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: How close should a PCM be to an mp3 w/o being a decoded one? (Read 2242 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

How close should a PCM be to an mp3 w/o being a decoded one?

I ran a bitcompare on a set of mp3s against FLACs of an album where I suspected transcodes were uploaded to Bandcamp. (Bandcamp only accepts uploads in lossless files, but can not control the source.)
It is a various artists compilation, so it could very well be "some but not all".

fb2k reported a few of them to be in a range around 0.0001221. (And a couple at 0.08, and at least one at 2.)

What to learn from this? (Roundoff errors are smaller than the -40 dB mark?)

How close should a PCM be to an mp3 w/o being a decoded one?

Reply #1 –
I don't know what those numbers mean but I don't see how you can bit-compare an MP3 to a FLAC...  Maybe if you have the same exact MP3 file that was up-converted to FLAC.

Quote
How close should a PCM be to an mp3 w/o being a decoded one?
Generally, every sample of the decompressed MP3 is going to be mathematically different from the uncompressed original (by 1 bit or more).   

You can do a subtraction with an audio editor and listen to the difference, and/or check the amplitude of the difference.  But, but remember that the sound of the difference is not the same as the difference in the sound.  The simplest way to demonstrate that is by delaying one file by about 1 millisecond or so, which makes absolutely no difference in the sound, but it creates a "big-loud" difference file when subtracted.  In fact with a simple delay, the amplitude of the difference is often greater than the amplitude of the original files.

How close should a PCM be to an mp3 w/o being a decoded one?

Reply #2 –
Typical mid-bitrate MP3 has an RMS error of about -25 to -30 dB relative to lossless, give or take some depending on the bitrate you use.

How close should a PCM be to an mp3 w/o being a decoded one?

Reply #3 –
I suspected lossy-sourced FLACS; mp3-->PCM. Not the other way around: not that the mp3s were encoded from the lossless files. And some of them are so nearly the same that the latter is hard to believe; - 40 dB is peak difference on some of them.

The label made released the files for free download at Mediafire, http://www.mediafire.com/download/npasml8h...apes%282%29.zip , and at Bandcamp, https://grimtown.bandcamp.com/album/caesare...onfession-tapes (I have not redownloaded and compared to my own files).

IIRC the mp3s first. And they have done some suboptimal mp3'ing before (the comment at http://grimtownmusic.com/2014/01/26/the-re...mtown/#comments is mine, based on what I learned at HA ...), so I wouldn't be surprised if I learned that they created mp3s, went to Bandcamp, found out that Bandcamp requires lossless files, decoded the mp3s, and uploaded them - that was what I suspected and the reason why I fired up the foo_bitcompare. If there were differences at the LSB level, I would have settled for that, but a diff peaking at -40 dB would mean an unusually large round-off error ... or what?

How close should a PCM be to an mp3 w/o being a decoded one?

Reply #4 –
Quote from: Porcus on
I ran a bitcompare on a set of mp3s against FLACs of an album where I suspected transcodes were uploaded to Bandcamp. (Bandcamp only accepts uploads in lossless files, but can not control the source.)
It is a various artists compilation, so it could very well be "some but not all".

fb2k reported a few of them to be in a range around 0.0001221. (And a couple at 0.08, and at least one at 2.)

What to learn from this? (Roundoff errors are smaller than the -40 dB mark?)


The first and most important thing to  know is that numerical differences and audible differences are not consistent with each other, but are related by a big broad deep technical area called Psychoacoustics.

At one extreme numerical errors on the order of 10% or more can be imperceptible. At the other numerical errors on the order of 0.1% or more can be perceptible. For example, short bursts of large errors can be imperceptible. Errors at some frequencies are far more significant than others.

Even how the magnitude of errors is calculated is not generally agreed upon.  The percentage of error calculated by one tool can disagree with that calculated from the same data by another tool.