Nice and useful add-ons.
A question about Hilbert transform at 48K with REW.
Does an interpolation (upsampling) to 96K is done before HT ?
At least,4 samples by period are needed to process a Hilbert transform (90° phase shift).
Saying that because a little deviation beyond 12KHz is noticeable at Fs=48K.
compare to a minimum phase generated by "hand".
Not very important,but just to understand what makes difference
A question about Hilbert transform at 48K with REW.
Does an interpolation (upsampling) to 96K is done before HT ?
At least,4 samples by period are needed to process a Hilbert transform (90° phase shift).
Saying that because a little deviation beyond 12KHz is noticeable at Fs=48K.
compare to a minimum phase generated by "hand".
Not very important,but just to understand what makes difference
Last edited:
No, that doesn't work for a Hilbert/cepstrum approach - upsampling would mean half the spectrum would be zero and zeroes in the spectrum are a problem for it. The limitations are inherent in that method (though they can be made even worse by insufficiently zero padding the time sequence). A better result needs a different method, which is on the todo list.
Hi All,
A brilliant and informative thread, which I have not yet read right through! I would like to use rePhase for a 3- or 4-way PA rig with a view to flattening the phase response. It is always necessary to use EQ 'on the fly', but I have seen no posts regarding its implementation. My current understanding is that any EQ anywhere in the chain will introduce new, unknown phase shift, destroying the flattened phase response originally set up. Am I missing something here or is it as simple as a software linear phase EQ somewhere in the chain?
Cheers, Carl.
A brilliant and informative thread, which I have not yet read right through! I would like to use rePhase for a 3- or 4-way PA rig with a view to flattening the phase response. It is always necessary to use EQ 'on the fly', but I have seen no posts regarding its implementation. My current understanding is that any EQ anywhere in the chain will introduce new, unknown phase shift, destroying the flattened phase response originally set up. Am I missing something here or is it as simple as a software linear phase EQ somewhere in the chain?
Cheers, Carl.
The FIR filters needed to adjust the phase response introduce time delay and the longer the filter the more time delay. This can be an issue for live sound. There are some pro solutions that use FIR like the Lake processors, so it can be done but I think they keep the latency under a few milliseconds to avoid sync issues.
If you used linear phase EQ to do the on the fly tuning that would not change the phase correction but I'm not sure how well phase correction will work in a live sound environment due to the changing distances. Using something like a Danley Synergy that has inherently flatter phase would probably work better.
DSP correction only works correctly for one point in space, which is why people use multiple measurements to average out the effects so that the correction is valid over a much greater area.
If you used linear phase EQ to do the on the fly tuning that would not change the phase correction but I'm not sure how well phase correction will work in a live sound environment due to the changing distances. Using something like a Danley Synergy that has inherently flatter phase would probably work better.
DSP correction only works correctly for one point in space, which is why people use multiple measurements to average out the effects so that the correction is valid over a much greater area.
I've yet to find any device, pro, studio, or home...that allows on-the-fly linear phase PEQ or shelving filters.
My solution for general eq, which IMO is absolutely necessary on a CD by CD basis, is to apply global eq before the signal goes off to the FIR processing channels which I use for 4 way speaker management.
That said, I've recently been toying with simply varying the output levels post FIR, going to the 4 way setup. Each pass-band, sub, mid, HF, VHF, spans 2-3 octaves and it surprises me how easy it is to achieve pleasant and smooth tonal balance moving them up and down like a GEQ.
ReaFIR isn't exactly parametric format nor nearly as flexible as rePhase, but it is 'on the fly' and I use it on my front-end DSP stack for that reason. It has linear phase mode (called 'reduce artifacts'). There is of course latency in the adjustments but it provides up to 32k FFTs and an easy to use GUI. Sound quality is quite good. RePhase continues to manage my crossovers and anything else relatively static.
Which leads me to the question I logged on to post: I have found that flat-top windowing produces quite well defined sound on those crossovers and seems preferable on my system to, e.g., Blackman-Nutall and other more generally recommended windows. Has anyone else tried this? It is certainly not highly recommended for hi-fi purposes.
Which leads me to the question I logged on to post: I have found that flat-top windowing produces quite well defined sound on those crossovers and seems preferable on my system to, e.g., Blackman-Nutall and other more generally recommended windows. Has anyone else tried this? It is certainly not highly recommended for hi-fi purposes.
Hello Carl
linear-phase EQ are a no go for sound reproduction.
Minimum-phase EQ (either IIR or FIR, or even analog for that matter) are what you are looking for, as flattening magnitude (I suppose that is what you are looking for when applying EQ) will also flatten phase in the same time.
The only thing you want to correct using non minimum-phase corrections are all-pass filters (more or less textbook ones...) caused by crossovers, and that kind of correction does not require tweaking from one place to another as it is related to the source itself.
So if you are looking for a linear phase system, what you should do is first measure your system in anechoic conditions (that can be done in a normal room, using gating and several measurements at different distances and angles...), EQ that to flat using minimum-phase EQ, and then apply gentle phase corrections (filter linearization tab as well as a tad of phase EQ if you need to) to flatten phase.
Once you have all that in a FIR you will only need minimum-phase EQs to tailor your system to a given venue, and that is perfectly done using IIR EQ devices, and in real time if you so wish
Hello leoman
Interesting.
What version of rephase are you using?
rephase's flat top implemention had a major bug prior to version 1.0.0.
Flat top is an interesting windowing function with the overhang in its attenuation, but it also leaves less taps for the actual correction.
https://en.wikipedia.org/wiki/Window_function#Flat_top_window
I never really used it for actual corrections.
Some tests are in order!
If one have a fully calibrated DSP steered speaker system then think global EQ (IIR) must be most right way to correct tonality because whatever it do to phase will be presented same to all speaker systems pass bands, adjusting level per pass band afterwards for correcting tonality will ruin systems pre calibration for driver offset and show up as new small ripple in phase/group delay plots, and having that offset set correct in a DSP system is normal an improvement compared to non stepped baffle's and analog XO systems.
IME, multiple measurements and spatial averaging <=> One measurement only but applying very smooth corrections.
One point valid correction is NO valid correction: does not work, even at one point in space.
Forget about perfection, corrections that "nail it" and naivetes of that kind!
Last edited:
Hi BYRTT, Not if drivers' phase are sufficiently flat (and zero) through extended crossover regions. Offsets don't change.... IMO this is some of the beauty of what can be done with rephase driver-by-driver
Last edited:
Hi BYRTT, Not if drivers' phase are sufficiently flat (and zero) through extended crossover regions. Offsets don't change.... IMO this is some of the beauty of what can be done with rephase driver-by-driver
Thanks that sound nice
admit haven't yet tried out that compensate model but have experienced the model with textbook IIR XO's slopes compensated FIR filter linearization, and these exercises show a change in pass band level will mean new settings for offset.That is the point of multiple measurements that it averages the differences. So yes a single point measurement with broad correction will work out to be similar. Which works best will probably depend on the room that it is used in.
That makes no sense, it is valid for that point in space but I agree that it is not a good way to go about correction.
Just because I tried to explain a concept in response to a question doesn't mean that I think it is a good idea. For myself I am only interested in overall corrections that work across a large area and have no intention of "nailing it" with a perfect single point correction.Forget about perfection, corrections that "nail it" and naivetes of that kind!
BTW whether it is intentional or not your posts come across as condescending. Insinuating that people are naive, ignorant, stupid or idiotic is a good way to get a negative response from them. I find you tend to catch more flies with honey, try it you might like it
So those of us using a one spot measurement as the base for DSP (while being able to check what it does off axis, it isn't that hard) must be out of their minds?IME, multiple measurements and spatial averaging <=> One measurement only but applying very smooth corrections.
One point valid correction is NO valid correction: does not work, even at one point in space.
Forget about perfection, corrections that "nail it" and naivetes of that kind!
I'll throw in a different view. Expecting DSP to be able to solve all room problems is asking for trouble.
Once you do consider the speaker and room together and hunt down the early reflections (trough measurements) even a single point measurement viewed with the right type of windowing starts to make a lot more sense.
In a room/speaker situation where things change faster once we move out of the sweet spot I'd recommend using multiple measurements as the base for DSP.
What might work varies so much with the speakers used and all of the different rooms we all have (and what we can do with it) that it's save to say there isn't a one size fits all solution.
There will always be people skeptical of using any form of DSP or EQ at all. I just hope they never find out how much of it was used to mix/master their favourite songs.
BTW whether it is intentional or not your posts come across as condescending. Insinuating that people are naive, ignorant, stupid or idiotic is a good way to get a negative response from them. I find you tend to catch more flies with honey, try it you might like it
I didn't say "people are naive (me not...
)". But there exist indeed a wellknown kind of fetichism which consists in... deafly flaten a room curve removing any bump or deep in it, or worse even the fetichism of the pure perfect impulse/step response.To what extent what we see on those beautiful graphs matches what we hear?
To what extent is the feeling of obtaining an audible inprovement of the sound is influenced by the cosmetic elegance of an almost flat curve?
I suppose we are all aware of the good old before/after advertising trick...
Last edited:
I didn't say "people are naive (me not...
No you just linked it directly with DSP correction
There is a lot of research to suggest that a flatter frequency response is preferred by most listeners. Which is most likely the reason why most strive to achieve that. There is also nothing wrong theoretically with trying to reproduce the transient response of the original file. As long as something else is not lost along the way then I don't see the harm.
Impossible to say without actually getting to hear it.
That's why it is important to listen as well as measure otherwise the answer to perfect sound would just be to press measure and correct.To what extent is the feeling of obtaining an audible inprovement of the sound is influenced by the cosmetic elegance of an almost flat curve?
I think what is more important is to try and understand why something sounds better and then to find a way to measure that so it can be repeated.
That looks like Dirac which is one of the better automated room correction systems as it uses multiple measurements to base the correction on. But that does look to be a brute force use of the correction which is clearly not ideal. In that case it probably does not sound as good as it looks.
It's just not helpful to run down other people's attempts to improve their system without explaining what you would do differently to make it better. It's easy to pick faults but much harder to give real answers.
But this thread is meant to be about the rephase program itself which is starting to disappear in the distance so that's enough from me.
Take it easy and do not shoot the messenger!
I often fooled myself to believe that my system seemed to sound "BETTER" when it only sounded "DIFFERENT".
My mind was influenced by some obscure rational arguments that drove me to the conclusion that the change was for positively better. But my ears were never quite sure...
Imho, the good thing with Rephase is that it does not come with an automated system for inversion and processing of the room curve, and lets the user choose what corrections he thinks necessary.