I gave it a shot as well yesterday. The result was not what I was expecting. It came out with a garbled impulse and step, and also a worse looking frequency sweep.
Could've been human error on my part though. I also had to check the normalize box when exporting from REW as the wav without normalize was barely wiggling. Almost a straight line.
Could've been human error on my part though. I also had to check the normalize box when exporting from REW as the wav without normalize was barely wiggling. Almost a straight line.
I'm glad it's working out for you. I've attached a pic of my speakers: Dayton RS100-4 drivers in "Hypercube" enclosures.Anyway, I wish I could explain the results. I can only speculate right now that maybe the RT stage uses a different window shape than the prefilter stage. This is sort of what I had in mind while writing my initial post last week where I suggested that the RT stage was smoothing things out while retaining some of the longer filter's attributes. If anyone has an idea about this, please share it; I'm kind of driving myself nuts wondering about it...
Attachments
Last edited:
What's the difference with running the larger préfilter window without the RT stage?
Can't the difference be accounted to 'just using' the longer préfilter window?
I don't get why the Ringing truncation would be the filter, to me it's the size of the window for the préfilter that determines how the rest within DRC works. That would mean this would only count as doubling that préfilter window size. Whatever the Ringing truncation does is minimal compared to the work done before it. Except it may smooth out some of the longer window's attributes (the actual ringing truncation part) which could make the resulting filter a bit more 'relaxed' than using the double sized préfilter alone without the ringing truncation stage.
Viewing the actual resulting convolution filters in REW (one with and one without RT stage active should shed some light on that. That would show us what the real differences are, at least if we dissect that IR.
Just thinking out loud here...
Can't the difference be accounted to 'just using' the longer préfilter window?
I don't get why the Ringing truncation would be the filter, to me it's the size of the window for the préfilter that determines how the rest within DRC works. That would mean this would only count as doubling that préfilter window size. Whatever the Ringing truncation does is minimal compared to the work done before it. Except it may smooth out some of the longer window's attributes (the actual ringing truncation part) which could make the resulting filter a bit more 'relaxed' than using the double sized préfilter alone without the ringing truncation stage.
Viewing the actual resulting convolution filters in REW (one with and one without RT stage active should shed some light on that. That would show us what the real differences are, at least if we dissect that IR.
Just thinking out loud here...
Last edited:
A longer prefilter window with no RT means high Q dips/peaks in the filter response (which cause it to ring longer). The filter that is generated by the custom config file (8 cycles prefilter/4 cycles RT) is smooth like the old 4 cycles filter, but just a bit more effective for some reason. It's sort of like doing the windowing after the inversion rather than before it, which if you think about it, isn't really all that different. The effect of the RT stage is really only "minimal" if it's using the same size window as the prefilter stage. Tomorrow we'll look at some graphs; let me know what you want to see.
What might shed a light is seeing the FR + Phase of the convolution filter itself in REW from your previous 4 cycle filter, an 8 cycle filter without RT stage and your latest template which includes the RT stage. That would show the actual correction that will be applied to the speaker.
The subjective descriptions, a fuller sound, make me smile. The more we get close to time coherency, the more harmonics will line up within the recordings. That to me is the good part of FIR processing. Being able to get the same "bloom" from an acoustical guitar as when it's actually played in front of you. The lower it reaches as a system, the more real it sounds and feels. It makes the "being right there" believable. At least to me it does.
The subjective descriptions, a fuller sound, make me smile. The more we get close to time coherency, the more harmonics will line up within the recordings. That to me is the good part of FIR processing. Being able to get the same "bloom" from an acoustical guitar as when it's actually played in front of you. The lower it reaches as a system, the more real it sounds and feels. It makes the "being right there" believable. At least to me it does.
Thanks for helping me think this through. The attached graphs show old vs. new filter response, and also the effect of the RT with the new filter. It's interesting to study both of these; despite the smoothness of the new filter, it does seem to be retaining some of the attributes of a longer windowing - things that the 4 cycles filter isn't seeing.
edit: remember that I made a couple of other small tweaks to the new filter, so there are other variables here...
edit: remember that I made a couple of other small tweaks to the new filter, so there are other variables here...
Attachments
Last edited:
I kind of figured this would be what's happening. Basically I've been doing my own RT stage by hand, with tiny PEQ's smoothing out what was overblown by DRC. I generally do this while observing the effects on both the impulse and early waterfall graphs while keeping the FR target.
Can you also show the phase? It could be likely that there are differences there too.
Can you also show the phase? It could be likely that there are differences there too.
