|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
58.107.189.194
I just read an article (by an EE in the higher end new tech audio hardware business) that as well as adding detail, that software upsampling (can) add dynamics.Anything that can add dynamics to dynamics-deficient pop rock is IMO well worth checking out.
Anyone experimented/ been able to compare the effect of upsampling on dynamics?
Follow Ups:
I've been "hella-busy" the last few days.No, I am not talking about clicks, pops and droputs. I keep a close eye on CPU utilization with each iteration and never let my PC get to the point where it is even close to doing that. There are some who have reported sonic degradation when they start reaching the *threshold* of CPU resources (especially when upsampling) and don't get dropouts per se, but they experience a different kind of sonic degradation. I guess I am more referring to added veils, or loss or detail and transparency and PRAT and involvement if too many DSP algorithms are used in sequence. My main concern is not really CPU usage. I have a 2.9 GHz PIV Hyperthreading CPU with 1 GB RAM and this thing EATS UP processor intensive tasks. I can run a convolver plugin (for room correction), a DSP Equalizer (for final tweaks) and a three way digital active crossover plugin ALL in Foobar ALL at once with less than 25% CPU usage. I can even add upsampling using SRC up to 24/96 but CPU usage starts to get higher than I would like - but I never get clicks or pops. (I have disabled quite a few XP services and run a pretty stripped down OS with no "XP Bling".)
There are three main ways (that I know of right now) to do active crossovers ONBOARD with a PC AND do realtime room correction.
1) Use Foobar with Convolver plugin (1st) and crossover plugin of your choice (2nd). Of course, use ASIO if possible.
2) Use Winamp with RealReverb Convolver plugin and Thuneau's VST plugin crossovers complete with the "phase arbitrator". Again, ASIO is availble and should be used if possible.
3) Use Brute FIR for implementing both crossover and room correction filters. Not for the faint of heart and quite a bit of PC and programming knowlege is probably in order. There is a software suite available to make the filters called the "Acourate Audio Toolbox" which is specifically designed to generate filters for Brute FIR. This also is not for the faint of heart and requires the user to have a measurement suite and the ability to use FRD files along with a crossover design tool to generate the filters.
Now, of course, you can do what AlJordan said and do some of these processes (like EQ or crossover functions) outboard and use the software player alone (or just for room correction). As AlJordan said "The DAC in the DEQ is not bad, but not outstanding either." this applies to the DCX2496 as well. But I think the ransparency and detail rendering ability of units like these has more to do with getting a low jitter spdif (or AES/EBU) input into the thing in the first place. Mark my words - if Behringer came out with a new DCX2496 that had a USB input that used I2S internallyt with NO conversion to SPDIF on the other end... put me down for two.
Behringer... did you hear that? :D
The trouble with Behringers is the digital inputs are only as good as the digital being supplied to them. Sure, they use the common phase-locked-loop for asynchronous connections and then upsample to 24/96... but I find that the Behringer will tell you very quickly when it does not like the digital stream you feed into it. Some folks have reported that the Monarchy Audio DIP Upsampler between their source and the Behringers open things up a bit, and I have had success with products like a Gensis digital lense as well.
But at the end of the day, and as it stands right now, the most transparency I am getting is doing the crossover work passively and only using the player for convolution. I am not upsampling now at all and since I've gotten the room correction thing working I've honestly put upsampling on the back burner.
Basically, what you do depends on how much you want to spend, how much PC horsepower you can afford, and what you personal "beliefs" are when it comes to "where to do" each individual process.
Right now I have my eye on a new system using shallow crossover slopes in a transient perfect filler driver design (using passive crossovers). This way I can use Winamp and Thuneau crossovers. Once the crossover "Phase Arbitration" is done for the 1st and 2nd order filters, THEN room correction can be done. I just need to ensure the Thuneau crossovers will work with Winamp before i get them - I think I need some sort of VST Plugin to make it go.
Anyways. The bottom line is that there are 1,000 different ways to go about doing this PC stuff. Add outboard options and then there are a 1,000,000 ways. Right now I am still experimenting with relatively cheap iterations to try and nail down what processes (and combination of processes) do the LEAST sonic "damage" to the music.
