I've got a few albums where I got ahold of 128kbps mp3's (probably fgh) and also got ahold of high-bitrate Lame -V0 and -V2 encodes. Former from free promo services, latter from paid download services such as eMusic.

I'm not interested in mp3's with average bitrate of 250+ and generally encode cd's that I rip at Lame -V3 or -V4.

So what I'm wondering: which is likely to be better audio quality - the 128 kbps fgh or the -V3 transcodes from really-high-bitrate -V0 or -V2? The transcodes come out around 170 kbps themselves.

I don't think I have the hearing quality to ABX differences here, but am interested in having better-quality mp3's on a level noticeable by people other than just myself.

I also think this is also an interesting question generally, about how much quality is lost of likelihood of artifact production of two high-quality lossy encodes vs. one lower-quality lossy encode. It's something I was curious about before I had a specific instance where I needed to make up my mind.

Thanks.

Well, you are going to get different opinions from different people as we all have different ears and equipment. The best thing you can do is try to ABX the material yourself. If you can't ABX a difference between the 128kbps file and the Lame -V 0/-V 2 version then you might as well just go with the 128kbps files. Try setting up a blind ABX test between the 128kbps and -V 0/-V 2 files first. If you can ABX a difference (in that the -V 0/-V 2 songs are higher in quality) then your next test should be against the source -V 0/-V 2 songs and the transcoded -V 3 files.

As you're interested in preserving quality for others rather than for yourself, it's impossible to say for sure what level of quality will be good enough out of the proposed alternatives. ABX testing won't prove to you what others will or won't hear so I don't think it helps much in this particular situation.

Is there any particular reason why you have to transcode the existing VBR files to a lower quality level? Can't you just use either the 128Kbps CBR encodes or the -V0/-V2 VBR encodes as suits you best at the time?

Anything below VBR at -V2 seems to be bordering on the point of being ABX-able for some people on the forums so it seems a little unwise to transcode down to anything lower when there is bound to be some loss of quality involved not only in the lowering of bitrate but due to the transcoding process as well.

If it were a straight choice between keeping either the 128Kbps CBR files or the -V0/-V2 VBR files then I'd definitely keep the VBR files as they will be perceptually transparent to the massive majority and the 128Kbps CBR ones certainly won't be to everybody.

Cheers, Slipstreem.

the -V2 or -V0 files will of course be way better quality than the 128 cbr files.

There's no reason I *have* to transcode. But for me that quality is overkill (specifically the -V0) and I'd rather have smaller files. Hence either transcoding or keeping the 128kbps files instead.

Since I don't like really high bitrates, if I ever get 320kbps mp3's I will transcode them. Same with -V0 files.

Since transcoding is usually a general no-no (i.e., it's better to rip straight from a cd again instead of transcoding), the actual amount of loss from a second lossless compression is rarely discussed.
As I said above, I've been generally curious: is transcoding likely to introduce different sorts of artifacts? Maybe the -V3 transcode from a -V0 file will be overall better than the 128 cbr (which I expect will be the case) but will have some weird artifacts in a couple of places as a result of encoding twice.

I understand that ABX testing is an individual exercise.
But it's clearly possible to make some generalizations (e.g., -V0 or -V2 is *much* more likely to be transparent to a given person than 128kbps cbr).

So I'm curious if there's any knowledge and experience out there about transcoding and quality loss from multiple iterations of lossy compression.

There are only a handful of people I would trust to make these generalizations and I haven't seen them make any about 128 CBR vs high quality to semi-high quality VBR transcoding.

ABX!

Hi. I have tried to transcode to 128 Lame mp3 from 320 lame files and the results are very poor with audiblle swishing in the highs. i was noticeable and annoying.

High bitrate mp3 to ogg or aac is much better if your device will support it.

Better to abx and see which you like better.

Quite possibly does not apply to your case, but is something which can be meaningfully generalized: dont transcode albums which have seamless trackchanges. While the overall level of quality decrease by transcoding is debatable, it is well known that seamless trackchanges can easily break by transcoding, especially when music complexity during the trackchange is low, so that a "pop" could easily be noticed.

