I'm using aucdtect to confirm if a file is lossless or not. What I'm finding are some flac file are said to be mpeg. I believe this is wrong as I'm ripping the tracks with EAC from premium CDs unlikely to be mastered from compressed sources.
if aucdtect know to give less than perfect results, and does anything else exist that might prove this one way or another?
I think mp3 compression started in 1993, so CDs before 1993 could not contain any. But is mp3 the only audio compression?
+ Reply to Thread
Results 1 to 30 of 35
AudioCDs, aka CD-DA, aka RedBook, is a very specific format. It doesn't support ANYTHING else. Only 16bit, 44.1kHz, 2ch, LPCM. No compression.
You may have Data CDs, or MIXED MODE/Enhanced CDs, which would have a data portion in addition to the Redbook. Files in those sections can be whatever you want to put on them, but they only be compatible what whatever you can verify is supposed to be compatible. They are not UNIVERSAL like RedBook is.
If you used EAC to extract/rip, it is designed to ONLY work from RedBook (or the RedBook section of mixed mode) CDs.
In the app, YOU choose how it is to be saved. It can be a similar LPCM like a WAV file, or can be lossless like FLAC, or can be lossy like mp3 or aac, etc.
If it is LPCM, it should be (barring issues with error correction, and barring hidden additional data*) an exact copy of the LPCM that was on the disc. If you chose FLAC, a round-trip compression & decompression should also be an exact copy of the original. Hence, "lossless".
Mp3 etc are lossy - and that is how and why they achieve the level of compression they are capable of.
Does aucdtect do a good job? Don't know. Haven't ever used it. I find that using EAC and having done full pregap scan and pre-calibrating with reference tracks should get you the best rips possible from any app or platform, thereby rendering something like aucdtect superfluous.
Seems to me that the utility of this app is in determining the origin of files of unknown provenance. But why would they be of unknown provenance if you are the one doing the ripping? And if you are NOT doing the ripping, how are you legitimately supposed to be getting these files?
Also, it isn't clear whether this app does its work via Hash signature comparison with known manufactured CD contents via a database, a la freedb, or whether if uses an analysis of the entropy of a file, where a compressed version would be different from an uncompressed one.
Either method isn't infallible.
Last edited by Cornucopia; 17th Jan 2021 at 21:06.
The actual rip (with EAC) is said to be as good as you can get from CDDA. The problem is the disk may have been mastered from a lossy source (a lot of cheap compilation disks are). aucdTect or tau analyser are tools to detect this. I've struck a few where tracks that should not be detected as MPEG are. Hash signatures in a database would not influence the process. It seems neither tool is perfect. I just wondered if anything else exists to compare results with.
Are you talking about oldies compilations of music where the copyright in the music has expired
and any old publisher can issue the tracks? There are many inferior cuts that make it onto this kind of CD,
but you don't need a program to tell you that, you can hear a difference.
I don't understand the scope of what you're doing. Are you trying to detect counterfeit products?
This has nothing to do with counterfeit or copyright. Or older CDs, the issue can occur with any disk where its source material has been compressed.
Detecting if it has or not is my aim. You cannot really "hear" the difference. Maybe on some, but the majority no. It could be an older recording fooling auCDtect because it just doesn't have the bandwidth.. I don't know. Although you'd think there more to these analysis that purely freq range.
Hence asking here about this in case someone has some ideas. Or know of another program like those already mentioned.
Loss of high frequencies is not enough, many oldies are like this, even the official studio master versions.
Do you have any example tracks you can share with the forum?
This is what auCDtect and Tau Analyser do.. by somehow anyalysing the psychoacoustic mp3 model. I think only MPEG and CDDA are valid as it reports only one or the other. The input file can be flac or wav (or for Tau read direct from the CD).
Yes I have examples, but is that practical? Maybe Google for those apps (they're free) and test with some CD you have. If you want a file I can upload one but they're quite big (7 meg)
The problem I see is that there is a general ideal psychoacoustic model (general in the sense that it is derived from and averages the response from multiple users), but the models implemented in mp1, mp2, mp3, aac, ac3, dts, opus, etc are all different and non-ideal.
Seems pointless if one cannot account for multiple forms these (and other) kinds of compression.
