Remark:
The attached paper titled
"Cross-frequency interactions in the precedence effect" by Barbara Shinn-Cunningham
may serve to emphasize, that precedence does not depend on "room reflections identical in time domain".
Giving up this notion, all the drawbacks of highly correlated room reflections - especially the early ones - become obvious.
Having highly correlated reflections in a listening room is synonymic to an inferior sound field, which hinders enjoyful stereo listening instead of supporting it.
The attached paper titled
"Cross-frequency interactions in the precedence effect" by Barbara Shinn-Cunningham
may serve to emphasize, that precedence does not depend on "room reflections identical in time domain".
Giving up this notion, all the drawbacks of highly correlated room reflections - especially the early ones - become obvious.
Having highly correlated reflections in a listening room is synonymic to an inferior sound field, which hinders enjoyful stereo listening instead of supporting it.
Happy new year, Oliver! 
My hypothesis is that the precedence effect relies on time domain information rather than on the frequency domain information.
This is simply because in the time frame of the effect that is in the first millisecond after the arrival of the first wavefront at the listener's ear there is yet no established frequency domain information - the physical sound source still goes only through the phase of fighting inertia in its own specific way.
Therefore for an early reflection to be sorted out by our hearing with the help of the precedence effect it is not the energy spectrum of the reflections that should resemble that of the direct sound. It must be - or so it seems - the shape of the first wavefront - because it is all there is in the time frame of the precedence effect.
My hypothesis is that the precedence effect relies on time domain information rather than on the frequency domain information.
This is simply because in the time frame of the effect that is in the first millisecond after the arrival of the first wavefront at the listener's ear there is yet no established frequency domain information - the physical sound source still goes only through the phase of fighting inertia in its own specific way.
Therefore for an early reflection to be sorted out by our hearing with the help of the precedence effect it is not the energy spectrum of the reflections that should resemble that of the direct sound. It must be - or so it seems - the shape of the first wavefront - because it is all there is in the time frame of the precedence effect.
Hi Oliver,
I tend to disagree.
To me this appears as a better and more logical requirement than yours:
Rudolf
Hi Rudolf,
Happy new year to you too !
I was not denying phase coherence playing an important role in the direct sound, just to prevent misunderstandings.
Direct sound is highly phase correlated in harmonics. Reflections are typically delayed and "scatte
red".
It is simply not true - to avoid the word "ridicoulus" - that humans need "plane mirror shaped walls" for the precedence effect to work.
The precedence effect in fact works even or even better with scattered reflections from a stone grot, which we left just a minute ago in cultural evolution, as you might remember.
Rudolf and graaf, my impression is that your opinions are influenced by "HiFi- thinking" and that those opinions are surely not in line with data from psychoacoustics and also some generic principles our brain uses in achieving
- pattern recogniton
- source-separation tasks
Kind Regrards
Last edited:
interesting, thanks! I must read though this.
Are those conclusions of the study?
I am sursprised. If it's true then perhaps the "cocktail party effect" does not rely on the "precedence effect" understood as something that happens during the first millisecond of the proces of hearing?
This is new to me.
I thought that it's rather the high interaural cross corellation that lead to and inferior sound field.
I have to read though the papers. Thanks for them!
well, this is not what I am saying
Did the study by Ms Shinn-Cunningham cover this question?
Well, there is such possibility
but I really like to learn
Last edited:
I am not saying that "I am presenting data". Just a hypothesis based on established science. No need to be rude, really.
All I am saying is that, let me repeat - I am sursprised. If it's true then perhaps the "cocktail party effect" does not rely on the "precedence effect" understood as something that happens during the first millisecond of the proces of hearing?
graaf said:
The conclusions due to HiFi and stereo are my own ones. But the "cross frequency ..." studies show, that precedence does not rely on "time domain" identity or even "close similarity" between direct and reflected sound.
BTW: It is hard to argue with "time frames" in many respects of auditive pattern recocnition as we should consider, that "data evaluation" in the higher domains of perception may be
- time delayed considerably
- and has access to acoustical short term memory, able to evaluate "signals and parameters" over longer time frames and/or with considarable delay
Kind Regards
Due to e.g. Prof. Dr. Wolf Singer
http://de.wikipedia.org/wiki/Wolf_Singer
HNF - Singer
correlation is the structuring principle for neural activity in representing patterns and also forming consciousness.
Separating correlated from uncorrelated input is in my view what the brain "does with high profession" and in many ways.
It is - to me - contradictory to the so far accepted generic functional principles of the brain, that a source separation task gets "easier", when highly correlated versions of the source pattern are floating around in the environment.
