[
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2007, Audio Asylum, all rights reserved ]
[ General ] [ Speakers ] [ Tubes ] [ Vinyl ] [ Digital ] [ Hi-Rez ] [ Video Asylum ] [ Cables ] [ Tweaks/DIY ] [ Music ] [ Films ]
|
Home
/ FAQ
/ News Classifieds / Events |
Audio Asylum Thread Printer |
Get a view of an entire thread on one page |
69.200.83.8
In the other FLAC versus WAV thread, Primiano has demonstrated that he is able to record bit-perfect WAV files from the digital out of his soundcard when it's playing FLAC or WAV files. That's an important step, but it does NOT demonstrate that FLAC and WAV files will sound the same. The reason is that there could be enough jitter in the S/PDIF outputs to make an audible difference between the two, but without rising to the level of causing actual bit errors.
However there is a very simple test anyone interested in this can do in ten minutes. Simply connect your soundcard analogue out to your soundcard analogue in. Now play a WAV file and record the input, then play a FLAC version of the same file and record the input. Then download and run audiodiffmaker, which time-aligns the tracks and then outputs the DIFFERENCE between the two. If the difference file is silent, it is impossible that the two files were audibly different to begin with. If you doubt that, think about it - you're listening to the differences between the tracks against a background of silence, rather than listening to two loud tracks and trying to hear just that small difference. It's the auditory equivalent of looking for a needle in a haystack once someone has removed all the hay :-).
Now, I tried this both for my MacBook Pro laptop, using itunes playing ALAC versus AIFF (apple's version of WAV) and on my Windows desktop, with foobar playing FLAC versus WAV. In both cases the recording was done with Audacity, and audiodiffmaker (running on the desktop) created the difference files.
In both cases, the difference file consisted of noise at about -85dB (for the Mac) and around -80 dB for the windows machine (which incidentally demonstrates the slight superiority of the MacBook, at least for noise floor). The spectrum of the Mac difference signal was almost exactly white according to Audacity's spectrum analyzer, whereas in the case of the Windows recording there were a few little peaks (maybe 10 dB above the rest of the noise) at around 14 and 17 kHz. The meaning of that is unclear - the next step is to redo the recording, only using identical source files this time, and see if those little peaks are still there (I suspect they are). I'll do that later if there is any debate.
In any case, the difference between WAV and FLAC and between ALAC and AIFF playback on my computers is a nearly white noise component at around -80 dB. Given that 16 bit audio has a headroom of only 96dB I strongly suspect the difference between two successive recordings of the same WAV or FLAC file will be just the same (i.e. the noise is probably introduced by the D->A->D process) but if anyone is in doubt, I suggest you try it yourself (or I will when I have time later).
This definitely proves that there is no significant difference between WAV and FLAC playback on my computers (which are not particularly well optimized for audio anyway). I suggest to anyone that thinks they hear a difference that they try this simple experiment.
There is no such thing as a definitive test for some people.
If there are errors caused by logic induced modulation etc., then they may be offset by using analog loopback on the same audio card.
Also, -80 or -85dB is not enough resolution to really measure differences, which are likely to be lower.
A better idea is to send the analog out onto a completely different PC with a different soundcard. And you need one with high resolution, recording at 24 bits ideally with a noise floor lower than -110dB. Remember, there is no requirement to match the recording resolution to the source resolution (ie. it's better record in say 96/24 or even 192/24 and look for differences).
Sorry to be a pain on this, but controlled testing on this is a lot more difficult than it may seem.
In these comaprisons, not enough attention is given to hardware interface, hardware perforamnce, and interfacial link effects.
It is the digital data entering the dac that counts, and this is not just a computer or software issue.
I think you're missing the point. While greater recording resolution might be preferable in other circumstances, it isn't necessary in this case. Worse resolution or more noise makes the difference signal MORE audible, not less, but the difference signal in this case was close to inaudible (over headphones at normal volume). To really hear it I had to max the volume - a level that would have been on the verge of painful with the original track.
There is simply no way that difference could have been audible when added to the original full-range musical signal. That's the nice thing about audiodiffmaker - while an audible difference signal does not prove the difference was audible to begin with, an inaudible difference signal DOES prove beyond doubt that the signals were indistinguishable. Again, noisy or low-resolution recording simply makes an inaudible difference signal that much harder to attain, therefore making the result even stronger.
*** I think you're missing the point. ***
No, I think you are missing the point.
You are assuming because you have captured at low resolution and can barely hear the difference, therefore there must be no difference.
However, that could also be because the capture resolution is not enough to allow any differences to be accurately captured. You've already stated the noise floor was around -80 to -85dB - this is way too high and will mask any differences below that (and yes, our ears have the ability to hear structured sounds well below the so called noise floor of a room).
Also, it may be that the difference manifests itself further down the signal chain, as Dawnrazor pointed out. For example, higher CPU loading whilst playing FLAC may cause more EMI to be generated which affects the power amp (even though the signal coming out of the PC itself may be relatively unaffected). Your test will not reveal such differences, nor differences caused by (for example) unshielded speaker cables being affected by the generated EMI.