And I don't mean clicks and pops.
I mean interfere with the LIFE... the EMOTION of the music!!
PrestoThanks for the * very informative posts
Not suggesting that Behringers are the best at all, but what DACs have you used for reasonable money that are better?
Cheers
Presto wrote:
"Right now I have my eye on a new system using shallow crossover slopes in a transient perfect filler driver design (using passive crossovers). This way I can use Winamp and Thuneau crossovers. Once the crossover "Phase Arbitration" is done for the 1st and 2nd order filters, THEN room correction can be done. I just need to ensure the Thuneau crossovers will work with Winamp before i get them - I think I need some sort of VST Plugin to make it go."Hi Presto,
I would be interested in hearing more details about the shallow TP crossover system you are thinking about. Who designed the speakers and crossover?
I recently purchased the Thuneau Allocator so I can try some experiments with Winamp for you if it helps. I have been using the kernel streaming output of Foobar mainly for playback, but I have successfully configured my PC so that every sound made by any application will go through the Thuneau crossover/arbitrator. I particularly like the Thuneau crossover because it has parametric eq capabilities on each crossover leg. I can use the crossover to reduce the amplitude of standing waves and eliminate the DEQ 2496. The main problem is that my Lynx ASIO drivers will only let one application at a time access the ASIO drivers, which is why I have to use the kernel streaming driver in Foobar. I may consider looking into Console to see if it will allow me to stack multiple apps into the ASIO driver. I am also going to try to see if the M-Audio 410 ASIO drivers will allow multiple apps to access ASIO, although I am not sure the M-Audio mixer is flexible enough to accomplish the routing I need.
What sound card are you using?
Also, Jan at Thuneau mentioned that he is coming out with a new software package called Frequency Affirmer that will do "very good room correction and crossover at the same time". I am hoping that it will mimic DRC/BruteFIR in a package that allows for an easier visual configuration, but I have no details. I was able to get an impulse response under Linux and use BruteFIR, but the process was difficult even though I have a lot of linux experience. I would pay for a package that is easier to configure.
Alan:AP "I would be interested in hearing more details about the shallow TP crossover system you are thinking about. Who designed the speakers and crossover?"
Nobody... yet. I would be designing it myself. The design would consist of a high efficiency wide range driver used as a midrange (filler driver in this case) to join a woofer and tweeter in a TP configuration.
You can examine what different filter Q values you need for a given filter "Gamma" (overlap) using this cool little program created by an authority on transient perfect and transient perfect subtractive delay crossover designs: The (great) John Kreskovsky. He's one of the 'greats' not just because he's so darned smart, but because he shares so much of his work with the DIY community. (This goes for all of the contributors over at the frd consortium... THANKS guys!)
You can get the (filler driver) transient perfect designer here:
http://www.pvconsultants.com/audio/tp/tpdfil.htm
AP:"I have been using the kernel streaming output of Foobar mainly for playback, but I have successfully configured my PC so that every sound made by any application will go through the Thuneau crossover/arbitrator."
You said you need to use kernel streaming - I am guessing this is because the VST plugin for Thuneau Allocator is using ASIO to map the (8) output channels to the multi-channel ASIO card, meaning you can't use ASIO to map the (2) INPUT channels simultaneously. Honestly? I think ASIO is over-rated for playback applications. There are a number of ways to "bypass" windows k-mixer, and latency is not a major concern for non-live playback systems. Thanks for the info on you trials and tribulations (and methodology). It's nice when people share this information!
Here is another trick you can try. You can use a product called "Virtual Audio Cable" with an ASIO4ALL wrapper to map the sound from your app (Foobar) to the Allocator. I spent $50 on Virtual Audio Cable a while back - and I have not yet really NEEDED it for anything but it's just so darned cool I decided to get it "just in case" and support the very creative guy who wrote that software. Basically you can "interconnect" any two software programs via their I/O as if you were selecting soundcards - but you are not - you are simply tying (directly) the output of one app to the input of another and the sample rate / bit depth is controlled by your software. The "cable" does not interfere so long as the "requested" samplerate and bitdepth is within the range of settings that YOU can select (up to 24/192 I believe!)