In fact the brain can manage even those situations, but the "natural and easy way" is always separating patterns of high correlation from those of lower correlation.
The notion precedence effect needing perfect copies of the "source pattern" is advocated by some persons having high influence in the HiFi scene:
E.g. Siegfried Linkwitz was operating with that "copy" nature of reflections needed for reflections to be "sorted out" by the earbrain as such in some of his lectures.
Unfortunately a large share of his audience seems to take (not "understand") that "copy" notion "literally": I do not even know, whether this was intended by S.L. .
But in my view by doing so large damage has occured due to misunderstanding the role of reflections and the preferred properties of loudspeaker/room interaction:
Many people now think "reflections should exactly resemble all aspects of the original signal" otherwise the earbrain cannot "sort them out": That is a complete nonsense of course, but it is easy (and fast) to communicate especially among "electrical engineering knowledge only, not interested in anything else" kind of designers and hobbyists, who actually provide the majority of "communicated opinions" by sheer number in the HiFi scene.
In my view this "reflection should be copy" notion also leads loudspeaker design - especially for small rooms - far astray, thereby possibly hampering progress (since and) even for (further) decades.
I strongly recommend reading studies in this field presenting credible data concerning effects in auditive pattern recognition.
Ad hoc assumptions from people without expertise in this field do not lead to better understanding of matters.
Unfortunately simplifications usually spread a lot wider and faster than more complex data and conclusions that seem more "multilayered".
Kind Regards
http://de.wikipedia.org/wiki/Wolf_Singer
HNF - Singer
correlation is the structuring principle for neural activity in representing patterns and also forming consciousness.
Separating correlated from uncorrelated input is in my view what the brain "does with high profession" and in many ways.
It is - to me - contradictory to the so far accepted generic functional principles of the brain, that a source separation task gets "easier", when highly correlated versions of the source pattern are floating around in the environment.
In fact the brain can manage even those situations, but the "natural and easy way" is always separating patterns of high correlation from those of lower correlation.
The notion precedence effect needing perfect copies of the "source pattern" is advocated by some persons having high influence in the HiFi scene:
E.g. Siegfried Linkwitz was operating with that "copy" nature of reflections needed for reflections to be "sorted out" by the earbrain as such in some of his lectures.
Unfortunately a large share of his audience seems to take (not "understand") that "copy" notion "literally": I do not even know, whether this was intended by S.L. .
But in my view by doing so large damage has occured due to misunderstanding the role of reflections and the preferred properties of loudspeaker/room interaction:
Many people now think "reflections should exactly resemble all aspects of the original signal" otherwise the earbrain cannot "sort them out": That is a complete nonsense of course, but it is easy (and fast) to communicate especially among "electrical engineering knowledge only, not interested in anything else" kind of designers and hobbyists, who actually provide the majority of "communicated opinions" by sheer number in the HiFi scene.
In my view this "reflection should be copy" notion also leads loudspeaker design - especially for small rooms - far astray, thereby possibly hampering progress (since and) even for (further) decades.
I strongly recommend reading studies in this field presenting credible data concerning effects in auditive pattern recognition.
Ad hoc assumptions from people without expertise in this field do not lead to better understanding of matters.
Unfortunately simplifications usually spread a lot wider and faster than more complex data and conclusions that seem more "multilayered".
Kind Regards
Last edited:
graaf said:
When we observe envelopes e.g. in narrowbanded frequency ranges, then we will find correlations (also interband correlations) within the direct sound and possibly even between direct sound and reflected sound.
But interband correlations will not be the same in direct and reflected sound, at least when the reflecting surface is e.g. a "naturally grown wall of rock" or even a man made diffuser.
http://static.panoramio.com/photos/large/1073108.jpg
https://www.gearslutz.com/board/att...70153732-qrd-diffuser-diy-sding-0094-copy.jpg
Last edited:
Happy new year to you too, Oliver
the argument in my Griesinger citation is clearly post-Cunningham and should be very well in line with Singer. Apart from that his argument is founded more on neuroacoustics than on psychoacoustics.
It seems improbable that you had the time to look into my Griesinger link before you answered my post. At least your answer suggests that.
If you are interested in a further discussion, this thread wouldn't be the best place.
Rudolf
Ok, then what should be those preferred properties in Your opinion?
"Exactly al aspects"? Well, that's certainly not me. I hypothesise instead that a certain degree of similarity in the time domain is required.
Ok, then what direction should it have in Your opinion?
Hi Oliver,
I tend to disagree.
To me this appears as a better and more logical requirement than yours:
Rudolf
very interesting, thanks for the link
Hi graaf,
regarding room reflections:
Diffuse reflections are to be preferred.