Not sure where you are getting these discs from, but a reputable professional producer wouldn't be caught dead putting out something from a compressed master. With one exception: if that was a restoration when an original was permanently lost (fire, flood, etc).
I still have my original recordings from nearly 40 years ago. (Nothing to play them on now. - Anybody know of a studio that has a 1/2" or 1" Otari 8/16 track 15ips reel unit & can bake a tape?)
Yes. 7MB for a 3 min mono flac. I will upload just a sample 45 secs.
> reputable professional producer wouldn't be caught dead putting out something from a compressed master.
In a perfect world, and while most Cds are fine you will get some cowboys. How do I know ? I have supplied some companies with compressed files, and while I don't like it, its only done as you suggest when no other source is available (or available cheaply). Every man and his dog were putting out CDs prior to the 75-year copyright being reintroduced. (From 50 years).
To detect if they're lossless or not becomes practical. In some cases the recordings are very old (1950s/60s).
I know little about audio formats. But I do know a little more about maths.
So the question is. How many minutes of audio is on the CD from which the above sample was extracted ?
You mean the whole CD ? 77:18
The track timing is 2:55.
Sample was 46 seconds.
Where are we going with this ?
Ok. So now you do the maths. Calculate the total size of 77 mins based on 3 mins at 7 mb.
Anywhere close to 700/800 mb ?
The CD is audio, not a data disk. The flac size has no relationship to CDDA. Unless you convert to a red book spec but then is no longer a 7MB flac!
But this doesn't help detect if the file is truly lossless or not. Or to what extent auCDtect is bulletproof.
I did get another copy of the sample and this did show as lossless (unlike the sample). I don't know it's true source though. The one that failed was from Bear Family which is a shame as they are normally very good.
But this still give no absolute conclusion.
did the A/D conversion?
The data from the CD master can certainly be converted to mp3 and then used on a new CDDA but what is the point of doing that?
you have a connection to the music business - can you explain the scenario that creates these files you're looking for?
In all that you've said so far, I'm still not able to infer the "big picture" of what you're talking about.
In the case of the Bear family, don't they do their own A/D conversion from the best available analog tapes?
SO in that sense they're lossless a far as I can see
Sorry I just assumed ppl here would know far more than me. AuCDTect is a tool to check if an audio file is lossless. Tau also does this (but from the CD itself).
You would expect it to be lossless, but it won't be if the file (or CD) source is lossy e.g mp3.
Someone may have extracted a file from CD (to wav) and then to mp3. Someone who doesn't know better might then convert that back into a flac (or wav).
You may not know this by just listening, hence using a tool to analyse the file.
So the "big picture" question is how good are these tools, how often is there a false positive. Or can they be considered (near to) 100%.
I found a CDR I burned many years ago for my mother, a bunch of songs from the 50's and 60's.
Some if the songs were tracks ripped from CD's I had and few were mp3's
I found Tau analyzer not to be accurate, one of the known mp3 songs was identified as CDDA
Interesting program. I hadn't heard of AuCDTect so I gave it a spin with two CDs worth of tracks I ripped to flac myself. If anyone cares, it was "Stop The Clocks", an Oasis best of.
Test 1 - They were all determined to be lossless.
Test 2 - I converted to flac again while reducing the volume to ReplayGain volume. Most were reduced by over 10dB. The new files were still determined to be lossless.
Test 3 - After a conversion to 320kbps MP3 and back to flac, they were all identified as mpeg, except for one, where the conclusion was "????".
While I was at it, I thought I'd see if changing the audio in some way could fool AuCDTect.
Test 4 - Converting the lossless files to flac again, while running them through an 18k low pass filter, 15 were determined to be lossless and 3 to be mpeg.
Test 5 - Another conversion to flac, this time with a 16k low pass filter. 7 were determines lossless, and 11 mpeg. It seems like the lower the cut-off frequency, the more it's weighted into the decision.
Test 6 - Converting the MP3 versions to flac again, this time while adding enough reverb to be annoying. Does anyone still manufacture hi-fi amps with built in reverb? I know a couple of people who actually listen to CDs with reverb enabled... on purpose.
Adding reverb didn't seem to fool AuCDTect. 15 determined to be mpeg, 3 undecided.
