Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Lossy Codecs and transparency (Read 5809 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Lossy Codecs and transparency

Couldn't find a good forum that was just about lossy codecs in general, so here goes.  Mods can move as they choose.

Lossy codecs and transparency are a great thing.  At 256K AAC, I have yet to find a song that I can ABX.  Admittedly, I have only tried a few dozen over the last year or so.  But I see no reason to continue.  I would rather enjoy my music, rather then fret about whether I can hear some minuscule artifact during critical listening.

But there is something I am curious about.  Supposedly, there are a small number songs that can be ABXed in a blind test.  I don''t dispute this.  I'm sure some small percentage of music doesn't lend itself to transparent lossy compression with the codec that the tester chooses to use.

Since I have yet to find a song that I can distinguish in an ABX test, I need to ask anyone that was able to ABX a lossy song compared to the lossless file they generated it from, the following question:

If you ABXed a FLAC file vs a say v0 vbr MP3 or a 320K MP3, and were able to tell the difference between the two, did you also try to create a file in a different lossy codec such as AAC or Vorbis, ABX that and then get different results in an ABX test?  Can you achieve transparency by changing codecs?

If anyone can name songs that are not transparent once lossy compressed, I'd be willing to try to do this test myself, assuming that I can tell the difference between the MP3 and the original file.

Re: Lossy Codecs and transparency

Reply #2 –
Interesting thread.  Thanks for linking to it.

Re: Lossy Codecs and transparency

Reply #3 –
Generally speaking, audio files which are at the edge of clipping (due to "loudness war") are much more prone to cause artifacts when  compressed by a lossy codec, see e.g. https://www.youtube.com/watch?v=BhA7Vy3OPbc

Re: Lossy Codecs and transparency

Reply #4 –
Quote from: butrus on
Generally speaking, audio files which are at the edge of clipping (due to "loudness war") are much more prone to cause artifacts when  compressed by a lossy codec
Having people wade through a 40-minute video that only tangentially touches upon the subject so that they may find something that constitutes actual corroboration is a bit too much to ask.

NB: I have seen this video before.  It offers no objective evidence that the chances of audible artifacts will increase as a result of heavy dynamic range compression.

Please summarize the points made that support your claim and at which times they can be found if you believe me to be mistaken.

Re: Lossy Codecs and transparency

Reply #5 –
Even if one can demonstrate this issue in ABX tests, the issue is caused by clipping, not the compression algorithm itself. In such cases a fair test should be performed by using replaygain.

Re: Lossy Codecs and transparency

Reply #6 –
Quote from: greynol on
NB: I have seen this video before.  It offers no objective evidence that the chances of audible artifacts will increase as a result of heavy dynamic range compression.

The problem is not the dynamic range compression itself, but the fact that it is used to make the signal much louder and very close to the clipping point. Lossy codecs can sometimes take an input signal that's close to clipping and make it actually clip. If you only have a handful of peaks that are close to clipping, that's usually not so much of a problem (especially since these tend to be non-tonal transients), but if your entire signal is close to the clipping point, you're not only much more likely to get clipping, but that clipping is also more likely to be in a tonal segment, where the distortion will be more easily audible.

Re: Lossy Codecs and transparency

Reply #7 –
Quote from: Porcus on
This one may interest you? https://hydrogenaud.io/index.php/topic,96108.0.html

Interesting sample. As far as I can tell, the main difficulty is the insanely close transients over the entire file. The Opus encoder only recognizes about half the frames as transients and just telling it "all frames are transients" improves the result quite a bit. I suspect that would be the same for AAC and Vorbis.

In the end, for rates like 192 kb/s and above, my opinion is that it's always an encoder issue Opus, Vorbis and AAC are all easily capable of being transparent at an *average* (over all music) of 192 kb/s. That means having good decisions from the transient detector. The difficulty is to not screw up anywhere. Also, if a segment is particularly hard, you can still increase the bitrate, but you have to realize it's hard.

Re: Lossy Codecs and transparency

Reply #8 –
Quote from: apastuszak on
If anyone can name songs that are not transparent once lossy compressed, I'd be willing to try to do this test myself, assuming that I can tell the difference between the MP3 and the original file.

I'd be interested in hearing people's thoughts about this too.

Last time I did an ABX test, I picked some tracks that I thought should be pretty revealing of codec issues, but not being a designer I may have totally missed the boat.  I focused on stuff with big dynamic range transients, complex high frequency content, and solo instruments with subtle overtones.  When testing, I pretty much wore myself out doing a 128k AAC comparison to the source.  I was - barely but consistently - able to detect a difference, but it was completely impossible to put into words what I thought I heard.  I thought I was listening for fine details, but I could never pinpoint a passage that was missing anything specific.  All I could say was "I think there's a difference" and I was almost always right (9/10 or 10/10 depending on the sample).

No, I didn't go to 15 or 20, because I was already developing a headache from straining to hear differences.  I'm really dubious about trying 192k because (a) I think it would be much harder if not impossible for me, and (b) the 128k test was so stressful, and I had to focus so intently to do the test, that it really wouldn't be enjoyable or worth the effort.  I've never heard any artifacts at 192k in normal listening, and I'm not hyper-focused when I do listen, so I can't really say I'm concerned at all.

If anyone can provide names of readily-available test tunes, I'll gladly re-run the test at least once.  I'm not averse to sharing the names of songs I used for a test if anyone cares to know.