Narrower critical bands at low frequencies ? In terms of Hz per band or fractions of an octave ?
I was under the impression that our hearing critical bands were wider at bass frequencies in terms of octaves, not narrower, and one cause of our relatively poor pitch perception at bass frequencies compared to middle frequencies.
Anyway, how exactly would narrower critical bands at low frequencies (if that is indeed the case) be consistent with longer integration times ? Surely low frequencies in and of themselves demand longer integration times to measure regardless of whether there was a proportional change in octave bandwidth compared to higher frequencies ?
Eg a 100Hz tone can't be integrated in 1ms when it takes 10ms for even a single cycle to be completed, regardless of whether the measuring device/ear has fractional octave band filtering or not.
Last edited:
Interesting insight Earl! I've been planning on reading further in Zwicker but I haven't finished Blauert yet (they're expensive books for a hobby, but very interesting). I've looked up the chapter you're referring to. Implementing the values in Zwicker in some sort of equalizing would yield something quite different from the variable window used in DRC, which was mentioned earlier. Do you think that the Zwicker data would be readily usable for a variable time window for equalization? Or might there be other factors to be considered?
Simon
The time-frequency relationship is in linear terms not octaves, so of course one uses linear frequency. The time-frequency tradeoff is inverse - as the filter gets narrower (in Hertz not in octaves) the associated time gets wider. Below 500 Hz the ears critcal band is a flat 100 Hz, rising to 1 kHz AT 5 kHz.
The time-frequency relationship is in linear terms not octaves, so of course one uses linear frequency. The time-frequency tradeoff is inverse - as the filter gets narrower (in Hertz not in octaves) the associated time gets wider. Below 500 Hz the ears critcal band is a flat 100 Hz, rising to 1 kHz AT 5 kHz.
Last edited:
As a first-cut its certainly what I would use. Are there other factors? - of course there are, but I strongly suspect that the critcal band concept, which is fundamental and physical, not cognitive, is the major effect and other effects would be of secondary importance.
But I want to be clear that I don't believe in Equalization in any frequency range where what I am saying is relavent. The idea is to find those reflections in a room that have the most detrimental effect and to deal with them in the room or the speakers, not with EQ.
Both Salmi and Kates used combinations of time window gates and critical bandwidth smoothing. In fact the critical bandwith was only used for smoothing to get to a perceptual frequency response. I don't know if it contributes, in and of itself, to the time window?
The Salmi data on critical bandwidth (which I believe he got from other sources) was approx. constant percentage of a little less than a 1/3rd Octave above 500 or 1k, and fixed bandwidth below. Is that about the same as Zwicker?
David S.
The Salmi data on critical bandwidth (which I believe he got from other sources) was approx. constant percentage of a little less than a 1/3rd Octave above 500 or 1k, and fixed bandwidth below. Is that about the same as Zwicker?
David S.
---------------------------
Dr Geddes, before you go too far with the Zwicker chapter, please keep in mind that the chapter was written a number of decades ago.
When auditory "filters" are measured there are a variety of techniques that are employed. It is beyond the scope of the current discussion to get into this topic. Historically, the filters were measured by making a couple of fundamental assumptions. This lead to two kinds of data sets. In one, the filters were derived from what are called "critical ratios" and the other(s) from what are called "critical bandwidths".
As a function of frequency, the two sets will parallel one another, except at low frequencies where one continues to act like a constant percentage filter and the other will flatten out and act as a constant number of Hertz in terms of bandwidth. There was more recent work (last decade) reconciling these seeming differences, but that discussion would become more difficult.
Also please keep in mind that while you can get a estimate of a temporal window by knowing the filter's bandwidth (which is where I think you are going), there are also other other temporal windows which could be the actual limiting process.
Last edited:
I have done some additional measurements with the in phase/out of phase model to see the impact of implementing unbalancies. If you don't get the distances from left and right speaker to the microphone exactly equal, you get something like this:
Only when I achieved less than 1 cm difference in the distances, the result became reasonable:
In fact this measurement is a more precise method to get to the exact point of the stereo triangle than my measuring tape hold free in the air.
Next I switched off a -6dB bandpass at 1.9kHz in the left speaker only:
Compare to the picture above. While the difference is not prominent in L+R, it becomes really evident in L-R, leading to less separation between L+R and L-R.
Putting some pillows on the floor between left speaker and microphone led to another sort of unbalance:
Again it is mainly the L-R curve closing the gap to L+R,nicely illustrated by these diagrams:
In the end this method is like the telephone call in the mid of the night: A muffled voice at the other end of the line is telling you "You are in deep trouble" and then hanging up. Now you know that things really go wrong, but you have no clue what to do about it.
Only when I achieved less than 1 cm difference in the distances, the result became reasonable:
In fact this measurement is a more precise method to get to the exact point of the stereo triangle than my measuring tape hold free in the air.
Next I switched off a -6dB bandpass at 1.9kHz in the left speaker only:Compare to the picture above. While the difference is not prominent in L+R, it becomes really evident in L-R, leading to less separation between L+R and L-R.
Putting some pillows on the floor between left speaker and microphone led to another sort of unbalance:
Again it is mainly the L-R curve closing the gap to L+R,nicely illustrated by these diagrams:
In the end this method is like the telephone call in the mid of the night: A muffled voice at the other end of the line is telling you "You are in deep trouble" and then hanging up. Now you know that things really go wrong, but you have no clue what to do about it.
