SIGNAL GENERATION FOR BINAURAL SIGNALS
Description
The present invention relates to the generation of a room reflection and/or reverberation
related contribution of a binaural signal, the generation of a binaural signal itself, and the
forming of an inter-similarity decreasing set of head-related transfer functions.
The human auditory system is able to determine the direction or directions where sounds
perceived come from. To this end, the human auditory system evaluates certain differences
between the sound received at the right hand ear and sound received at the left hand ear. The
latter information comprises, for example, so-called inter-aural cues which may, in turn, refer
to the sound signal difference between ears. Inter-aural cues are the most important means for
localization. The pressure level difference between the ears, namely the inter-aural level
difference (ILD) is the most important single cue for localization. When the sound arrives
from the horizontal plane with a non-zero azimuth, it has a different level in each ear. The
shadowed ear has a naturally suppressed sound image, compared to the unshadowed ear.
Another very important property dealing with localization is the inter-aural time difference
(ITD). The shadowed ear has a longer distance to the sound source, and thus gets the sound
wave front later than the unshadowed ear. The meaning of ITD is emphasized in the low
frequencies which do not attenuate much when reaching the shadowed ear compared to the
unshadowed ear. ITD is less important at the higher frequencies because the wavelength of
the sound gets closer to the distance between the ears. Hence, in other words, localization
exploits the fact that sound is subject to different interactions with the head, ears, and
shoulders of the listener traveling from the sound source to the left and right ear, respectively.
Problems occur when a person listens to a stereo signal that is intended for being reproduced
by a loud speaker setup via headphones. It is very likely that the listener would regard the
sound as unnatural, awkward, and disturbing as the listener feels that the sound source is
located in the head. This phenomenon is often referred in the literature as "in-the-head"
localization. Long-term listening to "in-the-head" sound may lead to listening fatigue. It
occurs because the information on which the human auditory system relies, when positioning
the sound sources, i.e. the inter-aural cues, is missing or ambiguous.

In order to render stereo signals, or even multi-channel signals with more than two channels
for headphone reproduction, directional filters may be used in order to model these
interactions. For example, the generation of a headphone output from a decoded multi-
channel signal may comprise filtering each signal after decoding by means of a pair of
directional filters. These filters typically model the acoustic transmission from a virtual sound
source in a room to the ear canal of a listener, the so-called binaural room transfer function
(BRTF). The BRTF performs time, level and spectral modifications, and model room
reflections and reverberation. The directional filters may be implemented in the time or
frequency domain.
However, since there are many filters required, namely Nx2 with N being the number of
decoded channels, these directional filters are rather long, such as 20000 filter taps at 44.1
kHz, and the process of filtering is computationally demanding. Therefore, the directional
filters are sometimes reduced to a minimum. The so-called head-related transfer functions
(HRTFs) contain the directional information including the interaural cures. A common
processing block is used to model the room reflections and reverberation. The room
processing module can be a reverberation algorithm in time or frequency domain, and may
operate on a one or two channel input signal obtained from the multi-channel input signal by
means of a sum of the channels of the multi-channel input signal. Such a structure is, for
example, described in WO 99/14983 Al. As just described, the room processing block
implements room reflections and/or reverberation. Room reflections and reverberation are
essential to localized sounds, especially with respect to distance and externalization -
meaning sounds are perceived outside the listener's head. The aforementioned document also
suggests implementing the directional filters as a set of FIR filters operating on differently
delayed versions of the respective channel, so as to model the direct path from the sound
source to the respective ear and distinct reflections. Moreover, in describing several measures
for providing a more pleasant listening experience over a pair of headphones, this document
also suggests delaying a mixture of the center channel and the front left channel, and the
center channel and the front right channel, respectively, relative to a sum and a difference of
the rear left and rear right channels, respectively.
However, the listening results achieved thus far still lack to a large extent a reduced spatial
width of the binaural output signal and a lack of externalization. Further, it has been realized
that despite the abovementioned measures for rendering multi-channel signals for headphone

reproduction, portions of voice in movie dialogs and music are often perceived unnaturally
reverberant and spectrally unequal.
Thus, it is the object of the present invention to provide a scheme for binaural signal
generation, yielding a more stable and pleasant headphone reproduction.
This object is achieved by devices according to any of claims 1, 3, 4 and 7 and methods
according to any of claims 16 to 19.
The first idea underlying the present application is that a more stable and pleasant binaural
signal for headphone reproduction may be achieved by differently processing, and thereby
reducing the similarity between, at least one of a left and a right channel of the plurality of
input channels, a front and a rear channel of the plurality of input channels, and a center and a
non-center channel of the plurality of channels, thereby obtaining an inter-similarity reduced
set of channels. This inter-similarity reduced set of channels is then fed to a plurality of
directional filters followed by respective mixers for the left and the right ear, respectively. By
reducing the inter-similarity of channels of the multi-channel input signal, the spatial width of
the binaural output signal may be increased and the externalization may be improved.
A further idea underlying the present application is that a more stable and pleasant binaural
signal for headphone reproduction may be achieved by performing - in a spectrally varying
sense - a phase and/or magnitude modification differently between at least two channels of
the plurality of channels, thereby obtaining the inter-similarity reduced set of channels which,
in turn, may then be fed to a plurality of directional filters followed by respective mixers for
the left and the right ear, respectively. Again, by reducing the inter-similarity of channels of
the multi-channel input signal, the spatial width of the binaural output signal may be
increased and the externalization may be improved.
The abovementioned advantages are also achievable when forming an inter-similarity
decreasing set of head-related transfer functions by causing the impulse responses of an
original plurality of head-related transfer functions to be delayed relative to each other, or - in
a spectrally varying sense - phase and/or magnitude responses of the original plurality of
head-related transfer functions differently relative to each other. The formation may be done
offline as a design step, or online during binaural signal generation, by using the head-related
PCT/EP2009/005548 as originally filed

transfer functions as directional filters such as, for example, responsive to an indication of
virtual sound source locations to be used.
Another idea underlying the present application is that some portions in movies and music
result in a more naturally perceived headphone reproduction, when the mono or stereo
downmix of the channels of the multi-channel signal to be subject to the room processor for
generating the room-reflections/reverberation related contribution of the binaural signal, is
formed such that the plurality of channels contribute to the mono or stereo downmix at a level
differing among at least two channels of the multi-channel signal. For example,, the inventors
realized that voices in movie dialogs and music are typically mixed mainly to the center
channel of a multi-channel signal, and that the center-channel signal, when fed to the room
processing module, results in an often unnatural reverberant and spectrally unequal perceived
output. The inventors discovered, however, that these deficiencies may be overcome by
feeding the center channel to the room processing module with a level reduction such as by,
for example, an attenuation of 3-12 dB, or specifically, 6 dB.
In the following, preferred embodiments are described in more detail with respect to the
figures, among which:
Fig. 1 shows a block diagram of a device for generating a binaural signal according to an
embodiment;
Fig. 2 shows a block diagram of a device for forming an inter-similarity decreasing set of
head-related transfer functions according to a further embodiment;
Fig. 3 shows a device for generating a room reflection and/or reverberation related
contribution of a binaural signal according to a further embodiment:
Fig. 4a and 4b show block diagrams of the room processor of Fig. 3 according to distinct
embodiments;
Fig. 5 shows a block diagram of the downmix generator of Fig. 3 according to an
embodiment;

Fig. 6 shows a schematic diagram illustrating a representation of a multi-channel signal
using spatial audio coding according to an embodiment;
Fig. 7 shows a binaural output signal generator according to an embodiment;
Fig. 8 shows a block diagram of a binaural output signal generator according to a further
embodiment;
Fig. 9 shows a block diagram of a binaural output signal generator according to an even
further embodiment;
Fig. 10 shows a block diagram of a binaural output signal generator according to a further
embodiment;
Fig. 11 shows a block diagram of a binaural output signal generator according to a further
embodiment;
Fig. 12 shows a block diagram of the binaural spatial audio decoder of Fig. 11 according to an
embodiment; and
Fig. 13 shows a block diagram of the modified spatial audio decoder of Fig. 11 according to
an embodiment.
Fig. 1 shows a device for generating a binaural signal intended, for example, for headphone
reproduction based on a multi-channel signal representing a plurality of channels and intended
for reproduction by a speaker configuration having a virtual sound source position associated
to each channel. The device which is generally indicated with reference sign 10, comprises a
similarity reducer 12, a plurality 14 of directional filters 14a-14h, a first mixer 16a and a
second mixer 16b.
The similarity reducer 12 is configured to turn the multi-channel signal 18 representing the
plurality of channels 18a-18d, into an inter- similarity reduced set 20 of channels 20a-20d.
The number of channels 18a-18d represented by the multi-channel signal 18 may be two or
more. For illustration purposes only, four channels 18a-18d have explicitly been shown in
PCT/EP2009/005548 as originally filed

