Audacity Forum

Gordon Coates wrote:Many thanks steve and Gale. I have imaged and exported the curve from the Windows machine - I'll see what happens when I import it under Mac, and I'll post screenshots as soon as I find the image editor (or maybe install one?) on the MacBook...

Very interesting... importing the curve exported from the Windows machine into Audacity on the Mac worked perfectly (many thanks again!) as can be seen from two of the attached screenshots.

However, attempting to CREATE the same curve by using the same points to draw a curve in Audacity under Mac in fact creates a completely different curve, as can be seen in the other attached screenshot.

Gordon Coates wrote:However, attempting to CREATE the same curve by using the same points to draw a curve in Audacity under Mac in fact creates a completely different curve, as can be seen in the other attached screenshot.

Incidentally, the curve I created under Windows and imported to my MacBook was created in Audacity 2.0.2, whereas the version of Audacity on my MacBook is 2.0.3. Perhaps this means that the problem with "Draw Curves" in version 2.0.3 could routinely be worked around by keeping version 2.0.2 handy on another computer, as in the case just posted above, at least for the time being...

Gordon Coates wrote:Very interesting... importing the curve exported from the Windows machine into Audacity on the Mac worked perfectly (many thanks again!) as can be seen from two of the attached screenshots.

However, attempting to CREATE the same curve by using the same points to draw a curve in Audacity under Mac in fact creates a completely different curve, as can be seen in the other attached screenshot.

Thanks. That appears to illustrate a bug.
Interestingly, on Linux I'm getting a green line that follows the blue line much more closely than either your Windows or Mac screenshots.

Could you post the XML files for both machines.

Gordon Coates wrote:Incidentally, the curve I created under Windows and imported to my MacBook was created in Audacity 2.0.2, whereas the version of Audacity on my MacBook is 2.0.3. Perhaps this means that the problem with "Draw Curves" in version 2.0.3 could routinely be worked around by keeping version 2.0.2 handy on another computer

When I commented on the current known bugs in EQ in 2.0.3, I said "Draw Curves" when I meant "Graphic EQ", which I've corrected now. Sorry about that. A potential fix for those bugs has now been committed, which seems to restore behaviour to 2.0.2 accuracy as far as I can tell so far.

Using a filter length of 8191 and importing Steve's test XML file, the green line curve has the same good fit for me on Windows 7 (Audacity 2.0.3 release and "fixed " 2.0.4 alpha) and Mac OS X 10.8.3 (Audacity 2.0.3 release).

If you (Gordon) can post the XML file you created using 2.0.2 on Windows I will try creating that curve on Mac and Windows.


Gale

steve wrote:Interestingly, on Linux I'm getting a green line that follows the blue line much more closely than either your Windows or Mac screenshots.

Steve, are you working with a 192KHz file open? I find that the green line matches drawn curves very closely with 44.1KHz files, reasonably closely with 96KHz files, and not closely at all with 192KHz files. By the same token, I find that, if a curve made for 44.1KHz files is applied to 96 0r 192KHz files, it doesn't work as expected at all - so I always name my Eq curves with a name that includes the sampling frequency they were made under, because that seems to be the only sampling frequency they will work properly with.

Gordon

Gale Andrews wrote:If you (Gordon) can post the XML file you created using 2.0.2 on Windows I will try creating that curve on Mac and Windows.Gale

Many thanks again, Gale. Here is the XML file I exported from 2.0.2 under Windows and imported into 2.0.3 under Mac OSX. (It only creates the equalization I need to flatten out the room interaction with my speakers when applied to 192KHz files - I use very different curves to create approximately the same equalization with 44.1 or 96Khz files).

Gordon

Gordon Coates wrote: Here is the XML file I exported from 2.0.2 under Windows and imported into 2.0.3 under Mac OSX. (It only creates the equalization I need to flatten out the room interaction with my speakers when applied to 192KHz files - I use very different curves to create approximately the same equalization with 44.1 or 96Khz files).

On Windows, there is no difference between 2.0.2, 2.0.3 and 2.0.4 alpha that I can see - whether importing that XML file or drawing its EQ points, the green line needs the maximum filter length to fit closely at 44100 Hz project rate, at 8000 Hz it fits closely with as short as 2500, and at 19200 Hz or higher the fit cannot be made tight (by tight, I mean that the line appears green or blue/green).

I'm out of time to test on Mac until later, but even if there are Mac differences between 2.0.2 and 2.0.3 as seen in your images they are slight in comparison to the sample rate differences. I did not know that relationship existed, so I'd be interested to read Steve's comments.



Gale

Gordon Coates wrote:Steve, are you working with a 192KHz file open?

Ah ha! That explains it

The "Size" of the filter is the "FFT window size". That is the number of samples that are used in each (overlapping) analysis window.
The larger the window is, the greater the accuracy of the filter at low frequencies.
The low frequency accuracy is limited by the FFT window size.

If a high sample rate is used, then the length of each window in seconds becomes much smaller (1024 samples at 44100 Hz sample rate is about 0.023 seconds, but at a sample rate of 192000 Hz it is only about 0.005 seconds.

Unfortunately you can't just keep increasing the window size, because some of the calculations become increasingly less accurate unless the calculations are performed in much higher precision. Currently I think Audacity uses 64 bit precision for these calculations, so the maximum window size could perhaps be increased to a bit more than the current "8191" maximum, but there is another down side, which is that as the window size increases, so the pulse response becomes worse, which can cause "ringing". For FFT filters there is a trade-off between sample rate, frequency precision, and time precision. If any two of these are increased, the third must be reduced. FFT is a remarkable piece of mathematics, but it has its limits.

steve wrote:

Gordon Coates wrote:Steve, are you working with a 192KHz file open?

Ah ha! That explains it

As it looks a bit like a "bug", I added a sentence about this to the Manual .

No doubt you'll correct/improve it if needed.


Gale