Seeking Advice on Improving FIR Filters for 4-Way Stereo System

First bit is to be 100% certain that your measurements are ok. Two, maybe three things are required. 1) If your audio devices create glitches in the audio streams, the ttd correction will suffer. 2) If your measurement microphone isn’t properly calibrated it is very difficult to dial in on the best voicing. 3) If you measure at such a loud level that tour speakers are strained when they play the sweeps.

If you’re past that, more specifics about the sound quality would be helpful.

With a good mrasurement you can pretty much voice the sound to your preferences. And the only thing after that might be left to improve then would be transparency. But I have reason to believe that Audiolense is on top in that department. So hang in there. Trust the app, but be diligent about measurement quality.

Nothing stands out as wrong with the impulses.

Does the timing difference between the tweeters look OK? And do you get similar timing on the tweeters from measurement to measurement?

And ... what is it that you don't like with the sound?

Fostex? I'm guessing comppression driver here. With horn loading too maybe?

The tweeter corrections do not look optimal. I'm sure there are better ways.

I could look at your measurement file and see what can be done if you put it up here.

Yes I beleve the tweeter is the main challenge here.

And before I continue this rather long response: What I explain below has relevance to all users.

The impulse response of your tweeters shows a strong reflection after 1 ms:




It would require a lot of correction to clean up the time domain here, and the cost would be too high. In fact, the default only has a duracion of 0.3 ms, give or take, around 10kHz. So what it "sees" and tries to correct is this:



It will see a bit more below 10 kHz and less above ... anyway this, too is quite an undertaking for a ttd correction. This is the resulting correction filter:


Such tweeter corrections are frequently reported to not sound good. And yes, compression tweeters and horn loaded tweeters generally do not express their best qualities when they are corrected too much. Actually no tweeter likes a lot of correction, but conventional tweeters on a flat baffle have a much cleaner impulse to begin with, so there's a lot for the correction to chew on.

The general fix is to use a partial correction. Below, a procedure where I stop the TTD correction at 4 and 10 kHz respectively.



And here are the resulting tweeter corrections:



As you see, both are much more quiet than the one I got from the default correction. Based on what I see here, 10 kHz TTD cutoff seems to be the better of the two, since it leads to a very quiet correction filter. There's not much more than the crossover duties before the main pulse here.

I've seen a lot more difficult-to-correct tweeters than yours, btw. Yours can handle frequency correction all the way up with grace. There are some horn tweeters that requires that even the frequency correction stops an octave or two below 20kHz.

The double impulse of your tweeters (top picture) means that you will not get that clean rise of the step response. It will rise in two stages, with a rugged part in the middle:


This rugged top is something that you should not try to fix. The tweeter is as good as it is even though it leads to this.

The pre-ringing in the step response here is btw most likely a measurement artefact and nothing to worry about. (And even if it were for real I doubt that it would be audible) The measurement signal has pre-ringing in the top, which is usually low pass filtered by tweeter + dac. But in this scenario here, with a sweep that stops way before FS/2, the tweeter + converter likely extends beyond the measurement signal.

Over to a slightly different topic:
In the decay of the step response above there are some deviations from ideal, especially the right speaker around the 286 mark. This most likely happens because the "selective prefinging prevention" is checked. A frequency region with excess phase-like behavior (very strong cancellations) has been omitted from the TTD correction.

From time to time better results are obtained with the preringing prevention unchecked. I'll leave to you to test that. If you do, you should first examine the simulated step response for low frequency pre-ringing. If it looks good, you should also examine the correction filters, just like I showed you with the tweeter. If the simulated step response looks great, but the correction filters look a tad too busy before the peak, you should use your ears to decide what sounds best.

BTW I narrowed your highest crossover point to 1 octave width, to keep the crossover region out of the TTD transition region. I don't know if it matters to the results I've shown here, but you can safely use a much narrower crossover for the tweeter if you see an advantage in it.

And to the rest of you: The general consensus seem to be that partial correction in the top to quieten the tweeter correction also has benefits for more conventional tweeters. Consider it a bonus from work that went into Audiolense to handle "unmanageable" horn tweeters with grace. I haven't automated this, because the same settings do not work equally well on all setups. This has to be tweaked manually.

I hope this helps, onousis!

No, you can try both and see what happens … in the same way that I just showed you. My experience is that ttd per driver unchecked gives better results more often, but it is far from rare to see that it improves things either.

TTD per driver is in theory better, but adds complexity….

With regard to noise there may be advantages to level the spl outside the digital domain. But Audiolense and anything else involved in streaming can handle that swing without problems.