Fig. 1. The plurality 18 of channels may, for example, comprise a center channel, a front left
channel, a front right channel, a rear left channel, and a rear right channel. The channels 18a-
18d have, for example, been mixed up by a sound designer from a plurality of individual
audio signals representing, for example, individual instruments, vocals, or other individual
sound sources, assuming that or with the intention that the channels 18a-18d are reproduced
by a speaker setup (not shown in Fig. 1), having the speakers positioned at predefined virtual
sound source positions associated to each channel 18a-18d.
According to the embodiment of Fig. 1, the plurality of channels 18a-18d comprises, at least,
a pair of a left and a right channel, a pair of a front and a rear channel, or a pair of a center and
a non-center channel. Of course, more than one of the just-mentioned pairs may be present
within the plurality 18 of channels 18a-18d. The similarity reducer 12 is configured to
differently process, and thereby reduce a similarity between channels of the plurality of
channels. , in order to obtain the inter- similarity reduced set 20 of channels 20a-20d.
According to a first aspect, the similarity between at least one of, a left and a right channel of
the plurality 18 of channels, a front and a rear channel of a plurality 18 of channels, and a
center and a non-center channel of the plurality 18 of channels may be reduced by the
similarity reducer 12, in order to obtain the inter- similarity reduced set 20 of channels 20a-
20d. According to a second aspect, the similarity reducer (12) may - additionally or
alternatively - perform - in a spectrally varying sense - a phase and/or magnitude
modification differently between at least two channels of the plurality of channels, in order to
obtain the inter-similarity reduced set 20 of channels.
As will be outlined in more detail below, the similarity reducer 12 may, for example, achieve
the different processing by causing the respective pairs to be delayed relative to each other, or
by subjecting the respective pairs of channels to delays of different amounts in, for example,
each of a plurality of frequency bands, thereby obtaining an inter-correlation reduced set 20 of
channels. There are, of course, other possibilities in order to decrease the correlation between
the channels. In even other words, the correlation reducer 12 may have a transfer function
according to which the spectral energy distribution of each channel remains the same, i.e. the
transfer function as a magnitude of one over the relevant audio spectrum range wherein,
however, the similarity reducer 12 differently modifies phases of subbands or frequency
components thereof. For example, the correlation reducer 12 could be configured such that
same causes a phase modification on all of, or one or several of, the channels 18 such that a
PCT/EP2009/005548 as originally filed

signal of a first channel for a certain frequency band is delayed relative to another one of the
channels by at least one sample. Further, the correlation reducer 12 could be configured such
that same causes the phase modification such that the group delays of a first channel relative
to another one of the channels for a plurality of frequency bands, show a standard deviation of
at least one eighth of a sample. The frequency bands considered could be the Bark bands or a
subset thereof or any other frequency band sub-division.
Reducing the correlation is not the only way to prevent the human auditory system from in-
the-head localization. Rather, correlation is one of several possible measures by use of which
the human auditory system measures the similarity of the sound arriving at both ears, and
thus, the in-bound direction of sound. Accordingly, the similarity reducer 12 may also achieve
the different processing by subjecting the respective pairs of channels to level reductions of
different amounts in, for example, each of a plurality of frequency bands, thereby obtaining
an inter-similarity reduced set 20 of channels in a spectrally formed way. The spectral
formation may, for example, exaggerate the relative spectrally formed reduction occurring,
for example, for rear channel sound relative to front channel sound due to the shadowing by
the earlap. Accordingly, the similarity reducer 12 may subject the rear channel(s) to a
spectrally varying level reductions relative to other channels. In this spectral forming, the
similarity reducer 12 may have phase response being constant over the relevant audio
spectrum range wherein, however, the similarity reducer 12 differently modifies magnitudes
of subbands or frequency components thereof.
The way in which the multi-channel signal 18 represents a plurality of channels 18a-18d is, in
principle, not restricted to any specific representation. For example, the multi-channel signal
18 could represent the plurality of channels 18a-18d in a compressed manner, using spatial
audio coding. According to the spatial audio coding, the plurality of channels 18a-18d could
be represented by means of a downmix signal down to which the channels are downmixed,
accompanied by downmix information revealing the mixing ratio according to which the
individual channels 18a-18d have been mixed into the downmix channel or downmix
channels, and spatial parameters describing the spatial image of the multi-channel signal by
means of, for example, level/intensity differences, phase differences, time differences and/or
measures of correlation/coherence between individual channels 18a-18d. The output of the
correlation reducer 12 is divided-up into the individual channels 20a-20d. The latter channels

may, for example, be output as time signals or as spectrograms such as, for example,
spectrally decomposed into subbands.
The directional filters 14a-14h are configured to model an acoustic transmission of a
respective one of channels 20a-20d from a virtual sound source position associated with the
respective channel to a respective ear canal of the listener. In Fig. 1, directional filters 14a-
14d model the acoustic transmission to, for example, the left ear canal, whereas directional
filters 14e-14h model the acoustic transmission to the right ear canal. The directional filters
may model the acoustic transmission from a virtual sound source position in a room to an ear
canal of the listener and may perform this modeling by performing time, level and spectral
modifications, and optionally, modeling room reflections and reverberation. The directional
filters 18a-18h may be implemented in time or frequency domain. That is, the directional
filters may be time-domain filters such as filters, FIR filters, or may operate on the frequency
domain by multiplying respective transfer function sample values with respective spectral
values of channels 20a-20d. In particular, the directional filters 14a-14h may be selected to
model the respective head-related transfer function describing the interaction of the respective
channel signal 20a-20d from the respective virtual sound source position to the respective ear
canal, including, for example, the interactions with the head, ears, and shoulders of a human
person. The first mixer 16a is configured to mix the outputs of the directional filters 14a-14d
modeling the acoustic transmission to the left ear canal of the listener to obtain a signal 22a
intended to contribute to, or even be the left channel of the binaural output signal, while the
second mixer 16b is configured to mix the outputs of the directional filters 14e-14h modeling
the acoustic transmission to the right ear canal of the listener to obtain a signal 22b, and
intended to contribute to or even be the right channel of the binaural output signal.
As will be described in more detail below with the respective embodiments, further
contributions may be added to signals 22a and 22b, in order to take into account room
reflections and/or reverberation. By this measure, the complexity of the directional filters 14a-
14h may be reduced.
In the device of Fig. 1, the similarity reducer 12 counteracts the negative side effects of the
summation of the correlated signals input into mixers 16a and 16b, respectively, according to
which a much reduced spatial width of the binaural output signal 22a and 22b and a lack of

externalization results. The decorrelation achieved by the similarity reducer 12 reduces these
negative side effects.
Before turning to the next embodiment, Fig. 1 shows, in other words, a signal flow for the
generation of a headphone output from, for example, a decoded multi-channel signal. Each
signal is filtered by a pair of directional filter pairs. For example, channel 18a is filtered by
the pair of directional filters 14a-14e. Unfortunately, a significant amount of similarity such
as correlation exists between channels 18a-18d in typical multi-channel sound productions.
This would negatively affect the binaural output signal. Namely, after processing the multi-
channel signals with a directional filter 14a-14h, the intermediate signals output by the
directional filters 14a-14h are added in mixer 16a and 16b to form the headphone output
signal 20a and 20b. The summation of similar/correlated output signals would result in a
much reduced spatial width of the output signal 20a and 20b, and a lack of externalization.
This is particularly problematic for the similarity/correlation of the left and right signal and
the center channel. Accordingly, similarity reducer 12 is to reduce the similarity between
these signals as far as possible.
It should be noted that most measures performed by similarity reducer 12 to reduce the
similarity between channels of the plurality 18 of channels 18a-18d could also be achieved by
removing similarity reducer 12 with concurrently modifying the directional filters to perform
not only the aforementioned modeling of the acoustic transmission, but also achieve the dis-
similarity such as decorrelation just mentioned. Accordingly, the directional filters would
therefore, for example, not model HRTFs, but modified head-related transfer functions.
Fig. 2, for example, shows a device for forming an inter-similarity decreasing set of head-
related transfer functions for modeling an acoustic transmission of a set of channels from a
virtual sound source position associated with the respective channel to the ear canals of a
listener. The device which is generally indicated by 30 comprises an HRTF provider 32, as
well as an HRTF processor 34.
The HRTF provider 32 is configured to provide an original plurality of HRTFs. Step 32 may
comprise measurements using a standard dummy head, in order to measure the head-related
transfer functions from certain sound positions to the ear canals of a standard dummy listener.
Similarly, the HRTF provider 32 may be configured to simply look-up or load the original