Pictures 11 and 12 (pages 14 and 15) in the attached PDF show the interaction of a non phase coherent loudspeaker with a smooth wall (page 15: bottom colored diagram), as it is found in many living rooms near the loudspeakers, compared to a conventional coherent loudspeaker (page 15: top colored diagram).
The interference artifacts that come with early reflections are detrimental to sound quality.
E.g. a DML type loudspeaker - like the one measured - is able to provide diffuse reflections having less interference with the direct sound, because of phase decorrelated radiation.
However most usual DML also have drawbacks, as their on axis behaviour is often neither flat in frequency response nor phase coherent too: But that's what we want to have.
Ideal would be a combination of on axis and near axis coherence and flatness, combined with a phase decorrelation at larger off axis angles.
"Traditional directivity" should also apply in combination and DI should be in same order like in dipoles and cardioids on average. Furthermore DI should be fairly constant with frequency, yielding a considerably flat and smooth - also due to narrow banded structure - energy response.
I think thats it - for the first time being and in simplified manner - to make a "living room adjusted" loudspeaker.
Kind Regards
regarding room reflections:
Diffuse reflections are to be preferred.
Pictures 11 and 12 (pages 14 and 15) in the attached PDF show the interaction of a non phase coherent loudspeaker with a smooth wall (page 15: bottom colored diagram), as it is found in many living rooms near the loudspeakers, compared to a conventional coherent loudspeaker (page 15: top colored diagram).
The interference artifacts that come with early reflections are detrimental to sound quality.
E.g. a DML type loudspeaker - like the one measured - is able to provide diffuse reflections having less interference with the direct sound, because of phase decorrelated radiation.
However most usual DML also have drawbacks, as their on axis behaviour is often neither flat in frequency response nor phase coherent too: But that's what we want to have.
Ideal would be a combination of on axis and near axis coherence and flatness, combined with a phase decorrelation at larger off axis angles.
"Traditional directivity" should also apply in combination and DI should be in same order like in dipoles and cardioids on average. Furthermore DI should be fairly constant with frequency, yielding a considerably flat and smooth - also due to narrow banded structure - energy response.
I think thats it - for the first time being and in simplified manner - to make a "living room adjusted" loudspeaker.
Kind Regards
Last edited:
Some unsorted thoughts ...
Concerning "off axis smoothness" of the frequency response of a loudspeaker:
What could be meant with "off axis smoothness" regarding the acoustic reality of a living room e.g. ?
No single reflections - e.g. from a booshelf placed at a side wall - will be "smooth" in spectrum. Even if the loudspeaker was radiating a "smooth spectrum" towards the shelf, in an appropriate angle to be reflected to the listening seat, that will not yield a reflection being "smooth" in spectrum: The spectrum of a single reflection will be "scattered" instead.
So "off axis smoothness" may IMO just refer to a smooth energy spectrum as a whole in a statistically defined sense.
- it does not mean, the loudspaker ought to have "as flat as on axis" response at all (also larger off axis ...) angles
- it does not mean the loudspeaker has to be phase coherent at all outside the "main listening window" relevant for the loudspaker's direct sound.
Concerning "floor coupled upfiring" position, which is the topic of this thread:
"Floor coupled upfiring" means: Edge position of the loudspeaker in the room.
Idealized edge and especially corner positions reduce the number of mirror sources in proximity to the loudspeakers.
When using monopole sources , this is an option to generate more coherent wavefronts as would be possible using monopole soures having distances to the walls in the order of low frequency wavelengths and fractions of those.
It is reasonable, that by this "room constructed directivity" an advantage in coherence especially from above the Schroeder Frequency until midrange can be achieved. That frequency range say from 200Hz to 800Hz or so is most important in localization because of
- the stereo system working well here (above about 2Khz does not give very proper cues depending on signal shape)
- there is spectral dominance in localization cues
But the "edge or corner" placed speaker has potential difficulties or drawbacks also:
- in providing an adjusted stereo triangle to the listener in every room's situation. In fact this is hard enough, with speakers that can be placed rather freely ...
- in having a - positive or negative - angle of elevation, exciting HRTF cues regarding that angle
- real edge and corner positions may not be ideal at higher frequencies, where practical distances of sound sources become relevant for interference, as real sound sources (drivers, tweeters) have non zero distances to the adjacent walls and the floor
- and possibly also in exciting the room in a balanced manner below Schroeder Frequency due to "non neutral positon" regarding LF mode excitation in many practical cases
The competitors of "edge or corner" placed designs without "inherent directivity" are naturally those systems that provide "inherent directivity" even and especially from the lower and mid frequency range on upwards.