I'm not trying to attack you, I'm simply pointing out that your results do not prove or disprove the hypothesis that there may be audible differences.
The question I have answered is: Is there an audible difference between the analogue outputs of my soundcard, through my headphones, when playing FLAC versus WAV? If you want to know if EMI from FLAC decoding affects your amplifier, do it yourself - simply record the ouputs of your amp and use audiodiffmaker. I'm not going to waste my time - and frankly if the tiny amount of increased EMI from FLAC decoding affects your speaker cables or amp, buy better ones, because so will your cordless phones, switchmode power supplies, dish washers, microwaves, neighbors, etc. etc. etc.
Now, back to the question I have answered. The fact that the noise level was higher than it could be (although not by much - these were 16 bit files, with a theoretical best 96dB of headroom) makes it EASIER to hear differences. You don't seem to be getting that. The noise added in this process is RANDOM - therefore two recordings will differ by an amount proportional to the noise even if they were in other respects identical. If the difference is inaudible even with that added noise, it was certainly inaudible to begin with.
If you like, I can say that more precisely: the result of my experiment is that the difference between FLAC and WAV playback on my system is AT MOST a white noise component -80dB down from the signal. A better experiment could constrain that even better, but that's more than good enough for me.
*** The fact that the noise level was higher than it could be (although not by much - these were 16 bit files, with a theoretical best 96dB of headroom) makes it EASIER to hear differences. ***
I'm sorry, but what you are saying does not make sense.
Think of it this way, As you say, CD resolution is 16 bits, which correspond to 96dB of dynamic range. A noise floor of -85dB corresponds to an effective resolution of 14 bits.
Also, every 6dB difference in noise corresponds to a doubling of noise level. Another way of explaining this is a system with a noise floor of say -84dB has FOUR times the minimum noise level required to reproduce CDs at full resolution.
In other words, the system you used for your test cannot even render wav files at full resolution, so why would you think that it would be useful (much less "definitive") in detecting differences between flac and wav?
*** If you like, I can say that more precisely: the result of my experiment is that the difference between FLAC and WAV playback on my system is AT MOST a white noise component -80dB down from the signal. ***
-80dB is effectively 13.5 bits of resolution. If that is good enough for you, why bother listening to wav at all, you might as well listen to mp3.
Put it another way, your test probably won't even be able to detect any differences between wav and high bitrate mp3. Or even between two wav files differing only in the last two bits. So exactly why is this test "definitive"?
Hey Christine,
If I took my regular computer and recorded the output of the dac attached to it with my other rig using my Lynx 2b, would that work?
Or is the regular pc chain a problem with a pc that upsamples and only puts out 48k fed into a dac that upsamples to 96 or 192?
ALso would a recording from a mike work even if it was on the same PC? Reading all the posts from AL Sekela and how some speakers (like my maggies) can be rf antennas, and how some amps may not take kindly to RF, perhaps we are looking in the wrong place? That sure could explain why there is no consensus. Then someone who hears differences can run the test.
*** ALso would a recording from a mike work even if it was on the same PC? ***
A mic recording the output from the speakers would be nice, but preferably on a separate PC, powered by a separate power circuit. Something like an AT822 recording onto a laptop running on battery power and a good sound card/USB recording device (also battery powered) will be good. What we need a mic capable of at least 120dB dynamic range. Problem is, we will need to play the music pretty loud, to capture the maximum possible dynamic range and resolution.
You are definitely on the right track though - the differences may not be in the analog output of the soundcard but could potentially be further down the chain (particularly with the EMI hypothesis).
It's nice to see that you are trying to maintain an open mind. That's a very valuable attitude to have in this hobby.
I would like to think all of us here had open minds...I mean we did embrace the PC route, and well I must think that most of Audiophilia are close minded about THAT choice.
ANyhow, I think I am close to being able to do the test you are discussing...sort of.
I have dedicated ac runs , and if that is OK, can power the computers and audio components seperately.
I do have a mic, but not sure if it is good enough. It is the behringer ECM 8000. I bought 2 for DRC, but haven't used them yet. I still need a mic pre.
Even if those components are correct, there are some things that are a concern. First, my 2nd pc is not that good as an output device and my dac is not as resolving as my Lynx card. It would be ideal if I had 2 of them. 2nd, I don't know if I can hear any differences since I never play Flac on the main rig, and the normal pc is not resolving enough to know, and I don't have any of the files I am familiar with there.
It would be best if someone who can hear the differences can do the test.
I also use an SB3, but synchronously reclocked in the DAC (low jitter clock in the DAC, SB3 synced to that clock). I CAN hear a difference between streaming flac or wav, I cannot hear difference between flac or wav on the server. The server is in a different room and power branch from the stereo system.This surprised my, the reclocking down in the DAC should prevent any jitter born difference from showing up. I did bit compares on the data actually going in to the DAC chip and its identical in all casses, so thats not the issue.
I have a pretty good spectrum analyzer that I use for looking at phase noise of clock signals, I can see very small changes in jitter with this, I can see changes that I cannot hear. Looking at the clock going into the DAC chips I see no difference in the spectrum when streaming flac or wav. This seems to indicate that the reclocking is working well.