HRTFs from a memory. Even alternatively, the HRTF provider 32 may be configured to
compute the HRTFs according to a predetermined formula, depending on, for example,
virtual sound source positions of interest. Accordingly, HRTF provider 32 may be configured
to operate in a design environment for designing a binaural output signal generator, or may be
part of such a binaural output signal generator signal itself, in order to provide the original
HRTFs online such as, for example, responsive to a selection or change of the virtual sound
source positions. For example, device 30 may be part of a binaural output signal generator
which is able to accommodate multi-channel signals being intended for different speaker
configurations having different virtual sound source positions associated with their channels.
In this case, the HRTF provider 32 may be configured to provide the original HRTFs in a way
adapted to the currently intended virtual sound source positions.
The HRTF processor 34, in turn, is configured to cause the impulse responses of at least a pair
of the HRTFs to be displaced relative to each other or modify- in a spectrally varying sense -
the phase and/or magnitude responses thereof differently relative to each other. The pair of
HRTFs may model the acoustic transmission of one of left and right channels, front and rear
channels, and center and non-center channels. In effect, this may be achieved by one or a
combination of the following techniques applied to one or several channels of the multi-
channel signal, namely delaying the HRTF of a respective channel, modifying the phase
response of a respective HRTF and/or applying a decorrelation filter such as an all-pass filter
to the respective HRTF, thereby obtaining a inter-correlation reduced set of HRTFs, and/or
modifying - in a spectrally modifying sense - the magnitude response of a respective HRTF,
thereby obtaining an, at least, inter-similarity reduced set of HRTFs. In either case, the
resulting decorrelation/dissimilarity between the respective channels may support the human
auditory system in externally localizing the sound source and thereby prevent in-the-head
localization from occurring. For example, the HRTF processor 34 could be configured such
that same causes a modification of the phase response of all of, or of one or several of, the
channels HRTFs such that a group delay of a first HRTF for a certain frequency band is
introduced - or a certain frequency band of a first HRTF is delayed - relative to another one
of the HRTFs by at least one sample. Further, the HRTF processor 34 could be configured
such that same causes the modification of the phase response such that the group delays of a
first HRTF relative to another one of the HRTFs for a plurality of frequency bands, show a
standard deviation of at least an eighth of a sample. The frequency bands considered could be
the Bark bands or a subset thereof or any other frequency band sub-division.

The inter-similarity decreasing set of HRTFs resulting from the HRTF processor 34 may be
used for setting the HRTFs of the directional filters 14a-14h of the device of Fig. 1, wherein
the similarity reducer 12 may be present or absent. Due to the dis-similarity property of the
modified HRTFs, the aforementioned advantages with respect to the spatial width of the
binaural output signal and the improved externalization is similarly achieved even when the
similarity reducer 12 is missing.
As already described above, the device of Fig. 1 may be accompanied by a further pass
configured to obtain room reflection and/or reverberation related contributions of the binaural
output signal based on a downmix of at least some of the input channels 18a-18d. This
alleviates the complexity posed onto the directional filters 14a- 14h. A device for generating
such room reflection and/or room reverberation related contribution of a binaural output
signal is shown in Fig. 3. The device 40 comprises the downmix generator 42 and a room
processor 44 connected in series to each other with the room processor 44 following the
downmix generator 42. Device 40 may be connected between the input of the device of Fig. 1
at which the multi-channel signal 18 is input, and the output of the binaural output signal
where the left channel contribution 46a of the room processor 44 is added to the output 22a,
and the right channel output 46b of the room processor 44 is added to the output 22b. The
downmix generator 42 forms a mono or stereo downmix 48 from the channels of the multi-
channel signal 18, and the processor 44 is configured to generate the left channel 46a and the
right channel 46b of the room reflection and/or reverberation related contributions of the
binaural signal by modeling room reflection and/or reverberation based on the mono or stereo
signal 48.
The idea underlying the room processor 44 is that the room reflection/reverberation which
occurs in, for example, a room, may be modeled in a manner transparent for the listener,
based on a downmix such as a simple sum of the channels of the multi-channel signal 18.
Since the room reflections/ reverberation occur later than sounds traveling along the direct
path or line of sight from the sound source to the ear canals, the room processor's impulse
response is representative for, and substitutes, the tail of the impulse responses of the
directional filters shown in Fig. 1. The impulse responses of the directional filters may, in
turn, be restricted to model the direct path and the reflection and attenuations occurring at the
head, ears, and shoulders of the listener, thereby enabling shortening the impulse responses of

the directional filters. Of course, the border between what is modeled by the directional filter
and what is modeled by the room processor 44 may be freely varied so that the directional
filter may, for example, also model the first room reflections/reverberation.
Figs. 4a and 4b show possible implementations for the room processor's internal structure.
According to Fig. la, the room processor 44 is fed with a mono downmix signal 48 and
comprises two reverberation filters 50a and 50b. Analogously to the directional filters, the
reverberation filters 50a and 50b may be implemented to operate in the time domain or
frequency domain. The inputs of both receive the mono downmix signal 48. The output of the
reverberation filter 50a provides the left channel contribution output 46a, whereas the
reverberation filter 50b outputs the right channel contribution signal 46b. Fig. 4b shows an
example of the internal structure of room processor 44, in the case of the room processor 44
being provided with a stereo downmix signal 48. In this case, the room processor comprises
four reverberation filters 50a-50d. The inputs of reverberation filters 50a and 50b are
connected to a first channel 48a of the stereo downmix 48, whereas the input of the
reverberation filters 50c and 50d are connected to the other channel 48b of the stereo
downmix 48. The outputs of reverberation filters 50a and 50c are connected to the input of an
adder 52a, the output of which provides the left channel contribution 46a. The output of
reverberation filters 50b and 50d are connected to inputs of a further adder 52b, the output of
which provides the right channel contribution 46b.
Although it has been described that the downmix generator 42 may simply sum the channels
of the multi-channel signal 18 - with weighing each channel equally-, this is not exactly the
case with the embodiment of Fig. 3. Rather, the downmix generator 42 of Fig. 3 is configured
to form the mono or stereo downmix 48, such that the plurality of channels contribute to the
mono or stereo downmix at a level differing among at least two channels of the multi-channel
signal 18. By this measure, certain contents of multi-channel signals such as speech or
background music which are mixed into a specific channel or specific channels o the multi-
channel signal, may be prevented from or encouraged to being subject to the room processing,
thereby avoiding a unnatural sound.
For example, the downmix generator 42 of Fig. 3 may be configured to form the mono or
stereo downmix 48 such that a center channel of the plurality of channels of the multi-channel
signal 18 contributes to the mono or stereo downmix signal 48 in a level-reduced manner

