I'm a speciman of the "I-believe-standard-is-transparent(enough)" people, so I am interested in decoding and reencoding "insane" MPCs to "standard" MPCs, in order to save HD space. Has anyone run some ABX tests on this yet, i.e. is:

Original -> MPC 7 -> WAV -> MPC 5

abxable from

Original -> MPC 5?

Is this reasonable, at all?
Please comment!

this seems only reasonable to me if you are *really* short on disk space. otherwise leave the files alone.

regards; ilikedirt

That may be true for one or two albums but considering that standard MPCs are only 75% as big as insane ones, you would save 250 MB per 1 GB of "insane" albums. That is quite a lot in my understanding and definitly worth the effort.

BTW: Where can I get this ABX software everyone is talking about?

What the...? I beg to differ. Not only is it not worth the effort, it's a perversion of MPC's goal to deliver transparent quality at --standard, since that's exactly what these files most likely won't do. If you think it's okay to re-encode MPC -> MPC for saving some megabytes of disk space, you may have chosen the wrong format (no offense).



So your proposition is based on space saving, not quality considerations...

ABX software can be found at http://ff123.net/abchr/abchr.html or http://www.pcabx.com/ . Also visit http://ff123.net/training/training.html.

edit: Sorry, didn't see that you're the same person that made the initial post

iirc others have made some rudimentary tests and it turned out that reencoding MPCs to a lower setting reduces the quality more than one would expect (since mpc transcodes well to other formats), so that should not be an option....

Ah: try here.

More extensive tests would need to be done, but if the results are as bad as SometimesWarrior says, then it shouldn't be too hard to verify that reencoding MPCs is a no-no. I would test myself, but I don't have much time on my hands atm.

Another ABX comparator here: http://www.kikeg.arrakis.es/winabx

Thx for all your info.

@citay: I think you already noticed that quality is indeed an issue for me, that's why I asked whether the quality will be reduced or not. My consideration was: If q 5 is transparent and reencoding q 7 -> q 5 results in "no"!
quality loss, I could reencode those albums and save 25% of HD space (which is alot if you have enough q 7+ albums). As I know now that there will be indeed a quality reduction I will of course not do it.

Another question that came to my mind (should perhaps be asked in a new thread though). If it is true that there are a few very difficult to encode samples, which should better be encoded using q 6 (or even more), why doesn't mppenc detect such samples and just allocate more bandwith? I means its a VBR encoder, isn't it? Shouldn't it be able to compensate for such rare samples or is it impossible to detect them? Another thing: Is it really only 1% of the music that is ABXable from the original at q 5?

Yes, there exist quite a few samples on which mpc -q5 can be ABXable.

This problem exists in all lossy audio coders: it's very difficult to determine exactly, at each precise moment, how much distorsion can be admitted at which frequencies. Thus, from time to time, a few tenths of milliseconds can get encoded using too low precision, which results in an ABXable (at least for some people here) difference.

Musepack is VBR by design, but that doesn't make it perfect !

Where musepack fails to provide 100% transparency, pretty much any other audio coder will also fail (often much more miserably).

I don't know if it's ABXable, but anyway rename the files like 'Artist' - 'Album' -- transcoded! should you share this somewhere on the net

[edit]It's a petty not to keep the original mpc files. I'd suggest transcoding to ogg in this case...[edit]

Vorbis might not fail at the same places as MPC (different design),
but surely will fail on larger number of samples.

@lucpes: As mentioned above I will not use reencoding from q7 to q5 because there is a severe quality loss. (At least SometimesWarrior claimed that.)

Can someone tell(or better send) me a sample where he/she was actually able to ABX original -> q5 or q6 from original -> q7 with a guessing probability < 5%?

The reason this hasn't been tested before is simple: No codec was and will be tuned for re-encoding/transcoding quality (regardless of the technical possibility). First of all, there are enough problem cases where direct encoding from WAV isn't transparent (not many with MPC these days, but still). As you can see right on this board, new problem samples keep getting discovered by the users. There is always room for improvement, and if it just means a bitrate reduction while the quality is maintained.

Then, there's the 'ideology': MPC is aimed at users who want the best possible quality from a lossy codec. They relinquish MP3 compatibility and put their focus on transparency. Now, this should already tell you that few of these users would ever think of re-encoding a lossy file to MPC again. Yes, some transcode to low-bitrate MP3 for their portable player, but they don't really aim for hi-fi audio when they are on the bus etc. So the re-encoding to MPC has (rightfully!) never gotten much attention from the developers or users. I also don't think it's too hard to ABX a sample that has been re-encoded from q7 to q5; as soon as you find a file that accentuates one type of artifact introduced during re-encoding, you can find a multitude of similar samples. But good luck on finding somebody that performs some listening on this; i pass. I think it's the wrong direction.

@Citay: Nice explanation but you should have read my posts, especially the last one:



Also, I didn't ask for samples where "original -> mpc-std" is abxable from "originial -> mpc-insane -> mpc-std" as you assumed but! for samples where "original -> mpc-std" is abxable from "original mpc-insane or mpc-extreme", which is something completely different.

Having Realized now that this question was apparently off-topic I will repost it again in an new thread.

Oh, you are right, i misunderstood that part.

Matter of fact, i ABXed a few of those samples where q5 showed an artifact and q7 didn't (to such an extent), for instance 2nd_vent_proper.wav or BSB.wav (Frank Klemm confirmed it via e-mail for q5).