Do you normalize the volume on your audio tracks, and if so what are the best practices to do this without affecting the audio quality?
I have playlists which have tracks from different recording sources (e.g. a CD quality FLAC, then a Hi-Res FLAC, then a Vinyl recording). The CD recordings from older CDs are quieter than more recently mastered files (and also usually have higher DR numbers), but in particular it’s the vinyl recordings that are really quiet in comparison to the others. It would be nice to not have to adjust the volume between songs.
My understanding is that volume normalisation in the digital domain may have an adverse effect on sound quality, though whether audible to a particular person in a particular system is another matter - and maybe if you are playing isolated tracks like that any adverse effect is less of a concern than the frequent changes in loudness.
It is possible for VN to be done on the fly, with a setting recorded with each track so that when it plays it adjusts itself, or it can be done permanently when saving the track, altering the file itself. If the software allows the two options, the former has the advantage that you can always revert to normal. Whether you can have it set so VN is applied only when playing via a playlist and not when playing as an complete album, so preserving sound quality for the latter I have no idea - again depends on the software.
You could make a copy of a few albums in a different location or with modified names/metadata to test the effect, especially if the software you are using might result in a permanent change.
There is an effective volume levelling option in Roon which might be worth considering. There may be a small price to pay in sound quality, but for me, this is less important in playlists as they are more for background music, and it’s unlikely to be required when listening to a whole album.
I sometimes use Audition (Adobe) and normalise to 95% or so. I can’t say I have found any sound quality issues - it is a purely mathematical process, AIUI. I think that it finds the loudest recorded value, works out what the ratio is between that and full volume (or in my settings 95% of full volume) and then raises all other values by the same ratio. I can’t see how that would affect SQ. However, I have found no way to apply this to each file without opening, editing and saving each one, which is a pain. So I only do it for particularly quiet pieces. Arguably, though, it shouldn’t be done across the board - some music should be quiet, particularly classical pieces. Normalising a whole concert/album/LP/CD would be better than individual tracks.
There is no point in ‘normalising’ by a standard amount of change. Yes the process is mathematical, but it is always a lossy process (except the specific case when you upscale the bit depth and reduce the volume by an integer number of 6dB intervals).
When normalising tracks to give an ‘even’ apparent volume setting across different tracks, you can minimise the loss by upsampling to a greater bit depth (e.g. by upsampling from 16 bit to 24bit before applying the volume correction) but even then the process is still lossy to some degree.
IIRC dBPoweramp has an option to set something called Replay Gain when ripping an album, as a tag without altering the sound data, which assesses the average (or whatever) level notes the adjustment factor needed to normalise, and then subsequently when playing the file some software (Audirvana is one) can be set to use that to adjust the volume. I know no more as I have not used it and haven’t looked into it because as I almost invariably play by album and have remote control over volume the need to sometimes adjust it manually is a non-issue,
I use Volume normalisation in Roon as using Qobuz and Tidal the volume can differ massively if your using its Radio feature which is track based. It upsamples to 64bit before applying any leveling. Tidal provides metadata to do this based on each album/ track but Qobuz does not. Roon analyses ripped files when ingested and gives a dynamic range and uses this to base on the reduction/ increase required or can read embedded replay gain tags . It’s all non
destructive of your files as you can choose to apply this or not. Can’t say I have noticed any difference in sq when using it. YMMV and depends on your listening habits if its useful or not.
Why is volume normalization at the file level destructive? If we’re just mathematically adjusting the amplitude up and not crossing a level where the file would clip, wouldn’t that be considered non-destructive? I’ve seen it mentioned as being destructive in several places, so I’m curious to understand why this is any different to just turning up the volume on the system.
Re roon… I’m not excited about paying another annual subscription (particularly a very high one) so I’ll pass on this one. It’s also quite resource intensive on the NAS and asks for a dedicated SSD.