relative to the other channels of the multi-channel signal 18. For example, the amount of level
reduction may be between 3 dB and 12 dB. The level reduction may be evenly spread over the
effective spectral range of the channels of the multi-channel signal 18, or may be frequency
dependent such as concentrated on a specific spectral portion, such as the spectral portion
typically occupied by voice signals. The amount of level reduction relative to the other
channels may be the same for all other channels. That is, the other channels may be mixed
into the downmix signal 48 at the same level. Alternatively, the other channels may be mixed
into the downmix signal 48 at an unequal level. Then, the amount of level reduction relative
to the other channels may be measured against the mean value of the other channels or the
mean value of all channels including the reduced-one. If so, the standard deviation of the
mixing weights of the other channels or the standard deviation of the mixing weights of all
channels may be smaller than 66% of the level reduction of the mixing weight of the level-
reduced channel relative to the just-mentioned mean value.
The effect of the level reduction with respect to the center channel is that the binaural output
signal obtained via contributions 56a and 56b is - at least in some circumstances which are
discussed in more detail below - more naturally perceived by listeners than without the level
reduction. In other words, the downmix generator 42 forms a weighted sum of the channels of
the channels of the multi-channel signal 18, with the weighting value associated with the
center channel being reduced relative to the weighting values of the other channels.
The level reduction of the center channel is especially advantageous during voice portions of
movie dialogs or music. The audio impression improvement obtained during these voice
portions over-compensates minor penalties due to the level reduction in non-voice phases.
However, according to an alternative embodiment, the level reduction is not constant. Rather,
the downmix generator 42 may be configured to switch between a mode where the level
reduction is switched off, and a mode where the level reduction is switched on. In other
words, the downmix generator 42 may be configured to vary the amount of level reduction in
a time-varying manner. The variation may be of a binary or analogous nature, between zero
and a maximum value. The downmix generator 42 may be configured to perform the mode
switching or level reduction amount variation dependent on information contained within the
multi-channel signal 18. For example, the downmix generator 42 may be configured to detect
voice phases or distinguish these voice phases from non-voice phases, or may assign a voice
content measure measuring the voice content, being of at least ordinal scale, to consecutive

frames of the center channel. For example, the downmix generator 42 detects the presence of
voice in the center channel by means of a voice filter and determines as to whether the output
level of this filter exceeds the sum threshold. However, the detection of voice phases within
the center channel by the downmix generator 42 is not the only way to make the afore-
mentioned mode switching of level reduction amount variation time-dependent. For example,
the multi-channel signal 18 could have side information associated therewith, which is
especially intended for distinguishing between voice phases and non-voice phases, or
measuring the voice content quantitatively. In this case, the downmix generator 42 would
operate responsive to this side information. Another probability would be that the downmix
generator 42 performs the aforementioned mode switching or level reduction amount
variations dependent on a comparison between, for example, the current levels of the center
channel, the left channel, and the right channel. In case the center channel is greater than the
left and right channels, either individually or relative to the sum thereof, by more than a
certain threshold ratio, then the downmix generator 42 may assume that a voice phase is
currently present and act accordingly, i.e. by performing the level reduction. Similarly, the
downmix generator 42 may use the level differences between the center, left and right
channels in order to realize the abovementioned dependences.
Besides this, the downmix generator 42 may be responsive to spatial parameters used to
describe the spatial image of the multiple channels of the multi-channel signal 18. This is
shown in Fig. 5. Fig. 5 shows an example of the downmix generator 42 in case the multi-
channel signal 18 represents a plurality of channels by use of special audio coding, i.e. by
using a downmix signal 62 into which the plurality of channels have been downmixed and
spatial parameters 64 describing the spatial image of the plurality of channels. Optionally, the
multi-channel signal 18 may also comprise downmixing information describing the ratios by
which the individual channels have been mixed into the downmix signal 62, or the individual
channels of the downmix signal 62, as the downmix channel 62 may for example be a normal
downmix signal 62 or a stereo downmix signal 62. The downmix generator 42 of Fig. 5
comprises a decoder 64 and a mixer 66. The decoder 64 decodes, according to spatial audio
decoding, the multi-channel signal 18 in order to obtain the plurality of channels including,
inter alia, the center channel 66, and other channels 68. The mixer 66 is configured to mix the
center channel 66 and the other non-center channels 68 to derive the mono or stereo signal 48
by performing the afore-mentioned level reduction. As indicated by the dashed line 70, the
mixer 66 may be configured to use the spatial parameter 64 in order to switch between the

level reduction mode and the non-level reduction mode of the varied amount of level
reduction, as mentioned above. The spatial parameter 64 used by the mixer 66 may, for
example, be channel prediction coefficients describing how the center channel 66, a left
channel or the right channel may be derived from the downmix signal 62, wherein mixer 66
may additionally use inter-channel coherence/cross-correlation parameters representing the
coherence or cross-correlation between the just-mentioned left and right channels which, in
turn, may be downmixes of front left and rear left channels, and front right and rear right
channels, respectively. For example, the center channel may be mixed at a fixed ratio into the
afore-mentioned left channel and the right channel of the stereo downmix signal 62. In this
case, two channel prediction coefficients are sufficient in order to determine how the center,
left, and right channels may be derived from a respective linear combination of the two
channels of the stereo downmix signal 62. For example, the mixer 66 may use a ratio between
a sum and a difference of the channel prediction coefficients in order to differentiate between
voice phases and non-voice phases.
Although level reduction with respect to the center channel has been described in order to
exemplify the weighted summation of the plurality of channels such that same contribute to
the mono or stereo downmix at a level differing among at least two channels of the multi-
channel signal 18, there are also other examples where other channels are advantageously
level-reduced or level-amplified relative to another channel or other channels because some
sound source content present in this or these channels is/are to, or is/are not to, be subject to
the room processing at the same level as other contents in the multi-channel signal but at a
reduced/increased level.
Fig. 5 was rather generally explained with respect to a possibility for representing the
plurality of input channels by means of a downmix signal 62 and spatial parameters 64. With
respect to Fig. 6, this description is intensified. The description with respect to Fig. 6 is also
used for the understanding the following embodiments described with respect to Figs. 10 to
13. Fig. 6 shows the downmix signal 62 spectrally decomposed into a plurality of subbands
82. In Fig. 6, the subbands 82 are exemplarily shown as extending horizontally with the
subbands 82 being arranged with the subband frequency increasing from bottom to top as
indicated by frequency domain arrow 84. The extension along the horizontal direction shall
denote the time axis 86. For example, the downmix signal 62 comprises a sequence of
spectral values 88 per subband 82. The time resolution at which the subbands 82 are sampled

by the sample values 88 may be defined by filterbank slots 90. Thus, the time slots 90 and
subbands 82 define some time/frequency resolution or grid. A coarser time/frequency grid is
defined by uniting neighboring sample values 88 to time/frequency tiles 92 as indicated by the
dashed lines in Fig. 6, these tiles defining the time/frequency parameter resolution or grid.
The aforementioned spatial parameters 62 are defined in that time/frequency parameter
resolution 92. The time/frequency parameter resolution 92 may change in time. To this end,
the multi-channel signal 62 may be divided-up into consecutive frames 94. For each frame,
the time/frequency resolution grid 92 is able to be set individually. In case the decoder 64
receives the downmix signal 62 in the time domain, decoder 64 may comprise of an internal
analysis filterbank in order to derive the representation of the downmix signal 62 as shown in
Fig. 6. Alternatively, downmix signal 62 enters the decoder 64 in the form as shown in Fig. 6,
in which case no analysis filterbank is necessary in decoder 64. As was already been
mentioned in Fig. 5, for each tile 92 two channel prediction coefficients may be present
revealing how, with respect to the respective time/frequency tile 92, the right and left
channels may be derived from the left and right channels of the stereo downmix signal 62. In
addition, an inter-channel coherence/cross-correlation (ICC) parameter may be present for tile
92 indicating the ICC similarities between the left and right channel to be derived from the
stereo downmix signal 62, wherein one channel has been completely mixed into one channel
of the stereo downmix signal 62, while the other has completely been mixed into the other
channel of the stereo downmix signal 62. However, a channel level difference (CLD)
parameter may further be present for each tile 92 indicating the level difference between the
just-mentioned left and right channels. A non-uniform quantization on a logarithmic scale
may be applied to the CLD parameters, where the quantization has a high accuracy close to
zero dB and a coarser resolution when there is a large difference in level between the
channels. In addition, further parameters may be present within spatial parameter 64. These
parameters may, inter alia, define CLD and ICC relating to the channels which served for
forming, by mixing, the just-mentioned left and right channels, such as rear left, front left,
rear right, and front right channels.
It should be noted that the aforementioned embodiments may be combined with each other.
Some combination possibilities have already been mentioned above. Further possibilities will
be mentioned in the following with respect to the embodiments of Figs. 7 to 13. In addition,
the aforementioned embodiments of Figs. 1 and 5 assumed that the intermediate channels 20,
66, and 68, respectively, are actually present within the device. However, this is not