Attachments
My comments were just guesses as to how to start, but if there is better data then I'm interested. Otherwise, Zwicker is right on hand. The Third edition that I have is from 2006, so if the data is no longer valid I would have suspected that Fastl would have updated it.
So where does the more recent data differ from the Zwicker data? Just the LF difference from using different approaches? Does one approach equate to the Zwicker data? I'm not clear on that from your comments.
And what would the limiting factor on the temporal windows be? Thats not obvious either.
-----------
1, I don't know that Fastl would have updated the text of Zwicker. It would have been quite a bit of work and there are probably some other issues as well.
2. If you are interested in pursuing this, I would look at the work on psychophysical tuning curves done with forward masking paradigms. That shows the sharpest tuning and is more alsigned with the physiological data. The reference material on auditory filters (and heavily cited) would probably come from Glasburg and Moore (various JASA pubs).
3. The issue of temporal windows is trickier since the question needs to be better defined. One approach to temporal resolution comes from classic work done by Viemeister. Otherwise, there needs to be specificity about exactly "what" is being windowed (integrated etc): eg, energy, a binaural cue etc and toward what end: eg, detection of a signal in noise or localization etc.
The book is by Fastl and Zwicker, and I strongly disagree that Prof. Fastl would not update a section of text that was known to be wrong. He is not that kind of academic. Further, to quote from the Preface to the Third Edition: "Moreover, in particular sections on ... were significantly expanded." but he fails to correct known faulty data? I don't think so.
I also took note of the fact that you specifically did not answer my key questions about how bad this data was, how did it differ from "newer" data?
I have to assume from both perspectives that the data is more than adequite for my usage here.
Perhaps your original warning was somewhat overstated. The fact that it is old and there is newer data does not make it wrong.
Last edited:
Well, I guess this is why I usually don't bother trying to be helpful.
My field, by training and experience, is in psychoacoustics. Your background is not.
It is difficult trying to describe large data sets and their implications in 25 words or less, especially to a general audience. When members of that audience are hostile and snide, then there is really no sense in me continuing to try to help and educate.
Since you have it all figured out, anything I have to say would not interest you. Good luck on your modelling efforts. It is an interesting topic, and perhaps at some point you will discover it is also a difficult topic.
My field, by training and experience, is in psychoacoustics. Your background is not.
It is difficult trying to describe large data sets and their implications in 25 words or less, especially to a general audience. When members of that audience are hostile and snide, then there is really no sense in me continuing to try to help and educate.
Since you have it all figured out, anything I have to say would not interest you. Good luck on your modelling efforts. It is an interesting topic, and perhaps at some point you will discover it is also a difficult topic.
Last edited:
You make unwarranted assumptions about our backgrounds and intelligence as a reason for patronizing us.
Furthermore, I don't see what is "hostile" or "snide" about Geddes's answers to you.
I appears you don't like folk to disagree with you even in a direct but polite fashion. That is not our problem so don't try to make it such.
---------------
Hold your horses Frank! I never commented on anyone's intelligence, you did.
But let's get real. Look at how the following terms have been used and mis-used in this thread: Precedence effect, integration, temporal window, auditory filter, HRTF, masking, detection, audibility etc, etc. Now your look at your comment about assuming someone's background is far from accurate.
These terms have been tossed around in the most haphazard fashion. So no, I do not think the majority of folks here have any background in psychoacoustics or auditory science. So much for trying to effectively summarize complex topics in 25 words or less. Why bother? Speculation seems to be what many value. It is so much easier that way. Isn't it?
Yes the thing I find most odd is that the two seem to add pretty much in phase across the entire frequency range (due to the fact that dropping off one speaker results in a curve around 6db lower that follows the two speaker curve fairly closely). However when the phase is reversed on one, it would appear that the speakers are anything but in phase throughout the audio band.
This is bugging me rather a lot as the two things seem to be at odds
Rudolfs recent measurements do show though that the accuracy of the mic placement does make quite a difference.
I'll have a play with speaker and mic positioning over the next few days and report back if I can get to the bottom of things.
Tony.
<i>So much for trying to effectively summarize complex topics in 25 words or less. Why bother?</i>
So why not try for one or two hundred words? If we're so ignorant and misplaced in our attention then perhaps you might be helpful, no?
Hell, you don't even give references! We just wouldn't get it, right? A qualified professional who likely would indeed understand asked you for references and you couldn't be bothered.
Instead, you tell us ex cathedra we got it wrong but it's not worth telling us what's right. Why should anyone take you seriously?
You got problems. Keep 'em at home!
So why not try for one or two hundred words? If we're so ignorant and misplaced in our attention then perhaps you might be helpful, no?
Hell, you don't even give references! We just wouldn't get it, right? A qualified professional who likely would indeed understand asked you for references and you couldn't be bothered.
Instead, you tell us ex cathedra we got it wrong but it's not worth telling us what's right. Why should anyone take you seriously?
You got problems. Keep 'em at home!
Guys remember that this is a forum and we cannot see each others faces. It's easy to misjudge someone's intention.
WithTarragon has already given some clues about where to look. So why not try to find the references he mentioned (Viemeister, Glasberg & Moore)? I know it's not possible for everyone here, but if we're familiar with those texts it would be easier to discuss this.
WithTarragon has already given some clues about where to look. So why not try to find the references he mentioned (Viemeister, Glasberg & Moore)? I know it's not possible for everyone here, but if we're familiar with those texts it would be easier to discuss this.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.