[PhonoPhi]

I asked everyone so you can see that most people commenting here took the absolute minimal effort to do some listening tests before commenting. Why would you ever feel the need to comment on this topic if you clearly didn't try to do some testing before? I would love to hear anyone thoughts on the topic but only after they set up some form of blind testing and I think most people in this thread feel the same way.

I did.
But then I clearly feel that the point of all the recent comments to declare me "a troll", again similar behaviour to cable forums.

The proper answer would require quite a bit of writing (all the conditions, otherwise just the numbers are not very meaningful), and it would hardly make any difference in this conversation.
I actually mentioned the simple numbers in this thread before: 192 mp3 were my border line, above I would not bet my money in distinguishing.

Again, my much simpler point is that with the current cost of storage, what is the justification for lossy files that can't be further converted.

 

[VNandor]

That cleared most things up for me. Making a point about which format is better to use and not about which format sounds better is quite different.

 

[bigshot]

As for the real points, please kindly give us an example of what Bell labs did in 1920s that is directly relevant to the main points of this discussion.

The absolute thresholds of hearing, auditory masking and psychoacoustics- some of that was a little later than their earliest studies, but the work of Bell Labs is something everyone who is interested in the science of the reproduction of recorded music should research. Bell Labs laid the foundation for just about everything home audio is based upon. If you are interested in learning something new, see this... https://acousticstoday.org/wp-content/uploads/2016/03/Allen96.pdf

 

[Blackwoof]

Been trying Lame encoder more, V2 fine 99.5% of the time the stuff that breaks or needs more than V0. I just use Wavpack hybrid at 448kbps for those. Noticed stereo crush on MPC/Ogg at <150kbps & QAAC artifacts <150kbps with ambient with acoustic/synth sounds like a 32kbps MP3 would???.

 

[KeithPhantom]

Been trying Lame encoder more, V2 fine 99.5% of the time the stuff that breaks or needs more than V0. I just use Wavpack hybrid at 448kbps for those. Noticed stereo crush on MPC/Ogg at <150kbps & QAAC artifacts <150kbps with ambient with acoustic/synth sounds like a 32kbps MP3 would???.

I used to use FAAC and at 256 kbps there wasn't a single difference my ABX at that time could reveal. Nowadays, I'm thinking of compressing some of my FLAC files into AAC since I will need them to be smaller and I think 256 kbps Apple AAC should be enough to be virtually indistinguishable from lossless.

 

[Blackwoof]

I used to use FAAC and at 256 kbps there wasn't a single difference my ABX at that time could reveal. Nowadays, I'm thinking of compressing some of my FLAC files into AAC since I will need them to be smaller and I think 256 kbps Apple AAC should be enough to be virtually indistinguishable from lossless.

Just like i used to try Vorbis for a while at 175kbps then kinda binned it, When some samples needed 490kbps yet 224kbps AAC/MP3 was enough. I can't stand that puffy wind sound when it's bit rate starved when electro & noise rock/metal have noise burst sections.

 

[ev666il]

Hello,

I have a technical question about the mp3 format and figured I'd ask it in this thread.

If I understand correctly, the mp3 compression algorithms are based on psycho-acoustic models with particular emphasis on the masking effect: If a sound is present that would mask other sounds that play at the same time, those other sounds are removed. Is this correct (albeit perhaps a tad of an oversimplification)?

If I were to compress individual track stems for drums, bass, guitars and vocals, would less information/data potentially be removed from each track than if I compressed the song after combining the stems? The assumption is that with all stems playing simultaneously, there would more likely be masking in place.

 

[castleofargh]

Hello,

I have a technical question about the mp3 format and figured I'd ask it in this thread.

If I understand correctly, the mp3 compression algorithms are based on psycho-acoustic models with particular emphasis on the masking effect: If a sound is present that would mask other sounds that play at the same time, those other sounds are removed. Is this correct (albeit perhaps a tad of an oversimplification)?

If I were to compress individual track stems for drums, bass, guitars and vocals, would less information/data potentially be removed from each track than if I compressed the song after combining the stems? The assumption is that with all stems playing simultaneously, there would more likely be masking in place.