necessarily the case. For example, the modified HRTFs as derived by the device of Fig. 2 may
be used to define the directional filters of Fig. 1 by leaving out the similarity reducer 12, and
in this case, the device of Fig. 1 may operate on a downmix signal such as the downmix signal
62 shown in Fig. 5, representing the plurality of channels 18a-18d, by suitably combining the
spatial parameters and the modified HRTFs in the time/frequency parameter resolution 92,
and applying accordingly obtained linear combination coefficients in order to form binaural
signals 22a and 22b.
Similarly, downmix generator 42 may be configured to suitably combine the spatial
parameters 64 and the level reduction amount to be achieved for the center channel in order to
derive the mono or stereo downmix 48 intended for the room processor 44. Fig. 7 shows a
binaural output signal generator according to an embodiment. A generator which is generally
indicated with reference sign 100 comprises a multi-channel decoder 102, a binaural output
104, and two paths extending between the output of the multi-channel decoder 102 and the
binaural output 104, respectively, namely a direct path 106 and a reverberation path 108. In
the direct path, directional filters 110 are connected to the output of multi-channel decoder
102. The direct path further comprises a first group of adders 112 and a second group of
adders 114. Adders 112 sum up the output signal of a first half of the directional filters 110
and the second adders 114 sum up the output signal of a second half of the directional filters
110. The summed up outputs of the first and second adders 112 and 114 represent the afore-
mentioned direct path contribution of the binaural output signal 22a and 22b. Adders 116 and
118 are provided in order to combine contribution signals 22a and 22b with the binaural
contribution signals provided by the reverberation path 108 i.e. signals 46a and 46b. In the
reverberation path 108, a mixer 120 and a room processor 122 are connected in series
between the output of the multi-channel decoder 102 and the respective input of adders 16
and 118, the outputs of which define the binaural output signal output at output 104.
In order to ease the understanding of the following description of the device of Fig. 7, the
reference signs used in Figs. 1 to 6 have been partially used in order to denote elements in
Fig. 7, which correspond to those, or assume responsibility for the functionality of, elements
occurring in Figs. 1 to 6. The corresponding description will become clearer in the following
description. However, it is noted that, in order to ease the following description, the following
embodiments have been described with the assumption that the similarity reducer performs a
correlation reduction. Accordingly, the latter is denoted a correlation reducer, in the

following. However, as became clear from the above, the embodiments outlined below are
readily transferable to cases where the similarity reducer performs a reduction in similarity
other than in terms of correlation. Further, the below outlined embodiments have been drafted
assuming that the mixer for generating the downmix for the room processing generates a
level-reduction of the center channel although, as described above, a transfer to alternative
embodiments would readily achievable.
The device of Fig. 7 uses a signal flow for the generation of a headphone output at output 104
from a decoded multi-channel signal 124. The decoded multi-channel 124 is derived by the
multi-channel decoder 102 from a bitstream input at a bitstream input 126, such as, for
example, by spatial audio decoding. After decoding, each signal or channel of the decoded
multi-channel signal 124 is filtered by a pair of directional filters 110. For example, the first
(upper) channel of the decoded multi-channel signal 124 is filtered by directional filters 20
DirFilter(l,L) and DirFilter(l,R), and a second (second from the top) signal or channel is
filtered by directional filter DirFilter(2,L) and DirFilter(2,R), and so on. These filters 110 may
model the acoustical transmission from a virtual sound source in a room to the ear canal of a
listener, a so-called binaural room transfer function (BRTF). They may perform time, level,
and spectral modifications, and may partially also model room reflection and reverberation.
The directional filters 110 may be implemented in time or frequency domains. Since there are
many filters 110 required (Nx2, with N being the number of decoded channels), these
directional filters could, if they should model the room reflection and the reverberation
completely, be rather long, i.e. 20000 filter taps at 44.1 kHz, in which case the process of
filtering would be computationally demanding. The directional filter 110 are advantageously
reduced to the minimum, the so-called head-related transfer functions (HRTFs) and the
common processing block 122 is used the model the room reflections and reverberations. The
room processing module 122 can implement a reverberation algorithm in a time or frequency
domain and may operate from a one or two-channel input signal 48, which is calculated from
the decoded multi-channel input signal 124 by a mixing matrix within mixer 120. The room
processing block implements room reflections and/or reverberation. Room reflections and
reverberation are essential to localize sounds, especially with respect to the distance and
externalization - meaning sounds are perceived outside the listener's head.
Typically, multi-channel sound is produced such that the dominating sound energy is
contained in the front channels, i.e. left front, right front, center. Voices in movie dialogs and

music are typically mixed mainly to the center channel. If center channel signals are fed to the
room processing module 122, the resulting output is often perceived unnaturally reverberant
and spectrally unequal. Therefore, according to the embodiment of Fig. 7, the center channel
is fed to the room processing module 122 with a significant level reduction, such as attenuated
by 6 dB, which level reduction is performed, as already denoted above, within mixer 120.
Insofar, the embodiment of Fig. 7 comprises a configuration according to Figs. 3 and 5,
wherein reference signs 102, 124,120, and 122 of Fig. 7 correspond to reference signs 18, 64,
the combination of reference signs 66 and 68, reference sign 66 and reference sign 44 of Figs.
3 and 5, respectively.
Fig. 8 shows another binaural output signal generator according to a further embodiment. The
generator is generally indicated with reference sign 140. In order to ease the description of
Fig. 8, the same reference signs have been used as in Fig. 7. In order to denote that mixer 120
does not necessarily have the functionality as indicated with the embodiments of Figs. 3, 5
and 7, namely performing the level reduction with respect to the center channel, the reference
sign 40' has been used in order to denote the arrangement of blocks 102, 120, and 122,
respectively. In other words, the level reduction within mixer 122 is optional in case of Fig. 8.
Differing from Fig. 7, however, decorrelators are connected between each pair of directional
filters 110 and the output of decoder 102 for the associated channel of the decoded multi-
channel signal 124, respectively. The decorrelators are indicated with reference signs 1421,
1422, and so on. The decorrelators 1421-1424 act as the correlation reducer 12 indicated in Fig.
1. Although shown in Fig. 8, it is not necessary that a decorrelator 1421-1424 is provided for
each of the channels of the decoded multi-channel signal 124. Rather, one decorrelator would
be sufficient. The decorrelators 142 could simply be a delay. Preferably, the amount of delay
caused by each of the delays 1421-1424 would be different to each other. Another possibility
would be that the decorrelators 1421-1424 are all-pass filters, i.e. filters having a transfer
function of a magnitude of constantly being one with, however, changing the phases of the
spectral components of the respective channel. The phase modifications caused by the
decorrelators 1421-1424 would preferably be different for each of the channels. Other
possibilities would of course also exist. For example, the decorrelator 1421-1424 could be
implemented as FIR filters, or the like.
Thus, according to the embodiment of Fig. 8, the elements 1421-1424,110,112, and 114 act in
accordance with the device 10 of Fig. 1.

Similarly to Fig. 8, Fig. 9 shows a variation of the binaural output signal generator of Fig. 7.
Thus, Fig. 9 is also explained below using the same reference signs as used in Fig. 7.
Similarly to the embodiment of Fig. 8, the level reduction of mixer 122 is merely optional in
the case of Fig. 9, and therefore, reference sigh 40' has been in Fig. 9 rather than '40, as was
the case in Fig. 7. The embodiment of Fig. 9 addresses the problem that significant correlation
exists between all channels in multi-channel sound productions. After processing of the multi-
channel signals with the directional filters 110, the two-channel intermediate signals of each
filter pair are added by adders 112 and 114, to form the headphone output signal at output
104. The summation of correlated output signals by adders 112 and 114 results in a greatly
reduced spatial width of the output signal at output 104, and a lack of an externalization. This
is particularly problematic for the correlation of the left and right signal and the center
channel within decoded multi-channel signal 124. According to the embodiment of Fig. 9, the
directional filters are configured to have a decorrelated output as far as possible. To this end,
the device of Fig. 9 comprises the device 30 for forming an inter-correlation decreasing set of
HRTFs to be used by the directional filters 110 on the basis of some original set of HRTFs.
As described above, device 30 may use one, or a combination of, the following techniques
with regard to the HRTFs of the directional filter pair associated with one or several channels
of the decoded multi-channel signal 124:
delay the directional filter or the respective directional filter pair such as for example by
displacing the impulse response thereof which could be done, for example, by displacing the
filter taps;
modifying the phase response of the respective directional filters; and
applying a decorrelation filter such as an all-pass filter to the respective directional filters of
the respective channel. Such an all-pass filter could be implemented as a FIR filter.
As described above, device 30 could operate responsive to the change in the loudspeaker
configuration for which the bitstream at bitstream input 126 is intended.
The embodiments of Figs. 7 to 9 concerned a decoded multi-channel signal. The following
embodiments are concerned with the parametric multi-channel decoding for headphones.