I know they are, that doesn't matter, it's what they do that counts. Minimum phase filters in DRC will still alter phase of the speaker, as these plots show. They will probably differ once you look trough them with cycled filters, though REW is too smoothed to tell the whole story. The phase manipulation of the longer pré filter window could show differences. This isn't much difference though, which means if the resulting FR graph is better behaved (by the RT stage removing over compensation) it will be potentially better overall.
This looks like an improvement, as the filters without RT stage look way more aggressive.
This looks like an improvement, as the filters without RT stage look way more aggressive.
Last edited:
Another question if I may, do you record the convoluted speakers at the listening position to verify? I balance them Left and Right with broad stroke PEQ by hand using a ~6 cycle window. Again with the very early waterfall graphs to confirm. Made a world of difference for stereo. We can get them awfully close to the same. Especially in the part above room modes.
Now I'm way further from the speakers and it is probably less severe in your near field setup. But all it costs is some time.
A further explanation, I only look at general trends over a wider area. small dips and/or peaks are not that audible, with FIR correction and looking at 6 cycle graphs you probably don't even see that. But small SPL differences are audible if they affect a large enough area. By looking at the 6 cycle you can easily see if left and right channel are tracking each other. A few low Q bumps or cuts can help align the general trends.
Doing this with a 6 cycle window will improve the more traditional graphs (where the room is included) as well.
Here's a 1/12 smoothed plot showing the balance between left and right channel:
Aside from the energy shuffle I use below 80 Hz they track each other pretty good.
In a more traditional 1/3 smoothing it starts to look like straight lines.
This is the resulting graph, meaning I actually applied the PEQ based on a 6 octave FDW.
The right channels has the disadvantage of a few higher level reflections (less damping on that side).
That makes it a bit more restless between 100-200 Hz.
Now I'm way further from the speakers and it is probably less severe in your near field setup. But all it costs is some time.
A further explanation, I only look at general trends over a wider area. small dips and/or peaks are not that audible, with FIR correction and looking at 6 cycle graphs you probably don't even see that. But small SPL differences are audible if they affect a large enough area. By looking at the 6 cycle you can easily see if left and right channel are tracking each other. A few low Q bumps or cuts can help align the general trends.
Doing this with a 6 cycle window will improve the more traditional graphs (where the room is included) as well.
Here's a 1/12 smoothed plot showing the balance between left and right channel:
Aside from the energy shuffle I use below 80 Hz they track each other pretty good.
In a more traditional 1/3 smoothing it starts to look like straight lines.
This is the resulting graph, meaning I actually applied the PEQ based on a 6 octave FDW.
The right channels has the disadvantage of a few higher level reflections (less damping on that side).
That makes it a bit more restless between 100-200 Hz.
Last edited:
No, I just trust that the predictions are close enough to reality to not really matter. Also, I'm trying to move in the direction of more automation. Speaking of which, I think it's interesting that this new config may be an attempt on my part to automate something which you've been doing manually...
edit: I'm seeing your edit after posting. Yes, I get your point and I do think I have better correlation now between speakers with the new filter.
edit: I'm seeing your edit after posting. Yes, I get your point and I do think I have better correlation now between speakers with the new filter.
Last edited:
The predictions are pretty close, but I did see differences from time to time.
Yes, this new found recipe seems to do exactly that. Which I would agree to is gaining a better result overall.
I'm a lost cause, I like to confirm every step of the way. It gained me a solution for the big dip I got due to the corner placement on the right, being able to making up for that was a huge step forward. In fact, it's the reason I'm playing with the thought of adding more cone area to be able to do what I did separately on the left and right. It would also strengthen the low end, which is to die for in the amount of atmosphere you get from that.
Yes, this new found recipe seems to do exactly that. Which I would agree to is gaining a better result overall.
I'm a lost cause, I like to confirm every step of the way. It gained me a solution for the big dip I got due to the corner placement on the right, being able to making up for that was a huge step forward. In fact, it's the reason I'm playing with the thought of adding more cone area to be able to do what I did separately on the left and right. It would also strengthen the low end, which is to die for in the amount of atmosphere you get from that.
As I'm always editing posts, I'll just use a new one for once. One final test I always do is play both speakers together to see how they add up. They should add up perfectly. And if all went well they really do. That's how I found the cure for a dip I experienced when I was only looking at predictions.
It would never have come to me to (use PEQ) cut at a dip to be able to cure it without that step.
It would never have come to me to (use PEQ) cut at a dip to be able to cure it without that step.
