More > JRiver Media Center 21 for Windows
NEW: Volume Leveling uses the additional Headroom provided by Internal Volume
mwillems:
--- Quote from: lisbethfox on August 14, 2015, 01:23:30 pm ---Ok interesting.
You seem to know what you're talking about. I'm a rookie at this gain structure stuff, I don't have much industry experience, just access to some fun content and the occasional indie job.
--- End quote ---
If you're curious JRiver actually has a spec-compliant R128 audio analyzer built in. You can run on it various videos and see "how they did." The only hitch is that JRiver's calibration target is (as I recall) 83dB at -20dBFS, rather than 85dB so you have to mentally add +2dB to JRiver's calculated volume leveling adjustment to translate it.
--- Quote ---I have often run into recordings without that extra headroom and craving to turn it up louder (some classical SACD's are quiet). I guess my next question for you is, JRiver does all this stuff at 64bits? It says processing internally, 64bits. If so that pretty much removes any chance of the gain changes affecting quality? (at least from what I understand)
--- End quote ---
Yes, JRiver's internal volume control is a 64-bit volume control, so there's no degradation during the DSP chain, but it will be dithered down to 16 or 24 bits at the output (depending on your DAC). So any digital attenuation does (necessarily) have the effect of reducing the theoretical dynamic range, but for a few reasons that reduction is likely to be either inaudible or better than the alternative.
For example, if you have a 24-bit DAC with a dynamic range of 110dB, and set your reference level to 80%, you'll be "losing" 10dB of dynamic range so at reference level you'll "only" have 100dB of dynamic range. Given that the highest possible peak in an 85dB calibrated system will be the LFE at about 115dB (because it's 10dB higher than everything else), your "loss" of dynamic range means you'd lose some theoretical sounds that are 15dB or quieter.
You can test how quiet that is by playing back near-peak audio at 1% internal volume in JRiver (1% is -97dB). Whatever you hear there is what you'd likely be losing by taking your reference level down -10dB. It's also an open question, even if you hear it during the test, whether you'd hear it with full volume signal roaring over top of it.
And that's all assuming that analog volume controls are all perfectly quiet and contribute no distortion. A well-designed one will have a very, very low distortion profile, but no circuit (even a passive one) can contribute 0 distortion. And not all volume controls are well-designed (some are surprisingly noisy). So you may also "gain" by getting some of the boxes out of your chain (no pun intended).
So bottom line, there are necessary technical costs to digital attenuation, especially significant digital attenuation, but they rarely have any practical significance because you only lose the very quietest parts of the sound which are typically inaudible especially during playback. And the alternative (analog controls) are not necessarily perfect either.
mwillems:
So I just had a quick back and forth with one of the devs (Hendrik) and his conclusion was that this was not a bug, but a natural side effect of the kind of DSP processing being done.
What I knew going in to the discussion was that minimum phase DSP (like JRiver's DSP) that cuts frequency will also affect phase relationships; what I hadn't thought about (that Hendrik pointed out) is that changing the phase relationships can change how the different frequencies sum up resulting in certain frequencies that used to cancel or partially cancel now summing. The result is that it can push the peak up slightly in actual program material (which effect is largely but not completely offset by the volume cut by the filter itself). This effect would not be noticeable when testing with single tone sine waves, noise, or with a swept tone.
If you google "why does cutting frequencies increase the volume of a master" you'll find lots of folks grappling with this exact issue, including a particularly good one Hendrik found:
https://www.gearslutz.com/board/7793641-post5.html
You can also see a neat illustration of how phase relationships affect peak level in the following article.
http://clas.mq.edu.au/speech/acoustics/waveforms/adding_waveforms.html
And some cases of people seeing 4 and 5dB boosts from high pass filters:
https://www.soundonsound.com/sos/dec10/articles/qa-1210-5.htm
https://sound.stackexchange.com/questions/24981/why-is-my-cutting-equalizer-increasing-output-level
So the short answer is things are probably working as intended, and we all need to leave some headroom when doing phase affecting DSP of any kind, even cutting frequencies. I never ran up against this as it doesn't show up in the kinds of measurements I do, and I always leave myself plenty of internal volume headroom. Always glad to learn something new!
mattkhan:
@lisbeth can you share a track that provokes this effect? If you have many, pick the worst one.
mattkhan:
--- Quote from: mwillems on August 14, 2015, 01:45:15 pm ---The only hitch is that JRiver's calibration target is (as I recall) 83dB at -20dBFS, rather than 85dB so you have to mentally add +2dB to JRiver's calculated volume leveling adjustment to translate it.
--- End quote ---
This is technically correct for a band limited signal as per SMPTE RP200. I think you should be careful to distinguish between gain structure and level setting, they are obviously related but still distinct things (so you might end up with suboptimal gain structure to ensure relative levels are correct)
mwillems:
--- Quote from: mattkhan on August 14, 2015, 04:28:25 pm ---This is technically correct for a band limited signal as per SMPTE RP200. I think you should be careful to distinguish between gain structure and level setting, they are obviously related but still distinct things (so you might end up with suboptimal gain structure to ensure relative levels are correct)
--- End quote ---
I'm not sure exactly what you mean? I'm familiar with the difference between gain structure and level setting, but I'm uncertain what you mean in this context? Obviously you don't want to introduce unnecessary amplification only to attenuate it back down, if that's what you mean?
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version