Generally speaking, spatial audio coding is a multi-channel compression technique that
exploits the perceptual inter-channel irrelevance in multi-channel audio signals to achieve
higher compression rates. This can be captured in terms of spatial cues or spatial parameters,
i.e. parameters describing the spatial image of a multi-channel audio signal. Spatial cues
typically include level/intensity differences, phase differences and measures of
correlations/coherence between channels, and can be represented in an extremely compact
manner. The concept of spatial audio coding has been adopted by MPEG resulting in the
MPEG surround standard, i.e. ISO/IEC23 003-1. Spatial parameters such as those employed in
spatial audio coding can also be employed to describe directional filters. By doing so, the step
of decoding spatial audio data and applying directional filters can be combined to efficiently
decode and render multi-channel audio for headphone reproduction.
The general structure of a spatial audio decoder for headphone output is given in Fig. 10. The
decoder of Fig. 10 is generally indicated with reference sign 200, and comprises a binaural
spatial subband modifier 202 comprising an input for a stereo or mono downmix signal 204,
another input for spatial parameters 206, and an output for the binaural output signal 208. The
downmix signal along with the spatial parameters 206 form the afore-mentioned multi-
channel signal 18 and represent the plurality of channels thereof.
Internally, the subband modifier 202 comprises an analysis filterbank 208, a matrixing unit or
linear combiner 210 and a synthesis filterbank 212 connected in the order mentioned between
the downmix signal input and the output of subband modifier 202. Further, the subband
modifier 202 comprises a parameter converter 214 which is fed by the spatial parameters 206
and a modified set of HRTFs as obtained by device 30.
In Fig. 10, the downmix signal is assumed to have already been decoded beforehand,
including for example, entropy encoding. The binaural spatial audio decoder is fed with the
downmix signal 204. The parameter converter 214 uses the spatial parameters 206 and
parametric description of the directional filters in the form of the modified HRTF parameter
216 to form binaural parameters 218. These parameters 218 are applied by matrixing unit 210
in from of a two-by-two matrix (in case of a stereo downmix signal) and in form of a one-by-
two matrix (in case of a mono downmix signal 204), in frequency domain, to the spectral
values 88 output by analysis filterbank 208 (see Fig. 6). In other words, the binaural
parameters 218 vary in the time/frequency parameter resolution 92 shown in Fig. 6 and are

applied to each sample value 88. Interpolation may be used to smooth the matrix coefficients
and the binaural parameters 218, respectively, from the coarser time/frequency parameter
domain 92 to the time/frequency resolution of the analysis filterbank 208. That is, in the case
of a stereo downmix 204, the matrixing performed by unit 210 results in two sample values
per pair of sample value of the left channel of the downmix signal 204 and the corresponding
sample value of the right channel of the downmix signal 204. The resulting two sample values
are part of the left and right channels of the binaural output signal 208, respectively. In case of
a mono downmix signal 204, the matrixing by unit 210 results in two sample values per
sample value of the mono downmix signal 204, namely one for the left channel and one for
the right channel of the binaural output signal 208. The binaural parameters 218 define the
matrix operation leading from the one or two sample values of the downmix signal 204 to the
respective left and right channel sample values of the binaural output signal 208. The binaural
parameters 218 already reflect the modified HRTF parameters. Thus, they decorrelate the
input channels of the multi-channel signal 18 as indicated above.
Thus, the output of the matrixing unit 210 is a modified spectrogram as shown in Fig. 6. The
synthesis filterbank 212 reconstructs therefrom the binaural output signal 208. In other words,
the synthesis filterbank 212 converts the resulting two channel signal output by the matrixing
unit 210 into the time domain. This is, of course, optional.
In case of Fig. 10, the room reflection and reverberation effects were not addressed separately.
If ever, these effects have to be taken into account in the HRTFs 216. Fig. 11 shows a
binaural output signal generator combining a binaural spatial audio decoder 200' with
separate room reflection/reverberation processing. The ' of reference sign 200' in Fig. 11 shall
denote that the binaural spatial audio decoder 200' of Fig. 11 may use unmodified HRTFs, i.e.
the original HRTFs as indicated in Fig. 2. Optionally, however, the binaural spatial audio
decoder 200' of Fig. 11 may be the one shown in Fig. 10. In any case, the binaural output
signal generator of Fig. 11 which is generally indicated with reference sign 230, comprises
besides the binaural spatial decoder 200', a downmix audio decoder 232, a modified spatial
audio subband modifier 234, a room processor 122, and two adders 116 and 118. The
downmix audio decoder 232 is connected between a bitstream input 126 and a binaural spatial
audio subband modifier 202 of the binaural spatial audio decoder 200'. The downmix audio
decoder 232 is configured to decode the bit stream input at input 126 to derive the downmix
signal 214 and the spatial parameters 206. Both, the binaural spatial audio subband modifier

202, as well as the modified spatial audio subband modifier 234 is provided with a downmix
signal 204 in addition to the spatial parameters 206. The modified spatial audio subband
modifier 234 computes from the downmix signal 204 - by use of the spatial parameters 206 as
well as modified parameters 236 reflecting the aforementioned amount of level reduction of
the center channel - the mono or stereo downmix 48 serving as an input for room processor
122. The contributions output by both the binaural spatial audio subband modifier 202 and the
room processor 122, respectively, are channel-wise summed in adders 116 and 118 to result in
the binaural output signal at output 238.
Fig. 12 shows a block diagram illustrating the functionality of the binaural audio decoder 200'
of Fig. 11. It should be noted that Fig. 12 does not show the actual internal structure of the
binaural spatial audio decoder 200' of Fig. 11, but illustrates the signal modifications obtained
by the binaural spatial audio decoder 200'. It is recalled that the internal structure of the
binaural spatial audio decoder 200' generally complies with the structure shown in Fig. 10,
with the exception that the device 30 may be left away in the case that same is operating with
the original HRTFs. Additionally, Fig. 12 shows the functionality of the binaural spatial audio
decoder 200' exemplarily for the case that only three channels represented by the multi-
channel signal 18 are used by the binaural spatial audio decoder 200' in order to form the
binaural output signal 208. In particular, a "2 to 3", i.e. TTT, box is used to derive a center
channel 242, a right channel 244, and a left channel 246 from the two channels of the stereo
downmix 204. In other words, Fig. 12 exemplarily assumes that the downmix 204 is a stereo
downmix. The spatial parameters 206 used by the TTT box 248 comprise the above-
mentioned channel prediction coefficients. The correlation reduction is achieved by three
decorrelators, denoted DelayL, DelayR, and DelayC in Fig. 12. They correspond to the
decorrelation introduced in case of, for example, Figs. 1 and 7. However, it is again recalled
that Fig. 12 merely shows the signal modifications achieved by the binaural spatial audio
decoder 200', although the actual structure corresponds to that shown in Fig. 10. Thus,
although the delays forming the correlation reducer 12 are shown as separate features relative
to the HRTFs forming the directional filters 14, the existence of the delays in the correlation
reducer 12 may be seen as a modification of the HRTF parameters forming the original
HRTFs of the directional filters 14 of Fig. 12. First, Fig. 12 merely shows that the binaural
spatial audio decoder 200' decorrelates the channels for headphone reproduction. The
decorrelation is achieved by simple means, namely, by adding a delay block in the parametric
processing for the matrix M and the binaural spatial audio decoder 200'. Thus, the binaural