ReplayGain seems like it has potential though… I ran a test and it works but it needs to be supported by the media server being used: Synology’s default Media Server doesn’t do it. Asset uPnP works, but it has to re-encode it to a WAV before sending it to the SuperUniti. It upscales to a 24bit WAV at least here. And it seems MinimServer can do it (also with re-encoding to a 24bit WAV) but there’s no GUI to access the configuration so I haven’t figured out that part yet.
Does the audio quality get affected with the above conversion to a 24bit WAV and application of the ReplayGain values?
A couple years ago, when I was using Audirvana+ as a player, I used its features to set Album Replay Gain (not sure of exact spelling) as a metadata field. This I believe was a number of dB change to the replay to achieve a standard normalized level for the album. While the capability was there to set replay gain on a track by track basis, I never felt that was appropriate. Preferences could be set to alter the normalization level, and I think the algorithm (not sure). My understanding was that tracks where the replay gain was not set, these tracks were played unaltered. So I processed only the offending albums, leaving well mastered albums alone. As music was playing over many months, I would make note of every album that made me reach for the remote to reduce the volume - then when in the mood, I would process these offending titles. I wish I had this capability in the Naim world. Some music that we find entertaining, lacking the audio quality of better recordings, does benefit from this ”digital abuse.”
Digital volume adjustment is always destructive unless performed in integer units of 6dB. It doesn’t matter whether it’s calculated in the source data or in hardware - the maths is the same. Upscaling to 24bit helps, but there’s still some loss.
Yes, in theory; but the effect may be so small as to not be noticeable (or it may be large enough to be noticed reasonably easily, this depends on a lot of factors!).
As a little bit of background, as to “why” it’s destructive (for typical audio files on computers and streamers):
The sound files don’t contain “the music/audio” and “the (maximum) volume”, where you could adjust the latter without changing the former. Adjusting it “in the files” is NOT the same like cranking up or down the volume of your analoge amplifier. (Given, it would be an “ideal” amplifier, which does not distort the signal at whatever level.)
Instead, there just a “relative volume” in the files, from “0 to max”. Across that range for the amplitude, you have a discrete number of “steps” the volume can have. The number depends on the “resolution” of the digitization of the signal. (For most formats, this is the “bits per sample” like 8bit for ISDN telephony, 16 bit for CD/audio level, and 24+bit for high-res audio.)
At playback, the digital system is converted to an analogue signal via the DAC - also this step (the DAC) has a “maximum volume” it will apply, if the digital content indicates “max” - otherwise it will generate lower volumes. Any “real world loudness” comes afterwards via amplification in pre/power amps. I.e. the “computer part” has no value, about actual “absolute loudness in the analogue world”.
So, if you change “volume” inside the digital world: you scale the amplitude within the digital steps you have between 0 and max. Since it’s discrete steps, an increase like “10%” will not find a perfect match value in the available given values. So, you have to “round” to the next matching value.
Imagine you have a black-white image on a chessboard, where each square can be only black or white. If you scale the image by 50%, it won’t match perfectly onto the same chessboard, even if you had a larger one. Since you don’t have a larger board, making the singling too loud would “clip” - that’s when you “loudest” black square would not fit into the board any more. If you regulate power down, you have to “squeeze” the same picture information into less squares, and will loose even more information.
Using an upscaled format means, you have more “chessboard squares” available, to better adjust the scaled picture into it. Hence this helps.
(Same thing applies to any digital processing on the sound signal - equalizers, adjustment for the play-back device, be it a speaker or head-phone, … - hence, any digital adjustment is practically always incurring a loss of information. It might still sound better. )
Thanks, that makes sense now. I did my first test with a quiet vinyl recording of Leonard Cohen (album gain was +4.83) and I can definitely hear the difference on the lower end frequencies. The unaltered one has more of an open sound stage and generally sounds enjoyable to listen to. Ahhhh I’m so upset! lol
Next test… I’m going to create a copy with a manually applied increase of 6dB and see how that compares to the original. If it’s good then I think that will tide me over: there are a few quiet recordings in my collection that would benefit from the bump up in volume and I can leave the rest as they’re close enough.