Good question. Be interested to know if this is the case.
It’s not noticeable on every track. But songs with impact and scale. A good example is Wilco’s live version Misunderstood on Kicking Television. Try it and see what you think.
You listen to crappy Wilco on your highly resolving system? With volume leveling set to auto, Roon reduces the volume of that track by -4.5db. Running a quick test with a decibel meter on my iphone, the dynamic range looks the same with volume leveling enabled vs disabled - specifically looking at the section of music starting around 4:20. If I adjust the volume on my preamp to match the two, the graphs appear damn near identical.
I’d like to hear from the Roon admins or developers if there is any compression going on with volume leveling before I listen to any more Wilco. I’m not saying you are wrong or right, I’d just like to hear from the people who wrote the code.
Yes! And the Velvet Underground is bone chilling! Another good one for dynamics is Muddy Waters Folk Singer. I match the volume when doing these comparisons. My wife though has golden ears. She will come into the room and right away tell something abs sure enough… it’s a blessing and a curse. Ignorance is bliss.
You redeemed yourself with Muddy Waters Folk Singer.
I listened to Wilco per your suggestion. It was difficult to get precise volume match using my iPhone. Additionally, resetting the volume and turning leveling on and off to took time which made a direct AB comparison difficult. However, based on my very crude assessment, you may be onto something. I have leveling set to Auto and -15LUFS. With that particular track, as you stated, volume leveling significantly reduces the signal, -8.3 dB. Hmmm. Went into this a skeptic, came out confused.
Curious - is your source Qobuz, Tidal, or do you own the album in your library? If it is Qobuz or Tidal, did you import it into your library first? Do you tell Roon to play the whole album or just the first track?
When I set my volume leveling to Auto and -15UFS, and then play the whole album, the first track gets adjusted by -5.5db. If I just play the first track, it’s only -2.4db. My source is Qobuz 44.1/16 and I have it (temporarily) imported into my library. In settings → library, I have “use replay gain when available” enabled and “replay gain settings override Roon’s audio analysis” disabled.
According to Roon, volume level adjustments are just that - multiplying the signal by a constant gain value. It should not be compressing dynamic range or “sucking the life out of the music.”
However, there are some corner cases where volume level may affect sound quality.
Maybe something like this is happening here, especially if it is applying -8.3db to this track (that has a large dynamic range) in your setup? A large adjustment might take you out of the sweet spot for the DAC (significant bits) or pre-amp/amp as suggested above.
You can always disable volume leveling to avoid any mystery and send the signal unmolested directly to your DAC. I’m a fan of this approach but I find the feature very useful and leave it enabled.
I’m going to delete this album from my library now
I have found all volume levelling to affect what I hear for the worse. I first experienced it in iTunes, then Roon then BluOS. I have used various set ups from valve to solid state and active. Can’t explain it but I hear it.
Like any DSP, volume leveling is not lossless. Depending on the quality of the DSP (and not only), SQ degradation may or may not be audible. For example, Roon has optimized DSP for version 1.8 and my system feels the difference from previous versions.
In addition, most DACs apply DSP anyway (at least upsampling, usually). If you also use the Roon DSP, this means that the digital signal goes through two DSPs, which have a cumulative impact on the SQ.
How is volume leveling lossy?
Roon volume leveling is meaning:
- bit depth conversion from initial value to 64 bit - may or may not be lossless, depending on the initial value
- altering the amplitude of the audio wave - it is not lossless
- bit depth conversion from 64 bits to the value accepted by DAC - may or may not be lossless, depending on the value accepted by DAC.
An example of the worst situations: 24-bit FLAC and 24-bit DAC - volume leveling will contain 3 transformations, none lossless (because 64 is not exactly a multiple of 24).
Moreover, the adjustment gain values are not multiples of 2, which means the alteration of the audio wave amplitude implies rounding in mathematical calculations. This means that by applying the reverse process, you cannot get the original version, so the process is not lossless.
I’ve always taken “lossy” to mean information is being discarded in order to make the data set smaller, not math rounding errors 24 bits or more below what my calculator can process (and what I can hear). If that’s considered lossy, then all digital audio is lossy, right?
Volume leveling is not “bit perfect” but I still don’t believe it changes the dynamic range of the audio, except for a some corner cases described in Roon’s documentation above.
The maximum error will be in the least significant bit of the 24bit resolution, i.e. less than what the noise of the best analog amplifier adds as error to the acoustic signal. At least for me not audible as degradation. A mosquito wing flap at a distance of 20m is probably much louder.
What I wrote in my previous post is a very minimalist way of looking at conversions (but enough to show that they are not lossless). The actual processing involves many more calculations and approximations / roundings. All of them can contribute to the alteration of the initial shape of the sinusoids, in many ways. Depending on the quality of the respective DSP (and other factors), the negative effect may or may not be noticeable. For example, I easily notice the effect on Roon versions before 1.8, but at 1.8 I notice it quite hard.
I haven’t tried it again with 1.8.
Didn’t pay that close attention. Could haven been Qobuz or Tidal, I have both. Did not import. Just played the one track. I did do with and without leveling to compare. Just to be clear, I’m not saying I absolutely heard a degradation with volume leveling. An AB comparison was difficult with changing settings and adjusting the volume to match. Imprecise at best. However maybe, just maybe, I perceived a slight degradation with “leveling” on that particular track. The track itself was not, to my ear, the best for picking out detailed minutia. I did this particular “test” on my main system. Gaia to Terminator into my Levinson preamp and mono blocks into BW802D3s. Maybe I will try to same experiment on my tube set up. Was skeptical that there would be any change. Now I’m a little less skeptical.
I also feel a slight decrease in dynamics and liveliness when using volume levelling, I wish @jussi_laako and Roon team could fix so HQPlayer could do the volume levelling instead (send volume information before tune).
Just tested again with Roon 1.8 and it’s much better than in previous versions. But still noticeable as mentioned by others in this thread.
Yes, Roon improved DSP in 1.8, but its still not as good as for example DSP in HQPlayer (I use all my DSP in HQPlayer, Roon is basically bit-perfect transporter).