Such systems like dipoles and cardioids (but also other concepts having higher DI like arrays and larger waveguides), may also provide more "coherent wavefronts" from upper bass to midrange, but may be placed more freely to make up a stereo triangle to match individual geometrical restictions of a listening room.
Opting for "edge or corner" placed designs in the end is opting for "DI around 6dB" in upper bass and midrange (e.g. dipoles and cardioids have DI around 4.8dB, monopole in front of a wall has 3dB).
If a well made "edge or corner placed" design can make up a proper stereo triangle, then it may also be a cost saving alternative to systems having "inherent directivity", which has to be payed in terms of higher cost in construction especially for the lower frequencies.
Kind Regards
Concerning "off axis smoothness" of the frequency response of a loudspeaker:
What could be meant with "off axis smoothness" regarding the acoustic reality of a living room e.g. ?
No single reflections - e.g. from a booshelf placed at a side wall - will be "smooth" in spectrum. Even if the loudspeaker was radiating a "smooth spectrum" towards the shelf, in an appropriate angle to be reflected to the listening seat, that will not yield a reflection being "smooth" in spectrum: The spectrum of a single reflection will be "scattered" instead.
So "off axis smoothness" may IMO just refer to a smooth energy spectrum as a whole in a statistically defined sense.
- it does not mean, the loudspaker ought to have "as flat as on axis" response at all (also larger off axis ...) angles
- it does not mean the loudspeaker has to be phase coherent at all outside the "main listening window" relevant for the loudspaker's direct sound.
Concerning "floor coupled upfiring" position, which is the topic of this thread:
"Floor coupled upfiring" means: Edge position of the loudspeaker in the room.
Idealized edge and especially corner positions reduce the number of mirror sources in proximity to the loudspeakers.
When using monopole sources , this is an option to generate more coherent wavefronts as would be possible using monopole soures having distances to the walls in the order of low frequency wavelengths and fractions of those.
It is reasonable, that by this "room constructed directivity" an advantage in coherence especially from above the Schroeder Frequency until midrange can be achieved. That frequency range say from 200Hz to 800Hz or so is most important in localization because of
- the stereo system working well here (above about 2Khz does not give very proper cues depending on signal shape)
- there is spectral dominance in localization cues
But the "edge or corner" placed speaker has potential difficulties or drawbacks also:
- in providing an adjusted stereo triangle to the listener in every room's situation. In fact this is hard enough, with speakers that can be placed rather freely ...
- in having a - positive or negative - angle of elevation, exciting HRTF cues regarding that angle
- real edge and corner positions may not be ideal at higher frequencies, where practical distances of sound sources become relevant for interference, as real sound sources (drivers, tweeters) have non zero distances to the adjacent walls and the floor
- and possibly also in exciting the room in a balanced manner below Schroeder Frequency due to "non neutral positon" regarding LF mode excitation in many practical cases
The competitors of "edge or corner" placed designs without "inherent directivity" are naturally those systems that provide "inherent directivity" even and especially from the lower and mid frequency range on upwards.
Such systems like dipoles and cardioids (but also other concepts having higher DI like arrays and larger waveguides), may also provide more "coherent wavefronts" from upper bass to midrange, but may be placed more freely to make up a stereo triangle to match individual geometrical restictions of a listening room.
Opting for "edge or corner" placed designs in the end is opting for "DI around 6dB" in upper bass and midrange (e.g. dipoles and cardioids have DI around 4.8dB, monopole in front of a wall has 3dB).
If a well made "edge or corner placed" design can make up a proper stereo triangle, then it may also be a cost saving alternative to systems having "inherent directivity", which has to be payed in terms of higher cost in construction especially for the lower frequencies.
Kind Regards
Last edited:
Remark:
In fact a placement directly on the ground will also add further 3dB in directivity index to dipoles and monopoles ...
Freefield:
Monopole...: 0dB
Dipole.......: 4.8dB
Cardioid....: 4.8db
"Room Contribution":
Floor (one plane).......: +3dB
Edge (two planes)......: +6dB
Corner (three planes).: +9dB
That means an "edge placed" monopole source has comparable DI (6dB) in midrange like e.g. a cardioid or dipole in the freefield (4.8dB) ...
And a "corner placed" monopole has comparable DI (9dB) in midrange like e.g. a dipole or cardioid placed on the ground (7.8dB) ...
LineArray said:
In fact a placement directly on the ground will also add further 3dB in directivity index to dipoles and monopoles ...