Laziness and lack-of-desire-to-find-source-CDs has made me a transcoding advocate, so my post may be slightly coloured in that direction. I have also been blessed with the lack of ability to ABX transcodes, at least for the first few generations. However, by hearing the facts and doing some experiments, I have been able to make a few generalizations:

1. Every time you transcode, another generation of artifacts will appear. Sometimes they will be audible, sometimes the will not, but they will become part of the fabric of the files. With each additional transcode the encoder will think that artifact is part of the music and try to preserve it to the best of it's ability, and with several generations, those artifacts will become more and more pronounced. The question is if/when will they become noticeable to you, and can only be determined through ABXing (or, if you wanted to know for the average person, then through listening tests). For this reason the fewest encoding processes should yield the fewest artifacts, and the most faithful to the original recording.

2. Because different encoders introduce different kinds of artifacts, often transcoding between different file formats is a better idea, but compatibility does not always allow this. If the first generation is MP3, you should theoretically get better quality by making the next generation Vorbis (for example.) This will still introduce more artifacts, but because they are from a different encoder, they will effect the music in a different way, and not make the original artifacts more pronounced.

3. Most modern encoders are transparent around 128, if it's the first encode. I am pretty sure this is the consensus based on a listening tests that are here... somewhere.

4. The whole reason for audio compression in the first place is to save space - if you keep the original then there is no point in worrying about transparency, just whether the transcode is good enough for the intended use. Also, you should aim for the smallest filesize possible that is good enough to you - the only way to know for this as well is based on ABXing (or using listening test, made up of many ABXs).

For those reasons I would make this conclusion -> If you want to have files that are approximately 128kbps, keep the FhG encodes. If you want to make them any smaller still, then use the higher bitrate files as your source, and transcode away - if possible transcoding to a different file format or using a different encoder. And don't throw away those source files!

Feel free to tell me if any of my advice is contradictory to what you have heard.

Have you ABX'ed that or is it just a general statement?

I feel it's safe for me to say that I can more easily distinguish CBR 128 from my library than Lame -V2.

Pick the codec and I'll share some samples.

I think that the general assumption that those distinctions are likely to be audible to some people may come from the wording of the Hydrogenaudio LAME WIKI page itself that clearly states...

If a person is convinced that they can ABX a 128Kbps CBR encoding, then isn't it a reasonable assumption to make that they'll be less likely to be able to ABX a VBR encoding at -V0, -V1, -V2, or even maybe -V3 based on the wording above?

You seemed to be saying two posts ago (and please excuse me if I read it wrongly) that there are only a handful of people you'd trust to make such a generalisation. Joe Public is likely to take the WIKI entry at face value and as fact. Are you saying that people shouldn't believe the WIKI entry? Is it wrong or outdated?

Cheers, Slipstreem.

I tried a lot of transcodes from V2 / V0 a while back to mp3 V5. Very easy to abx and artifacts were really bad at times. I would say an fhg 128k would be better. the only somewhat acceptable level was ABR / CBR 256.

You won't see any mention regarding VBR to VBR transcoding vs original to CBR 128 encoding in the wiki.

EDIT: In looking back it does appear that I went astray from what I quoted of timcupery, though it does get back to the original question. My apologies.

I have re-read the post I was referring to and your response was in direct response to a specific question making no mention of transcoding. I just wanted to clear the point up before anyone else misinterpreted it.

I'm not being deliberately pedantic. I'm just pointing out that the answer to the specific question quoted in that particular post was wrong according to the information given in the official Hydrogenaudio LAME WIKI.

Cheers, Slipstreem.

EDIT: You edited as I was posting. No need to apologise. I'm just a stickler for details and maybe a little too fond of LAME which makes me jump to its defence rather quickly at times.

EDIT (noting your response below): I answered the part regarding "some generalizations" and at the time chose to ignore what was given as the example. Speaking for no one but myself, I have no problem agreeing with his specific example, but extrapolating this to compare the artifacts between the 128 CBR encode and the transcode is not something I think can be generalized about.

I meant that the comment you were quoting here... http://www.hydrogenaudio.org/forums/index....st&p=557056 ...and replying to contained no reference to transcoding except by yourself. Never mind.

Cheers, Slipstreem.

EDIT: Agreed.