Test 7 - I thought I should check the program doesn't decide mpeg because of added reverb, so lossless to flac with the same reverb applied. They were still all seen as lossless.
I also tried some home-made CDs, or rather the flac files I created instead of a CD, by extracting the audio from a Bluray and a DVD. Both containing live concert video. Well... I guess if it's a concert it must be live....
The Bluray contained PCM audio. I split it into individual tracks with fade-in and fade-outs etc. Same for the DVD audio, only the original was AC3, and I think it was 5.1ch I down-mixed to stereo. For both, the sample rate of the flac files is 48k, not 44.1k.
Test 8 - All the tracks from the Bluray source were identified as lossless.
Test 9 - For the AC3 DVD audio it was 6 lossless and 11 mpeg.
Test 10 - Tracks created from DVD AC3 again, and from memory without having to down-mix, but I couldn't be bothered checking, and the DVDs might have more than one audio stream anyway, so I'd still have to remember which one I used. For the second lot of tracks ripped from DVD audio, it was 5 lossless and 12 mpeg. Maybe the algorithm isn't as finely tuned for AC3 compression, lowering the accuracy, but it's probably not designed for that anyway.
I suspect, but haven't tried to prove, the accuracy may be lower for low fidelity audio, as it'd be harder to determine if it was low pass filtered by a lossy compressor.
That's very interesting and thanks for doing the tests. To Dave I've not yet found a result from Tau that differs from auCDtect. So a known mp3 being ID'd as CDDA isn't good. Usually its the other way around.. CDDA as MPEG. I've been finding new sources to make a complete set of 'CDDA' flacs figuring if you have them they may as well present as lossless, where possible. Hello_hello results were more promising... I know zilch about AC3 (seems it's another name for Dolby, but Doldy isn't compressed at least with tape, afaik). Perhaps auCDtect can't handle that format, or the higher sampling. Good that all PCM stuff passed and that added reverb didn't seem to matter. Altering freq range would fool it you'd think. My results are usually a 1st gen rip on a good audio CD gets a pass on all tracks. Individual (data) files can vary a lot and not unexpected.
What percentage of accuracy would you think auCDTect deserves? Perhaps it all depends on the source and thus impossible to say.
Ok. My bad. I was not allowing for the reduced size of flac compared with CDDA. But size reduction is at the absolute max if what I read (50-70%) is true.
Now my reporting program may not be accurate either. But it does report your sample as flac ie lossless.
And we have to rely on the reporting program. If the original source would not play in a conventional CD player then it is not CDDA in the first place.
Dolby Digital, and on a PC it generally has an ac3 extension. It's everywhere. It's lossy like MP3.
AC3 also is available on tape-based recording formats. E.g. D-VHS decks that capture broadcast channel streams (MPEG2TS w MPEG2 video & AC3 audio). Some Digibeta decks can layback produced AC3 for distribution & archiving.
> But it does report your sample as flac ie lossless.
What reported it ? It should be lossess but isn't according to the tools mentioned so far.
When you say "i.e. lossless" you mean it knows it a flac file so must be lossless?
The program does not look at the file extension. It reads headers. Try it yourself with a simple renaming of the file extension (mp3 will suffice)
mediainfo knows it is a Flac file regardless
But when kirkmc says the flac is not lossless he's talking about what the file was derived from as I mentioned
an example just now
Actually Flac is a confusing thing in this whole discussion, should stick to talking about WAV and mp3
I get my material only from reputable major publishers or from known audiophile or engineer friends, so would never need to encounter or look for such tools.
But you never answered my earlier question: if this only works on SOME pychoacoustic models, how is it expected to be definitive to your judgement? Does it matter if it's 80% or 40% correct if it puts ANY of your judgements in question (positive or negative)?
Also, if you cannot tell the difference via hearing, WHAT DOES IT MATTER? Seriously, if the quality is still great, isn't that good enough?
If you CAN hear it, no need for a tool to tell you to return it.
Me, depending on the kind of material, I can hear the difference somewhere between 320 and 256kps (on modern mp3s). I have a method of A/B'ing that makes very clear where that threshold is (need to have an uncompressed original for reference though), and this can be applied to test any form of compression.