Freefield:
Monopole...: 0dB
Dipole.......: 4.8dB
Cardioid....: 4.8db
"Room Contribution":
Floor (one plane).......: +3dB
Edge (two planes)......: +6dB
Corner (three planes).: +9dB
That means an "edge placed" monopole source has comparable DI (6dB) in midrange like e.g. a cardioid or dipole in the freefield (4.8dB) ...
And a "corner placed" monopole has comparable DI (9dB) in midrange like e.g. a dipole or cardioid placed on the ground (7.8dB) ...
Last edited:
LineArray said:
Of course this is not true as a generalization. I was referring to the sketch decribing the loudpeaker's positions in the initial post of the thread:
http://www.diyaudio.com/forums/atta...ges-floor-coupled-up-firing-speakers-room.jpg
BTW most loudspeakers may be considered as "floor coupled" in the bass ...
Edge or corner position is IMO like "wave guiding" the loudpeaker just using the room's walls. Aligning the radiation axis of the drivers at higher frequencies with the radiation axis of the wave guide is just kind of consequent here.
The improper match of the drivers to the "room corner waveguide's" throat at higher frequencies - which is ineviteable especially for the tweeters in that approach - would result in even more severe reflections otherwise.
Last edited:
Cocktail party effect:
Surely binaural hearing and estimating a "direction of arrival" plays a role here, which is important for source separation.
But solving the task can hardly be explained without using some kind of interband correlation and pitch detection.
Pitch variations during vowel articulation happen to be synchronously and at same rate for the fundamental frequency and the harmonics of a vowel's sound as well.
And because articulatory movements filter the fundamental frequency and the harmonics as well, there are also close interband correlations between the envelopes in different frequency bands.
The glottal pulses of a talking individual - as the primary sound source - places also a phase correlation between fundamental and harmonics of a certain individuals speech sounds, at least for the vowels.
Without estimating interband correlations in the modulation patterns of speech (patterns in amplitude and pitch as well) source separation seems impossible.
A rather narrow banded filtering in the frequency domain seems a prerequisite for correlating e.g. envelopes between different frequency bands, which in turn can be viewed as a "time domain" task if you like. But the time frames relevant her seem to be in order of articulatory movements rather than in order of periods of the glottal signal ... articulatory movement has a time frame e.g. of 4 syllables per second.
In fact the source separation task uses both domains, time and frequency.
Articulatory movements of human speech production are not too different from processes in playing musical instruments. In fact brass instruments e.g. are quite close to the operating principle of the vocal tract.
Perceptive strategies used in source separation for speech may apply rather closely to source separation in musical events also.
The human earbrain has developed to be an "organ of speech perception" to a large extent.
Kind Regards
graaf said:
Surely binaural hearing and estimating a "direction of arrival" plays a role here, which is important for source separation.
But solving the task can hardly be explained without using some kind of interband correlation and pitch detection.
Pitch variations during vowel articulation happen to be synchronously and at same rate for the fundamental frequency and the harmonics of a vowel's sound as well.
And because articulatory movements filter the fundamental frequency and the harmonics as well, there are also close interband correlations between the envelopes in different frequency bands.
The glottal pulses of a talking individual - as the primary sound source - places also a phase correlation between fundamental and harmonics of a certain individuals speech sounds, at least for the vowels.
Without estimating interband correlations in the modulation patterns of speech (patterns in amplitude and pitch as well) source separation seems impossible.
A rather narrow banded filtering in the frequency domain seems a prerequisite for correlating e.g. envelopes between different frequency bands, which in turn can be viewed as a "time domain" task if you like. But the time frames relevant her seem to be in order of articulatory movements rather than in order of periods of the glottal signal ... articulatory movement has a time frame e.g. of 4 syllables per second.
In fact the source separation task uses both domains, time and frequency.
Articulatory movements of human speech production are not too different from processes in playing musical instruments. In fact brass instruments e.g. are quite close to the operating principle of the vocal tract.
Perceptive strategies used in source separation for speech may apply rather closely to source separation in musical events also.
The human earbrain has developed to be an "organ of speech perception" to a large extent.
Kind Regards
Last edited:
[QUOTE="LineArray]
But the time frames relevant her seem to be in order of articulatory movements rather than in order of periods of the glottal signal ... articulatory movement has a time frame e.g. of 4 syllables per second.
[/QUOTE]
...of course duration of certain consonants is much shorter than eg. 110 ms duration of a long vowel.
But the time frames relevant her seem to be in order of articulatory movements rather than in order of periods of the glottal signal ... articulatory movement has a time frame e.g. of 4 syllables per second.
[/QUOTE]
...of course duration of certain consonants is much shorter than eg. 110 ms duration of a long vowel.