A while ago I asked how bad transcoding really is. I never really got a satisfying answer. It seems there is a consensus in HA that transcoding is bad, which may be justified, but in many cases these claims seem mighty close to a violation of TOS #8 since there haven't been many ABX tests for transcoding...

Anyway, my situation is that I have about 80Gb of lame 3.90.3 -aps mp3 files, and have a brand new iRiver iHP-140. I am considering transcoding a fair amount of this to lower bit rate ogg (how low should I go?) to fit more music on the player. I honestly have trouble ABXing even ogg -q0 from lame -aps (but I haven't spent that much time trying)., so I don't really see the problem in doing this. It isn't for archival purposes, its just so I can carry around as much music as possible.

Is there anything wrong with this? My point is that there may be legitimate reasons to transcode, and the quality loss may be practically imperceptible. If this is true, why do so many people say that you should never transcode (unless you deliberately want bad files? )

Dibrom recently wrote:



... but how much lower??

.dd.

The only real answer here is to just try for yourself and decide what you're comfortable with. If you can't ABX Vorbis -q0 from LAME aps, and nobody else is going to be listening to your files except yourself, then maybe it makes sense to transcode to that level.

Do you expect that you're going to be more likely to be able to discern artifacts in the future? If not, again, probably doesn't make much sense then to use the extra space for quality.

As a general rule I think HA should still discourage most people from transcoding because it does seriously impact quality in many cases, and although there may not have been any centralized tests measuring this, it has popped up enough in individual cases to support this claim.

So if you're sure it's not going to make a difference for you, then there's not much reason to bother. But for other people, as a general rule, it should still be discouraged until they realize the possible downsides and are able to make an informed decision from there.

I haven't noticed really bad loss from transcoding audio from video sources, but that's about my only experiences with transcoding. I am aware the quality loss is there from further quantisation but it's just up to you.

Try transcoding a few files and see if the quality is acceptable to you. If it is then it would probably be OK. If it's not you might have to end up doing some reripping.

Thanks Dibrom, that's the answer I expected.
Thanks too DreamTactix291. I guess the real test I should do is compare ogg -q0 files created from the original cd with ogg -q0 files transcoded from lame -aps. If there is no difference, and if I am satisfied with ogg -q0, then its pretty easy to set up a mass transcode with foobar

I guess my main point was that there is very little information in HA about transcoding, other than the sort of blanket warnings like Dibrom made here. Although I agree with Dibrom's reasoning in this case, shouldn't people be a just bit more careful when they diss transcoding?

I'm not sure.

People will transcode regardless of what most of us here on HA say. Since most of the people who come here are usually interested in quality, it makes sense in my opinion for transcoding to be frowned upon in most cases.

Most people don't even realize that transcoding is possibly a bad thing, so even though the typical response here might seem like an overreaction to someone who has a legitimate use for it, for less experienced users, it might be something they should learn.

One really has to understand I think that for every general rule, there's usually going to be an exception, and transcoding is no different. It's not really a great idea if it can be avoided, but in some cases it might make sense.

Dibrom, you know more about LAME than me. How much quantisation is done in general in a lossy audio codec and how badly would it be compounded with each encode? Just curious.

I don't really know how to calculate that in the way that I think you mean.

But the best way to tell "how badly it would be compounded", at least in a measurement that would be meaningful, is to listen to how much worse it sounded from a first generation encode in a listening test

I guess that was answer I was expecting. I haven't ABXed anything in a while, so I guess a transcoding test wouldn't be a bad thing to try.

Fun, fun.

According to my recent test, MP3 is not very good for transcoding so I don't see the need to do more tests on it. MPC is better but can also produce artifacts. The near-lossless hybrids like wavpack transcoded transparently from 320k (I tested 400k, 320k transcoded to 128k mp3). Dualstream seems good as well but I need to test more. Using the perceptual coders, the loss will depend on the sample and will introduce artifacts. The hybrids can only produce hissing noise - no traditional artifacts.