So where is the difference coming from? The only two things I can think of are EMI radiated through space or noise on the power line. I haven't come up with some definitive tests on this so I don't have hard data, but I do have a hypothesis:
I've been finding out that most power supply transformers have resonances in the 100KHz to 500KHz range, some have very strong resonances and others pretty week. If you have a piece of gear with a transformer with a strong resonance its very susceptible to noise on the power line near that frequency. (intersetingly enough "high quality" transformers tend to have much stronger resonances, the core losses in cheap transformers tend to damp the resonances) The hypothesis is that the flac decoding process is generating some extra noise in this frequency range that gets coupled to the power line, exciting these resonances in some gear.
In order for you to hear the difference your gear has to have a resonance that matches that prodeced by the computer, its going to be very gear and computer dependant. Some will have it, some will not. Using a different computer might change the effect.
Ive used the spectrum analyzer to measure the the power line that a computer is plugged into and I CAN see significant changes in the noise on the line depending on what the computer is doing.I tried a laptop and a desktop and the laptop injected significantly less noise into the line. So there is at least some possibility that whats happening in the computer could be effecting your gear independantly of the "bits" comming out of the computer, or even the jitter on the signal.
One intersting aspect of this frequency range that the transformers are sensitive to is that it can sail right through our usual attemps at "power conditioning". The frequency is high enough that it sails right through most "line frequncy" transformers and chokes, but low enough that most RF filters don't touch it.
There ARE methods to damp the resonances in transformers, but they are almost never used. I have never seen it applied in any of the commercial gear I've looked at. Part of the problem is that its transformer dependant and not so easy to measure the proper electrical paramters to compute the right damping network. I'm in the process of doing this with my own equipment, I'm going to repeat the test after I've treated all the parts of the system and see if I can still hear a difference.
Thats pretty much where it stands right now. Sometime in the future I want to do some tests measuring the output of the power transformers of different parts of the system and see if I can see a difference when streaming flac or wav. And try it with or without the damping networks in place.
John S.
Hi,
Interesting observation. I'd like to look into this and would like to start by trying to duplicate your results. Would you mind describing in a bit more detail what was resonating and how you measured it? I presume these are "linear" supplies rather than switchers.
Thanks, Rick
Your post is interesting. On my system, the digital and analog boxes are powered separately by two ac regenerators, and my computer based transport, when powered, is suppled direct from the mains via a transistor follower regulated dc supply, in an attempt to isolate the PC from the Mains and from the regenerators.
However, poweering components on and off has an effect on the sound; and powering the computer (laptop based destop) has a greater effect.
The system has been set up with an emc meter to space components apart from each other abut the greatest offending item is the satellite receiver (about 2.3 GHz) which needs to be turned off for audio replay. Unfortunately, I have a Sky system which needes rebooting every time I disconnect the box from the mains.
So, even with great care and fanaticism, it is difficult to have a system which is immune to mains and emc effects.
The SB3 itself is probably a generator of emc itself. I have found it beneficial switching my broadband modem, home nertwork and computers off completing when listening crtical to the hifi system.
Thanks John for taking the time to post your results.
Your theory seems plausible, and does seem to account for why some hear it, and why some don't.
It sounds also like you are saying that in some cases, the better the gear, the more likely one is to hear such differences. Is this correct?
HOw does ones processor speed factor in to this?
DO dedicated lines help?
Dedicated and isolated lines do help, but they don't totally isolate. Listening to my VIA 1.3G with RME card whilst using my download computer is not ideal. The sound is a lot coarser than with PC and modem OFF. The VIA is powered by an isolating transformer, and the PC thru a filtered Belkin box.
The Behringer ECM800 is a good measurement mic (I have one myself) but doesn't have a high dynamic range, so the measurements may be limited by the noise floor of the mic (I suspect it uses internal equalization to achieve a flat frequency response).
I'm not really advocating doing a test - in fact the opposite. I would only recommend doing a test if you are pretty sure you hear a difference and wanted to confirm. If you don't hear a difference, just relax and enjoy the music.
Remember: a system that is working well should NOT exhibit a difference playing back WAV and FLAC. So if a difference is noticed, it actually indicates there is a problem that needs to be fixed - it's certainly not something to be desired.
since the sound output device is what clocks the data to make a signal
and primiano showed that the data is the same
isnt this irrelevant if you can use the same output device for FLAC and WAV?
The only (semi-)reasonable hypothesis is that the increased or different processing tasks the computer is performing during FLAC versus WAV playback affect the jitter spectrum differently, thereby producing a different analogue signal in the two cases (remember that the analogue out of a DAC is determined not just by the digital data, but also by the timing with which the bits arrive). My test was an attempt (successful, I think) to demonstrate that's not the case, at least on my system.
nt
P.S. - I forgot to add that I found a loopback cable which eliminated the high-pitched tone I mentioned in the other thread. Bad news for the Tascam, but good for this experiment.
Top of Page ]
[ Contact Us ]
[ Support/Wish List ]
[ Copyright © 2007, Audio Asylum, all rights reserved ]