spatial audio decoder 200' may apply the following modifications to the individual channels,
namely
delaying the center channel preferably at least one sample,
delaying the center channel by different intervals in each frequency band,
delaying left and right channels preferably at least one sample and/or
delaying left and right channels by different intervals in each frequency band.
Fig. 13 shows an example for a structure of the modified spatial audio subband modifier of
Fig. 11. The subband modifier 234 of Fig. 13 comprises a two-to-three or TTT box 262,
weighting stages 264a-264e, first adders 266a and 266b, second adders 268a and 268b, an
input for the stereo downmix 204, an input for the spatial parameters 206, a further input for a
residual signal 270 and an output for the downmix 48 intended for being processed by the
room processor, and being, in accordance with Fig. 13, a stereo signal.
As Fig. 13 defines in a structural sense an embodiment for the modified spatial audio subband
modifier 234, the TTT box 262 of Fig. 13 merely reconstructs the center channel, the right
channel 244, and the left channel 246 from the stereo downmix 204 by using the spatial
parameters 206. It is once again recalled that in the case of Fig. 12, the channels 242-246 are
actually not computed. Rather, the binaural spatial audio subband modifier modifies matrix M
in such a manner that the stereo downmix signal 204 is directly turned into the binaural
contribution reflecting the HRTFs. The TTT box 262 of Fig. 13, however, actually performs
the reconstruction. Optionally, as shown in Fig. 13, the TTT box 262 may use a residual
signal 270 reflecting the prediction residual when reconstructing channels 242-246 based on
the stereo downmix 204 and the spatial parameters 206, which as denoted above, comprise the
channel prediction coefficients and, optionally, the ICC values. The first adders 266a are
configured to add-up channels 242-246 to form the left channel of the stereo downmix 48. In
particular, a weighted sum is formed by adders 266a and 266b, wherein the weighting values
are defined by the weighting stages 264a, 264b, 264c, and 264e which might apply to the
respective channel 246 to 242, a respective weighting value EQLL, EQRL and EQCL. Similarly,
adders 268a and 268b form a weighted sum of channels 246 to 242 with weighting stages
264b, 264d, and 264e forming the weighting values, the weighted sum forming the right
channel of the stereo downmix 48.

The parameters 270 for the weighting stages 264a-264e are, as described above, selected such
that the above-described center channel level reduction in the stereo downmix 48 is achieved
resulting, as described above, in the advantages with respect to natural sound perception.
Thus, in other words, Fig. 13 shows a room processing module which may be applied in
combination with the binaural parametric decoder 200' of Fig. 12. In Fig. 13, the downmix
signal 204 is used to feed the module. The downmix signal 204 contains all the signals of the
multi-channel signal to be able to provide stereo compatibility. As mentioned above, it is
desirable to feed the room processing module with a signal containing only a reduced center
signal. The modified spatial audio subband modifier of Fig. 13 serves to perform this level
reduction. In particular, according to Fig. 13, a residual signal 270 may be used in order to
reconstruct the center, left and right channels 242-246. The residual signal of the center and
the left and right channels 242-246 may be decoded by the downmix audio decoder 232,
although not shown in Fig. 11. The EQ parameters or weighting values applied by the
weighting stages 264a-264e may be real-valued for the left, right, and center channels 242-
246. A single parameter set for the center channel 242 may be stored and applied, and the
center channel is, according to Fig. 13, exemplarily equally mixed to both, left and right
output of stereo downmix 48.
The EQ parameters 270 fed into the modified spatial audio subband modifier 234 may have
the following properties. Firstly, the center channel signal may be attenuated preferably by at
least 6 dB. Further, the center channel signal may have a low-pass characteristic. Even
further, the difference signal of the remaining channels may be boosted at low frequencies. In
order to compensate the lower level of the center channel 242 relative to the other channels
244 and 246, the gain of the HRTF parameters for the center channel used in the binaural
spatial audio subband modifier 202 should be increased accordingly.
The main goal of the setting of the EQ parameters is the reduction of the center channel signal
in the output for the room processing module. However, the center channel should only be
suppressed to a limited extent: the center channel signal is subtracted from the left and the
right downmix channels inside the TTT box. If the center level is reduced, artifacts in the left
and right channel may become audible. Therefore, center level reduction in the EQ stage is a
trade-off between suppression and artifacts. Finding a fixed setting of EQ parameters is
possible, but may not be optimal for all signals. Accordingly, according to an embodiment, an

adaptive algorithm or module 274 may be used to control the amount of center level reduction
by one, or a combination of the following parameters:
The spatial parameters 206 used to decode the center channel 242 from the left and right
downmix channel 204 inside the TTT box 262 may be used as indicated by dashed line 276.
The level of center, left and right channels may be used as indicated by dashed line 278.
The level differences between center, left and right channels 242-246 may be used as also
indicated by dashed line 278.
The output of a single-type detection algorithm, such as a voice activity detector, may be used
as also indicated by dashed line 278.
Lastly, static of dynamic metadata describing the audio content may be used in order to
determine the amount of center level reduction as indicated by dashed line 280.
Although some aspects have been described in the context of an apparatus, it is clear that
these aspects also represent a description of the corresponding method, wherein a block or
device corresponds to a method step or a feature of a method step. Analogously, aspects
described in the context of a method step also represent a description of a corresponding
block or item or feature of a corresponding apparatus such as a part of an ASIC, a sub-routine
of a program code or a part of a programmed programmable logic.
The inventive encoded audio signal can be stored on a digital storage medium or can be
transmitted on a transmission medium such as a wireless transmission medium or a wired
transmission medium such as the Internet.
Depending on certain implementation requirements, embodiments of the invention can be
implemented in hardware or in software. The implementation can be performed using a digital
storage medium, for example a floppy disk, a DVD, a CD, a ROM, a PROM, an EPROM, an
EEPROM or a FLASH memory, having electronically readable control signals stored thereon,
which cooperate (or are capable of cooperating) with a programmable computer system such
that the respective method is performed.

Some embodiments according to the invention comprise a data carrier having electronically
readable control signals, which are capable of cooperating with a programmable computer
system, such that one of the methods described herein is performed.
Generally, embodiments of the present invention can be implemented as a computer program
product with a program code, the program code being operative for performing one of the
methods when the computer program product runs on a computer. The program code may for
example be stored on a machine readable carrier.
Other embodiments comprise the computer program for performing one of the methods
described herein, stored on a machine readable carrier.
In other words, an embodiment of the inventive method is, therefore, a computer program
having a program code for performing one of the methods described herein, when the
computer program runs on a computer.
A further embodiment of the inventive methods is, therefore,, a data carrier (or a digital
storage medium, or a computer-readable medium) comprising, recorded thereon, the computer
program for performing one of the methods described herein.
A further embodiment of the inventive method is, therefore, a data stream or a sequence of
signals representing the computer program for performing one of the methods described
herein. The data stream or the sequence of signals may for example be configured to be
transferred via a data communication connection, for example via the Internet.
A further embodiment comprises a processing means, for example a computer, or a
programmable logic device, configured to or adapted to perform one of the methods described
herein.
A further embodiment comprises a computer having installed thereon the computer program
for performing one of the methods described herein.
In some embodiments, a programmable logic device (for example a field programmable gate
array) may be used to perform some or all of the functionalities of the methods described
herein. In some embodiments, a field programmable gate array may cooperate with a
microprocessor in order to perform one of the methods described herein. Generally, the
methods are preferably performed by any hardware apparatus.

The above described embodiments are merely illustrative for the principles of the present
invention. It is understood that modifications and variations of the arrangements and the
details described herein will be apparent to others skilled in the art. It is the intent, therefore,
to be limited only by the scope of the impending patent claims and not by the specific details
presented by way of description and explanation of the embodiments herein.

Device for generating a binaural signal based on a multi-channel signal representing a
plurality of channels and intended for reproduction by a speaker configuration having
a virtual sound source position associated to each channel, comprising:
a similarity reducer (12) for differently processing, and thereby reducing a similarity
between, at least one of a left and a right channel of the plurality of channels, a front
and a rear channel of the plurality of channels, and a center and a non-center channel
of the plurality of channels, in order to obtain an inter-similarity reduced set (20) of
channels;
a plurality (14) of directional filters for modeling an acoustic transmission of a
respective one of the inter-similarity reduced set (20) of channels from a virtual sound
source position associated with the respective channel of the inter-similarity reduced
set of channels to a respective ear canal of a listener;
a first mixer (16a) for mixing outputs of the directional filters modeling the acoustic
transmission to the first ear canal of the listener to obtain a first channel (22a) of the
binaural signal; and
a second mixer (16b) for mixing outputs of the directional filters modeling the
acoustic transmission to the second ear canal of the listener to obtain a second channel
(22b) of the binaural signal;
a downmix generator (42) for forming a mono or stereo downmix of the plurality of
channels represented by the multi-channel signal; and
a room processor (44) for generating a room-reflections/reverberation related
contribution of the binaural signal, including a first channel output and a second'
channel output, by modeling room reflections/reverberations based on the mono or
stereo signal,