For casual portable listening its ok for *me* to transcode from say mpc to mp3. I feel too uneasy say its 'ok' for others though.

there's probably not much ABX results out there because they don't tell the full story if all you ever get is 16/16

i've tried some lame --aps to lame --aps transcodes, and there hasn't been one test i couldn't abx in several seconds (and i'm not as trained to artefacts as a lot of people here).

i'm not completely sure the exat figures, but a transcoding will increase quantization noise by up to twice as much, in theory. this means that a codec tuned for transparency will break on the first transcode.

of course, this doesn't mean the output will sound [i]bad[/u] as such. but it will certainly be distinguishable from the original.

if you have a codec tuned for transparency, and choose extreme settings (like musepack at --quality 7), you may not be able to ABX the transcode, even from the original sample. however, most people do their encoding at "transparent" settings (--preset standard, or q5) or less, in which case transcodes will sound worse no matter what you do.

if you apply a lossy compression that isn't based on psychoacoustics, you might stand a better chance of being able to transcode. this is why MPEG video transcodes reasonably well - no doubt if it exploited the human visual system more than it does, it would fail dismally on transcoding.

In my test 384k mp3 was no better than APS

http://www.hydrogenaudio.org/forums/index....2&hl=results+of

MPC to MP3 is decent even from q5. Did a few quick abx tests with Dualstream tonight at Q5, 3 and 0. Q5 and 3 are transparent like wavpack.

This is some interesting info written by Ghido (optimfrog dev) concerning dualstream:

- suitable for further editing and transcoding (the TC make use of
psycho acoustics, and produce distortions which generally make
editing and further reencoding (even with the same coder) give
inferior results

- suitable for computer processing of audio data (from the point of
the computer voice recognition, the TC make the audio data near
unusable. For example, when encoding vocal formants, the TMN level
in the respective critical band will add important noise,
diminishing the clear tonal characteristics of the formants)

I recently bought a MuVo 128meg and played a bit transcoding APS files down to abr 128.

I don.t want to violate TOS 8 or anything but the results were obviously artifacted and not acceptable, and I am not overly picky for my portable.

On the other hand APS files transcoded with AOUTV b2 ogg to 100kbs were nicely listenable but won't play in the MuVo of course.

So I think the transcoder as well as the quality of the original file has a bearing. MP3 to MP3 is not so good for sure.

another point which people who listen to gapless CDs may want to consider:

transcoding an album with seamless trackchanges is a very dangerous thing - at least thats my experience.

Even if i take a 320kbit mp3 (directly ripped from CD with EAC) and then trascode it to --aps, some tracks which were seamless in the 320kbit version, have audible clicks in the --aps version. Of course this depends on the music you encode - trackchanges which have a steady tone, especially if its a high-pitched one, seem to be most affected by this.

- Lyx

Just a few things I'd like to mention about transcoding:

1) Sometimes it can't be avoided (ie, for portable use)

2) One of the biggest reasons for just yelling out that transcoding is bad is because of the false belief that many people have (especially on file sharing networks) that you can take a low bitrate file and transcode it to a higher bitrate and it will gain quality. This obviously is not true.

3) It's been mentioned before, but the general theory on transcoding is that if it it must be done, it is better to use a different codec for the transcode due to different psychoacoustics and methodology used.

There's one question, that I always had:

Isn't it possible to transcode lossless from one lossy format to another by using the same quantization scheme (if allowed by the codec) and turning psychoacustics off (since it's already reduced)?

Sure, the file would not have least possible size, but it's losslessly transcoded ...

Did you ever think about that theory?

The problem is that lossy codecs use different methods (e.g. transforms and windowing) to take the time domain data into the frequency domain. So even before you apply psychoacoutics and quantization, you're starting off differently. Furthermore, even if the transform sections were the same, the techniques used to deconstruct and reconstruct the frequency domain data in the psychoacoustics section of most codecs are different.

Problem is that it is not "reduced" by the first encoding, but instead noise has been added.

This also applies in biology.

I havent ABXed, but it seems to be that you can do it better applying mp3gain to files before transcoding (or check ReplayGain in foobar diskwriter + check UseDSP). MP3's could be a bit overdriven(effect of encoding), if you transcode
MP3---->Wave---->OggVorbis
you'll can get "flat tops". This can affect both quality and compressability

MP3--[RG]-->Wave---->OggVorbis
(RG in this case is playing role VolumeDown)
this effect will be avoided

You could use VolumeControl as well, but i think Transcoding+RG perfectly suits portable use.

Correct me if i am wrong

P.S. sorry for bad english

I applied mp3 gain to the files I transcoded. "Max no-clip gain" was the option. Still sounded like ass, but I suppose clipping would make it worse.

You people are all nuts. Transcoding is embraced by The Sony Corporation.

The latest Sony Digital Walkman automatically transcodes any MP3s you feed it to the superior ATRAC3 format. Just think, no more lousy MP3 sound!