Right now, I am using a (cheap but decent) M-Audio Revolution 7.1 which has ASIO capability to it's onboard DACs but no ASIO capability to it's SPDIF output. The analog outputs of this card are amazingly clean and sound really good - it's good enough (for now) for experimentation purposes. I really want a USB solution with I2S (not SPDIF) on the other end that is MULTICHANNEL so I can tri-amp, but the only folks doing more than 2 Channels with USB are the prosound/prosumer guys and with them it's hard to know how the thing will truly "sound" despite amazing specifications. I also do not want USB BUS power running my DACs. (I think if you're using USB bus power to drive your DACs in an external box you may as well use a good PCI card!!)
AP:"Also, Jan at Thuneau mentioned that he is coming out with a new software package called Frequency Affirmer that will do "very good room correction and crossover at the same time".
Like we're not going to try THAT software!! lol
I love this PC audio stuff!
For me, when using foobar, I prefer to do no upsampling *IF* I am doing convolution or PC crossover plugins.If not doing convolution, then using the PC for crossover functions is fine - quite good sounding actually.
If doing convolution, I prefer to use outboard passive crossovers (or at least an outboard digitial active crossover). It seems that once you start doing more and more *different* digital processes (aka plugins) it seems the "effect" of doing all this processing seems to be cumulative.
I (personally) would take convolution with the *right* impulse file any day over upsampling - even Secret Rabbit Code. Upsampling seems to have a high "penalty to performance" ratio at the best of times, where convolution offers far greater benefits and a surprisingly low penalty. In fact, I find I hear the improvements SO much with DRC that any deleterious effects are vastly outweighed.
Cheers,
Presto
Hi Presto> when using foobar, I prefer to do no upsampling *IF* I am doing convolution or PC crossover plugins. If not doing convolution, then using the PC for crossover functions is fine - quite good sounding actually.
~ Due to over high demands on the processor, causing pops etc?> If doing convolution, I prefer to use outboard passive crossovers (or at least an outboard digitial active crossover). It seems that once you start doing more and more *different* digital processes (aka plugins) it seems the "effect" of doing all this processing seems to be cumulative.
Another approach is room correction with a Behringer DEQ2496 street price $300??
www.behringer.com/DEQ2496/index.cfm?lang=ENG
which would free up resources for other things> Upsampling seems to have a high "penalty to performance" ratio at the best of times, where convolution offers far greater benefits and a surprisingly low penalty.
Is the penalty ~ pops etc?
--Another approach is room correction with a Behringer DEQ2496 street price $300??
-- www.behringer.com/DEQ2496/index.cfm?lang=ENG
-- which would free up resources for other thingsHi,
I own a Behringer DEQ2496. Performance in the digital domain is excellent when going from transport to DEQ2496 to external DAC. The DAC in the DEQ is not bad, but not outstanding either.
The DEQ only performs room correction so far as it changes the amplitude of the signal; it does nothing with time and phase as DRC does. The DEQ is very good at reducing room nodes caused by standing waves and can therefor offer good improvements in the bass. It has many flexible EQ stages, including an excellent parametric EQ, but that is as far as it goes.
Thanks for the veery informative postsPardon my ignorance, but is DRC ~ a plug in?
No,DRC is an acronym for digital room correction. There is also an open source software package called DRC that is used to create the listening room compensation correction filters that can be used by other programs to modify digital music on the fly so that it is correct for your room. You can check out the link below; just prepare to spend a week or two if you really want to learn more about it. You can also check out the wiki at http://www.duffroomcorrection.com/wiki/Main_Page
Alan
Thanks
Room correction
Replying this time with some specifics from my systems.When I had a Benchmark DAC1, which resamples all inputs to 110K internally via hardware, upsampling to 96K on my computer improved the sound (I guess one could say it improved the dynamics). I attributed this to the Secret Rabbit resampler doing a better job of upsampling (from 44K to 96K at least) than the DAC1 (although the DAC1 still had to go from 96K to 110K, the additional data provided by the Secret Rabbit resampler probably helped quite a bit).
