.l and SF.sub.r are calculated as follows.
SF.sub.l =10,
and
SF.sub.r =20.
According to the present invention, under the error minimization mode, left
and right scale factor modifiers 5 and 12 modify the left and right scale
factors SF.sub.l and SF.sub.r to SF.sub.l " and SF.sub.r ", respectively,
in the steps as described below.
First the power P.sub.l and P.sub.r of the left and right signals L and R
are obtained by the following equations:
##EQU13##
Then, an interaction term I is calculated by the following equation:
##EQU14##
Also, an average between the left and right channel signals are calculated
as follows:
{(l.sub.i+r.sub.i)/2}
={11, 11.5, 10, 14, 11, -12, -12.5, -10.5, -9, -11.5}
Then, a combined scale factor SF.sub.m, which is the maximum absolute
sampled data, is obtained. In the above example, SF.sub.m =14.
According to the present invention, scale factor modifier K.sub.l " and
K.sub.r " are calculated by the following equations:
##EQU15##
and the following relationships are given.
SF.sub.l "=K.sub.l "*SF.sub.l
SF.sub.r "=K.sub.r "*SF.sub.r
Then, modified left and right scale factors SF.sub.l " and SF.sub.r " are
calculated by the following equations:
##EQU16##
Then, by the use of the modified left and right scale factors SF.sub.l "
and SF.sub.r ", reproduced left and right signals L" and R" are calculated
by the following equations.
##EQU17##
To obtain equations (13a) and (13b), other approaches can be used.
The errors D.sub.l " and D.sub.r " of left and right channel signals L" and
R" with respect to original signal L and R are calculated as follows.
##EQU18##
Since the powers P.sub.l " and P.sub.r " of the left and right signals L"
and R" can be calculated as follows,
##EQU19##
the powers P.sub.l " and P.sub.r " are very close to the powers P.sub.l
and P.sub.r of the original signal L and R.
As apparent from the above, in the other case where the left and right
channel signals are in phase and in which case the power condition would
have been satisfactory, emphasis is given to the finer temporal structure
of the audio by ensuring that the signals are reproduced with the minimum
error.
FIGS. 2a, 2b and 2c taken together as shown in FIG. 2 is a flowchart
showing operation of the stereo audio encoding apparatus of FIG. 1.
At step S11, left channel signal processing starts. At step S14, the
subband analysis is carried out at filter Bank 2 for dividing the signal
into subbands (32 frequency ranges) and also dividing into frames.
At step S19, scale factor SF.sub.l is calculated in left scale factor
detector 3.
In the meantime, steps S15, S17, S20, S23, S24 and S26 are carried out in
the psychoacoustics model generator 13 to determine a bit length of the
data in each subband so that the total bit rate, including left and right
channels, in one frame is equal to a predetermined bit rate. This is
disclosed in detail in the Audio Coding Standard ISO/IEC JTC1/SC2/WG11
N0043 MPEG 90/001.
Also, at step S16, it is detected whether or not the data is in the high
frequency range, such as above 2 kHz, or in the low frequency range which
should be in intensity stereo. The threshold frequency 2 kHz is not
constant but is variable and is determined in transcoding unit 4.
Then, at step S18, scale factor modifiers K.sub.1 ' and K.sub.r ', or
K.sub.l " and K.sub.r " are calculated. The detail of step S18 is shown in
FIG. 3.
Referring to FIG. 3, at steps S1 and S2, it is detected whether or not the
left channel signal L and the right channel signal R have a similar phase
or an opposite phase. If the signals L and R have a similar phase, step S4
is carried out, and if they have an opposite phase, step S3 is carried
out.
At step S3, operation under the power equalization mode is carried out, as
explained above so as to produce scale factor modifiers K.sub.1 ' and
K.sub.r '.
At step S4, operation under the error minimization mode is carried out, as
explained above so as to produce modified scale factors SF.sub.l " and
SF.sub.r ".
Referring back to FIG. 2a, at step S21, using the scale factor modifier
K.sub.l ' or K.sub.l ", scale factors SF.sub.l ' or SF.sub.l " is
calculated. This signal is applied to Steps S25, S26, S27, S29 and S31 for
a further processing.
In FIG. 4a a segment of the original input audio signal L (real line) and
signal R (dotted line) are shown in which the left and right channel
signals have almost the same phase. FIGS. 4b and 4c show the plot of the
right and left channels after the coding process, in which a real line
shows the original signal, a dotted line shows the results of the prior
art, and a dash line shows the results of the invention. It can be
observed from the plots that while the results for the right channel of
both dotted and dash lines are comparable, the plots of the left channel
for the dash line is closer to that of the original than the dotted line.
According to the present invention, the error minimization and power
equalization method to modify scale factors may be applied in various
audio coding algorithms which use feedforward quantization. The method
according to the present invention will improve the quality of high
fidelity sounds irrespective of the correlation between the sound
channels. The present invention is particularly useful in improving sound
quality at lower bit rates where an optimal stereophonic bit reduction
scheme is more significant. Subjective quality of the reconstructed audio
sequences at 64 kbps per audio channel has shown that the method according
to the present invention has resulted in a highly improved sound quality
over the intensity stereo coding method suggested in the MPEG audio coding
standard.
As will be known to those skilled in the art, the stereo audio encoding
method according to the present invention can suppress the loss of sound
quality in the reproduced audio signal.
The invention being thus described, it will be obvious that the same may be
varied in many ways. Such variations are not to be regarded as a departure
from the spirit and scope of the invention, and all such modifications as
would be obvious to one skilled in the art are intended to be included
within the scope of the following claims.
Although the present invention has been fully described in connection with
the preferred embodiments thereof with reference to the accompanying
drawings, it is to be noted that various changes and modifications are
apparent to those skilled in the art. Such changes and modifications are
to be understood as included within the scope of the present invention as
defined by the appended claims unless they depart therefrom.
* * * * *
|
|
 |
|
 |
Description |
|