a first adder (116) configured to add the first channel output of the room processor to
the first channel (22a) of the binaural signal; and
a second adder (118) configured to add the second channel output of the room
processor to the second channel (22a) of the binaural signal.
The device according to claim 1, wherein the similarity reducer (12) is configured to
perform the different processing by
causing a relative delay between, and/or performing - in a spectrally varying sense -
phase modification differently between, the at least one of the left and the right
channels of the plurality of channels, the front and the rear channels of the plurality of
channels, and the center and non-center channels of the plurality of channels, and/or
performing - in a spectrally varying sense - a magnitude modification differently
between, the at least one of the left and the right channels of the plurality of channels,
the front and the rear channels of the plurality of channels, and the center and non-
center channels of the plurality of channels.
Device for generating a binaural signal based on a multi-channel signal representing a
plurality of channels and intended for reproduction by a speaker configuration having
a virtual sound source position associated to each channel, comprising:
a similarity reducer (12) for causing a relative delay between, and/or performing - in a
spectrally varying sense - a phase and/or magnitude modification differently between
at least two channels of the plurality of channels, in order to obtain an inter-similarity *
reduced set (20) of channels;
a plurality (14) of directional filters for modeling an acoustic transmission of a
respective one of the inter-similarity reduced set (20) of channels from a virtual sound
source position associated with the respective channel of the inter- similarity reduced
set of channels to a respective ear canal of a listener;

a first mixer (16a) for mixing outputs of the directional filters modeling the acoustic
transmission to the first ear canal of the listener to obtain a first channel (22a) of the
binaural signal;
a second mixer (16b) for mixing outputs of the directional filters modeling the
acoustic transmission to the second ear canal of the listener to obtain a second channel
(22b) of the binaural signal;
a downmix generator (42) for forming a mono or stereo downmix of the plurality of
channels represented by the multi-channel signal;
a room processor (44) for generating a room-reflections/reverberation related
contribution of the binaural signal, including a first channel output and a second
channel output, by modeling room reflections/reverberations based on the mono or
stereo signal;
a first adder (116) configured to add the first channel output of the room processor to
the first channel (22a) of the binaural signal; and
a second adder (118) configured to add the second channel output of the room
processor to the second channel (22a) of the binaural signal.
Device for forming an inter-similarity decreasing set of HRTFs for modeling an
acoustic transmission of a plurality of channels from a virtual sound source position
associated with the respective channel to ear canals of a listener, the device
comprising:
an HRTF provider (32) for providing an original plurality of HRTFs implemented as
FIR filters, by looking-up or computing filter taps for each of the original plurality of
HRTFs responsive to a selection or change of the virtual sound source positions; and
an HRTF processor (34) for causing impulse responses of the HRTFs modeling the
acoustic transmissions of a predetermined pair of channels to be delayed relative to
each other, or differently modifying - in a spectrally varying sense - phase and/or

magnitude responses thereof, the pair of channels being one of a left and a right
channel of the plurality of channels, a front and a rear channel of the plurality of
channels, and a center and a non-center channel of the plurality of channels.
5. Device according to claim 4, wherein
the HRTF processor (34) is configured to cause the impulse responses of the HRTFs
modeling the acoustic transmissions of a predetermined pair of channels to be delayed
relative to each other by displacing the filter taps.
6. Device according to claim 4 or 5, wherein
the HRTF processor (34) is configured to cause the impulse responses of the HRTFs
modeling the acoustic transmissions of a predetermined pair of channels to be delayed
relative to each other, or differently modify - in a spectrally varying sense - phase
and/or magnitude responses thereof such that group delays of a first one of the HRTFs
relative to another one of the HRTFs, show, for bark bands, a standard deviation of at
least an eighth of a sample.
7. Device according to any of claims 4 to 6, wherein the HRTF provider (32) is
configured to provide the original plurality of HRTFs based on the virtual sound
source positions and HRTF parameters.
8. Device according to any of claims 4 to 7, wherein the HRTF processor (34) is
configured to differently all-pass filter the impulse responses of the predetermined pair
of channels.
9. Method for generating a binaural signal based on a multi-channel signal representing a
plurality of channels and intended for reproduction by a speaker configuration having
a virtual sound source position associated to each channel, comprising:
differently processing, and thereby reducing a correlation between, at least one of a
left and a right channel of the plurality of channels, a front and a rear channel of the

plurality of channels, and a center and a non-center channel of the plurality of
channels, in order to obtain an inter-similarity reduced set (20) of channels;
subject the inter-similarity reduced set (20) of channels to a plurality (14) of
directional filters for modeling an acoustic transmission of a respective one of the
inter-similarity reduced set (20) of channels from a virtual sound source position
associated with the respective channel of the inter-similarity reduced set of channels to
a respective ear canal of a listener;
mixing outputs of the directional filters modeling the acoustic transmission to the first
ear canal of the listener to obtain a first channel (22a) of the binaural signal;
mixing outputs of the directional filters modeling the acoustic transmission to the
second ear canal of the listener to obtain a second channel (22b) of the binaural signal;
forming a mono or stereo downmix of the plurality of channels represented by the
multi-channel signal;
generating a room-reflections/reverberation related contribution of the binaural signal,
including a first channel output and a second channel output, by modeling room
reflections/reverberations based on the mono or stereo signal,
adding the first channel output of the room processor to the first channel (22a) of the
binaural signal; and
adding the second channel output of the room processor to the second channel (22a) o£
the binaural signal.
0. Method for generating a binaural signal based on a multi-channel signal representing a
plurality of channels and intended for reproduction by a speaker configuration having
a virtual sound source position associated to each channel, comprising:

performing - in a spectrally varying sense - a phase and/or magnitude modification
differently between at least two channels of the plurality of channels, in order to
obtain an inter-similarity reduced set (20) of channels;
subject the -similarity reduced set (20) of channels to a plurality (14) of directional
filters for modeling an acoustic transmission of a respective one of the inter-similarity
reduced set (20) of channels from a virtual sound source position associated with the
respective channel of the inter- similarity reduced set of channels to a respective ear
canal of a listener;
mixing outputs of the directional filters modeling the acoustic transmission to the first
ear canal of the listener to obtain a first channel (22a) of the binaural signal; and
mixing outputs of the directional filters modeling the acoustic transmission to the
second ear canal of the listener to obtain a second channel (22b) of the binaural signal;
forming a mono or stereo downmix of the plurality of channels represented by the
multi-channel signal;
generating a room-reflections/reverberation related contribution of the binaural signal,
including a first channel output and a second channel output, by modeling room
reflections/reverberations based on the mono or stereo signal,
adding the first channel output of the room processor to the first channel (22a) of the
binaural signal; and
adding the second channel output of the room processor to the second channel (22a) of
the binaural signal.
Method for forming an inter-similarity decreasing set of head-related transfer
functions for modeling an acoustic transmission of a plurality of channels from a
virtual sound source position associated with the respective channel to ear canals of a
listener, the method comprising:

providing an original plurality of HRTFs implemented as FIR filters, by looking-up or
computing filter taps for each of the original plurality of HRTFs responsive to a
selection or change of the virtual sound source positions; and
differently modifying - in a spectrally varying sense - phase and/or magnitude
responses of impulse responses of the HRTFs modeling the acoustic transmissions of a
predetermined pair of channels such that group delays of a first one of the HRTFs
relative to another one of the HRTFs, show, for bark bands, a standard deviation of at
least an eighth of a sample, the pair of channels being one of a left and a right channel
of the plurality of channels, a front and a rear channel of the plurality of channels, and
a center and a non-center channel of the plurality of channels.
Computer program having instructions for performing, when running on a computer, a
method according to any of claims 9 to 11.

A device for generating a binaural signal based on a multi-channel signal representing a
plurality of channels and intended for reproduction by a speaker configuration having a
virtual sound source position associated to each channel, is described. It comprises a
correlation reducer for differently processing, and thereby reducing a correlation between, at
least one of a left and a right channel of the plurality of channels, a front and a rear channel of
the plurality of channels, and a center and a non-center channel of the plurality of channels, in
order to obtain an inter-similarity reduced set of channels; a plurality of directional filters, a
first mixer for mixing outputs of the directional filters modeling the acoustic transmission to
the first ear canal of the listener, and a second mixer for mixing outputs of the directional
filters modeling the acoustic transmission to the second ear canal of the listener. According to
another aspect, a center level reduction for forming the downmix for a room processor is
performed. According to even another aspect, an inter-similarity decreasing set of head-
related transfer functions is formed.