The general idea is correct. The difficulty with specific examples is to know if the algorithm takes them into account or not, and how far the psychoacoustic understanding is pushed. For example the model for 2 tones played simultaneously is pretty much perfectly understood. In term of amplitude and time, we can very well predict when one will mask the other or just attenuate the perceived loudness of the tone by X amount. But with more complicated signals, something supposedly masked because of the relation between 2 tones, can become audible again in some cases because of other tones kicking in at about the same time. So using the basic models and then extend them to more complex signals, just ends up not working, or at least not working all the time. For that reason, most encoders tend to play it safe and don't go as far as they probably could. For a high bitrate MP3 it's not rare to see close to no difference from 0 to -50dB. Even though we have psychoacoustics models where one tone would mask another as soon as -20 or -25dB when the frequencies are really close to each other.
So to not answer your second question, I have no idea if things would go as you imagined, because of the pretty huge "safety net" used at high bitrate.

 

[bigshot]

I think you're right Castle... I've spent the morning thinking about this... I'm not sure how compressing stems separately would sound compared to compressing them all at once in the final mix. Assuming they are all normalized, I suspect there might not be much difference at all. Thinking in pure theory, I would assume that if something is masked as a solo, it would be masked perhaps even more as an element buried in the mix. I'd certainly recommend not using compressed audio to mix with, but I don't think it would make any difference if the individual stems are compressed no further than the threshold of audible transparency.

 

[Blackwoof]

Hello,

I have a technical question about the mp3 format and figured I'd ask it in this thread.

If I understand correctly, the mp3 compression algorithms are based on psycho-acoustic models with particular emphasis on the masking effect: If a sound is present that would mask other sounds that play at the same time, those other sounds are removed. Is this correct (albeit perhaps a tad of an oversimplification)?

If I were to compress individual track stems for drums, bass, guitars and vocals, would less information/data potentially be removed from each track than if I compressed the song after combining the stems? The assumption is that with all stems playing simultaneously, there would more likely be masking in place.

That only one aspect, If say your guitar & drums are 85% of the song the codec will focus on that more than vocals & quieter stuff. But since those two are hard on the MP3 codec it could fail, Metal at V2 can pump out a 275kbps file if fast & quite noisy. With AAC/Vorbis the same rules applies but since they can handle 22KHz vs MP3s 16KHz, Have much more advanced model & can do up to 1mbit in VBR mode. You should not tell any difference if things go right, It why even with 320kbps MP3 can die with hard electronic samples because it can't do masking or chop off the treble without quality loss, This is why AAC/Vorbis can be 495kbps even at the 128 ~ 192kbps settings & have extra stuff to fake noise if they do cut the information.

 

[ev666il]

Thank you very much for your replies, @castleofargh , @bigshot , and @Blackwoof

To provide a little context: I own two albums that came with a bonus disc containing the individual track stems for each song in 192kbps mp3 format. I used Audacity to mix them together and saved them (in FLAC format to avoid a second pass of lossy compression) to get a final mix without the annoying Loudness War limiting.

Now, I do think it sounds pretty good (and I have been unable to tell FLAC from AAC/MP3 in the past; besides, my hearing pretty much caps at 16.5Khz)—but for my curiosity and peace of mind, I figured I’d ask my question. Thanks again!

 

[bigshot]

Balancing the mix might be more of a challenge than the compression. If they are stems, they might not be balanced to the mix. They may be flat level. That means you would have to mix each element riding it up and down as necessary throughout the whole song. It would be fun to play with though.

 

[ev666il]

It would be fun to play with though.

I can send you stems for a song or two if you'd like to try your hand at it

 

[bigshot]

That's OK. My plate is very full right now. Thanks for the offer though.

 

[Blackwoof]

Really love how even at 96kbps Opus is still transparent 99.5% of the time, On par with 160k AAC/Vorbis & 192Kbps Lame mp3. Could fit 12,500 Songs on my 128gb card since many of them are either <7 min or 56+ min Noise/Ambient/Noise metal freak outs, Which I'm still floored even possible at 96kbps since HA only expected above 128kbps for it. Wish xHE AAC support grows since that even better at 96kbps with Exhale 1.8.0

With USAC/xHE AAC cue better than FM stereo on AM stations, For the USA.