When I switched to a Lavry DA10 I found that upsampling to 96K no longer improved the sound.
So the answer to this question probably depends on your DAC. Based on my experience it seems that DACs that are doing hardware resampling will benefit from some of the work being done via the Secret Rabbit code. But DACs that don't resample will probably suffer from resampling (because, in my experience resampling decreases the quality of the sound).
If someone or a magazine (is there an audiophile PC audio magazine or e-zine??) had the time & resources to do a comparison of a few of each of hardware & software upsamplingI just read an ealier thread poster that Secret Rabbit is demanding of processor, but does a better job than some. Better algorithm needs more grunt . . horses for courses
There is no doubt, it adds dynamics. The peaks are higher with upsampling, assuming the right algorithm.
Unless by "right algorithm" you mean a dynamic range expander.
Which algorithm do you find works 'best'?
The older SRC with Foobar 0.8.3.
Thanks
The dynamics are superior and it is smoother and more natural on vocals. The newer Foobar and SRC seem more harsh to me.
Cheers
No, its just the original SRC.
I thought (X)dBFS was (X)dBFS, but perhaps there's more to it?I understand that if you double the number of samples, there will be a greater number of samples at the predicted maximum excursions of the waveform by virtue of interpolation, but I was under the impression a D/A converter does that anyway. Besides, aren't we talking about extremely small differences for signals in the audible range? While I believe that extremely small differences can be perceived in many cases, I'm not sure how that relates to the issue of dynamics.
Maybe I'm missing something? Thanks.
-Anthony
The algorithms that interpolate between the actual sample points will often go higher and lower voltage than the actual sample points because of the "trajectory" of the waveform. For instance if there are two sample points at the peak of a transient (such as an impulse) on either side of the the maxima and they are both the same voltage, it is unlikely that the "missing" sample points would be at the same voltage. This would create a "clipped" or squared-off waveform, which is quite distorted and unnatural and would contain a lot of ultrasonic harmonics. The correct added sample points would be higher in voltage between two identical voltage points at the top of a transient "impulse" in order to "smooth-out" the top of the impulse and bring it down below ultrasonics in spectra.
I understand what you mean, but again--doesn't a D/A converter do that [the interpolation] anyway? I suppose some are better at it than others, but that's still the idea, right? If not, as the frequency of a signal approached the sampling rate, passably accurate reconstruction of the signal from the data would become highly problematic. In the simplest case as an example, when a 20kHz sine wave is digitized at 44.1kHz and then reconstructed, from what I've seen on an oscilloscope, it doesn't result in a clipped or unreasonably distorted waveform.I could be mistaken, but that's been my understanding. If I'm wrong, I certainly appreciate the information.
I believe most D/A chips also do a digital smoothing function, and this can actually result in clipping depending on the waveform and the DAC analog section and its headroom.It depends on the level of the signal and how many bits are involved in creating the waveform. The signal that you were looking at on the scope was probably in the volts RMS. At low levels, there are just a few bits involved, so the waveform is much more like a square-wave than a sine-wave, so the resampling algorithm can make a big difference in the amplitude and accuracy of this waveform, a bigger difference than the D/A.
Upsampling consists of creating (additional) new data out of existing data via a software algorithm that knows nothing about the particular type of sounds/music it is upsampling (other than the numbers in a buffer it uses while processing this data). It will sound different, but I'd be very surprised if a generic sound processing algorithm can create data that sounds as good as the musicians that created the original data.
But it's about doing what we can, with what we have
I have found on my system that upsampling creates 'air' at the expense of dynamics. That's why I don't use it.
Cheers
I have a TVC passive and a ribbon tweeter, and have plenty of 'air' to begin with. It just seems that to me that impact is slightly reduced with SRC. It doesn't sound bad. SRC definitely sounds better than SSRC, which sounded to me like a typical DSP.
I will be building a system with Aurum Cantus 1 ribbons, which do you use
Fountek with a magnesium Seas Excel in a sealed box.
We can always add information to, or change, a signal with a filter. Some day we may be able to turn a digital signal into a potato.
This post is made possible by the generous support of people like you and our sponsors: