Is this right or wrong or what?

[Doughboy]

I've noticed this several times now. I match up my mix to a reference, almost identically, (sorry for the unoriginality), then record it. Later I have brought the recorded, (eq'd) version up and it is slightly different from the reference. I'm not talking about the fact that I may have gain boosted, but at different frequencies, there are changes that I did not make. I know this because the reference has not changed.

At first I thought maybe it was just visual and not auditory, but if I readjust the EQ to match the reference again, the sound quality definately improves.

I have the latest 1.5 update. At least I'm pretty sure I do.

Thanks!

[HarBal]

First off, if you want the best results possible then you should never match your track with the reference exactly. It makes no sense too because each recording has unique nuances that belong to that recording alone. Please read the tutorial in the application help for an explanation and you can find additional information scattered through this forum.

Secondly, the matching will never be exact if you compare the HarBalized track to the original plus filter because the filter resolution is finite. This is particularly true at the low frequency end where the effect of finite filter length is most prominant. I'd assume this is where the departure in your case is greatest. Also note that the higher sampling rate you use the worse it gets as far as LF matching is concerned because the filter is a fixed 8192 points in length but at a higher sample rate this represents a smaller time window.

Regards,


Paavo.

[Doughboy]

Thanks Pavvo, you answered my question expertly.

Still seems I have better luck, when I match it up near perfectly. Just seems to sound a bit better. Not a lot, just a bit. I will continue to experiment though. One way or the other I'll never be totally satisfied, satisfaction is just not in me when it comes to my songs.

but that's just with my songs, I am totally satisfied with Har-Bal. I hope you get filthy, filthy, rich if you're not already. What a great program!

[HarBal]

Thanks for that. I'm just doing alright and that's all I really want from it, that and people actually finding the software useful in improving the quality of their music productions.

Paavo.

[Guest]

.....the filter resolution is finite. This is particularly true at the low frequency end where the effect of finite filter length is most prominant. I'd assume this is where the departure in your case is greatest. Also note that the higher sampling rate you use the worse it gets as far as LF matching is concerned because the filter is a fixed 8192 points in length but at a higher sample rate this represents a smaller time window.

Regards,


Paavo.

Why doesn't Har-Bal filter resolution scale proportional to sample rate?

[electro]

Why doesn't Har-Bal filter resolution scale proportional to sample rate?

Same Post as "Guest" above

electro

[Paavo]

Why doesn't Har-Bal filter resolution scale proportional to sample rate?

Same Post as "Guest" above

electro

I've answered that one before in the HarBal forum somewhere though I can't tell you where. I'll give you the short answer here.

Though it sounds simple enough in theory, in practice it takes a great deal of management to make it function satisfactorily. Fixed length is much easier to implement than arbitrary length, primarily because efficient FIR filter implementations are generally optimised and it is difficult to optimize for an arbitrary length. How many other products out there can you name that have FIR's whose lengths scale with the sampling rate? My guess is you'll be pushing to find one. That's not to say it isn't on the agenda but it is a low priority.

Paavo.

[electro]

So for now, the rule is the lower the sample rate, the better the resolution? I understand that it is difficult to implement and optomize an arbitrary impulse length for scaling sampling rates. As an alternative, will you consider making a second impulse length optomized for 192Khz whenever it is convenient? Even if it has 4x the CPU hit of the current impulse length, it will be worth it. That will be the preferred sampling rate in the near future. I think it is more important that 96Khz. After that 88.2 seems to be the next most important samplerate. Multiple impulse lenghts optomized for higher sample rates could be a feature of Har-Bal 2.0.

[Paavo]

electro,

I will be looking into this and will most likely implement complete scaling but I can't tell you when. To implement a plugin version requires some redesigning of the filter engine to run with arbitrary process buffer lengths. Under that re-implementation I can add the necessary changes to support longer filters without significant SNR performance degradation.

Cheers,


Paavo.