Wow, it's that easy to get CD quality sound?!?!

In my opinion transcoding should be done only when there's absolutely no way to encode from original data. Sure it may take some time to rip a huge collection again, but I'd do it in every case just for the peace of mind (and nobody is going ABX a huge collection to be sure that there is no audible loss in any songs).

thats why many people have gone the lossless/lossy hybrid-route:
rip to lossless for archiving - then later rip to lossy on demand (while keeping the lossless files) - that way, the don't have to rerip their entire collection..... and with fb2k mass-transcoding really is a one-click job - you can just start in the evening, and next morning, it will have transcoded some hundred songs from lossless to lossy - compare that to diskjokeying all of those CDs while re-ripping.

It seems to depend on the interaction between the source and destination codecs. I did a few tests this morning and it seems that WMA and Vorbis don't get on well at all:

WMA 96kBps -> Vorbis 128kBps
ABX Vorbis from CD 10/10 (very easy)
ABX Vorbis from WMA 10/10 (not as easy, but still not difficult)
The transcode adds a strangely variable quality. At some points it sounds fine and at others it sounds as though Seether are playing in a metal pipe. Cymbal clashes and the singer's voice are the most obviously effected.

MP3 96kBps -> Vorbis 128kBps
ABX Vorbis from CD 10/10 (fairly easy, about what I would expect from 96k mp3)
ABX Vorbis from MP3 8/10 (not conclusive - another try without the morning traffic noise would be better)
This one doesn't have the obvious artifacts of the WMA->Vorbis conversion. The final ABX is pretty hard as it became a matter of playing "hunt the artifact" rather than an obvious change in sound.

cabbagerat, what settings did you use for Vorbis? If you used bitrate management at 128kbps, Vorbis is NOT a CBR codec and bitrate management produces suboptimal quality. Try a quality setting instead.

Here comes some ABX results. Note that I don't have golden ears. Hearing damage from being a musician has made my ears probably representative of the average casual listener when I listen hard. I'm careful to wear hearing protection always, but it just creeps up on you. =(

I'm trying to duplicate the conditions of the original poster. Encoded a FLAC rip using LAME 3.90.3 --alt-preset standard and Vorbis 1.1 RC1 -q0, then transcoded the MP3 to OGG using the same settings.

The transcode is slightly smaller in file size (1 kbps less average bitrate). ABX results follow. Note that the ABX was *completely* blind, total score is only revealed at the end of the test.

Original vs. Normal Ogg
5 out of 5, p = .031
Piece of cake. HF boost gives it completely away

Original vs. Transcoded Ogg
5 out of 5, p =.031
Ditto. I thought this was easier to ABX than the first, but based on the next test I guess it was my imagination. I have seen the placebo effect first hand!

Normal vs. Transcoded
9 out of 18, p = 0.593
Sometimes, I swear that X is neither A or B. Well, my ears can't tell the difference. I say go for it!

For portable/casual listening, I go all the way down to Vorbis Q -1. About 40 kbps average. I love Vorbis because the quality degrades so nicely, there are no real distracting artifacts even at absolutely tiny bitrates, it just sounds generically lower-fi.

Anyway, conclusion: Transcoding from --alt-preset standard should be okay, especially for portable use.

I tried mp3 APS to vorbis .85 with aotuv b2 with about 70kbps. This was listenable but easily a/b'd but I didn't use an abx tool.

Q 2.5 gave me around 100kbps and this wasvery nice for portable use for me.

I used -q3 (the default), which on this sample made an average bitrate of ~128kBps.

There is an interesting difference in our results. I take a low quality original (96kBps) and transcoded it at higher quality (128kBps) - something that many traders do to "gain extra quality". Analogy took a high quality original (LAME -ps) and transcoded to a lower quality (vorbis -q0). I got easily ABXable results and analogy got unABXable results. This suggests (but does not prove) an interesting result: that low->high transcoding does more damage to quality than high->low transcoding.

Having said that, I don't think I could ABX my samples through a pair of earbuds with traffic noise in the background - the typical portable listening environment. It seems to me that high->low transcoding for portable use is perfectly acceptable. Low->high transcoding in an attempt to gain quality is (as most knew already) stupid.

Then again, we have different ears, and were using different samples. What's un-ABXable for me might be easily ABXable for you. I might not have been using a particularly difficult sample, too. And remember, I'm a hearing-damaged musician (hearing stops at about 18 khz).