Skip to main content

Topic: Short re-encoding blind listening test (Read 60847 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.
  • senjuuni
  • [*]
Short re-encoding blind listening test
Reply #25
What about transcoding from mp3 cbr 320 to vbr q2?
Would the results still be as bad?

  • Silversight
  • [*][*][*][*]
Short re-encoding blind listening test
Reply #26
What about transcoding from mp3 cbr 320 to vbr q2?
Would the results still be as bad?


Holy thread resurrection Batman!

In case you failed to understand it the first time I wrote it: Perform an ABX test, hear for yourself and stop asking impossible questions!
Nothing is impossible if you don't need to do it yourself.

  • senjuuni
  • [*]
Short re-encoding blind listening test
Reply #27
Oh my.. sorry
I forgot to enable email notifications to that thread and neither have I bookmarked it.
I need to dig into this abx... I barely know what it's about.

  • pepoluan
  • [*][*][*][*][*]
Short re-encoding blind listening test
Reply #28
I need to dig into this abx... I barely know what it's about.
I think it is explained in HA's Wiki. So you can start there
Nobody is Perfect.
I am Nobody.

http://pandu.poluan.info

  • senjuuni
  • [*]
Short re-encoding blind listening test
Reply #29
I did that. I tried winabx. Thank you.

  • shadowking
  • [*][*][*][*][*]
Short re-encoding blind listening test
Reply #30
Interesting test. I guess with so many rumours and so little testing, its easy toget sucked in to superiority of subband encoder etc .

I tried a simple test on half a dozen songs using mp3, aac, vorbis. I was really interested in the poor mp3 performance and also AAC performance.

128k NERO AAC> 128k AAC: Not great to my ears but I expected worse. Ringing , unstable sound, distorted hihats etc

192 AAC> 128k AAC: Better some distortions on hihats, guitars. I think a non fussy listener might be pleased.

256 AAC> 128k AAC: Very good in general. Most people will be pleased. Only slight artifact in some places.

256 AAC > 128k [other formats]: Very good in general.

128 MP3 > 128k MP3: Terrible. loud knocking, ringing, unstable and other sometimes scary sounds not present in the original.

192k V2-V0> 128k mp3: Better , still artifacts are in a lot of places, ringing, distorted hihats. Additional sounds (artifacts) are still persisting to a degree. Maybe a non-fussy listener would be OK with this ?.. Hmmm

260k ABR > 128 MP3: There is a noticeable quality bump. Additional sounds seem to disappear or become really quite. I must point out hihats are still distorted to a degree in places and there are some artifacts if you look for them, but things have become transparent to a degree. A lot of people might be pleased with the quality.

260k ABR > 128 MP3 [Helix/Xing - FHG] - Very good quality. Quality rises further when transcoding to a different MP3 PSY !!

260k ABR > 128 [AAC / Vorbis] - Very good quality on a casual listen.

Another interesting point: [Vorbis > Vorbis]

Vorbis 128k > vorbis 128k : Horrible. Similar mp3 128> mp3 128
Vorbis 192k: Still bad
Vorbis 256k: Again nice quality bump, still some noise artifacts.
Vorbis 320~384: Hmm.. Still not perfect ..Seems going higher than 256k doesn't yield much better quality ?

I have learned some things:

a) Transcoding from 128k is to my ears an abomination. At 256k there is enough juice to do this with reasonable results.

b) There seems to be some reaction when transcoding to the same PSY model - LAME nspsytune > nspsytune, Vorbis > Vorbis, probably even MPC > MPC.. So I think this MPC subband encoder ability is probably a myth.

c) LAME MP3 has the worse reputation for transcoding and this is also due to testing popular bitrates settings and same PSY model reaction as I described in point b);

At 256 K ABR and higher there is a rise in quality also for DSP like fake surround. It seems that at this point MP3 is a good transcoding source to other formats and to different MP3 encoders. Nero AAC is very nice even when transcoding to itself.
  • Last Edit: 24 July, 2007, 03:58:21 AM by shadowking
wavpack -b4x4s1c

  • halb27
  • [*][*][*][*][*]
Short re-encoding blind listening test
Reply #31
... 260k ABR > 128 MP3 [Helix/Xing - FHG] - Very good quality. Quality rises further when transcoding to a different MP3 PSY !! ...

Hi shadowking, very interesting.
Which version and setting did you use for 160k ABR, Helix and FhG?
lame3995o -Q1

  • shadowking
  • [*][*][*][*][*]
Short re-encoding blind listening test
Reply #32

... 260k ABR > 128 MP3 [Helix/Xing - FHG] - Very good quality. Quality rises further when transcoding to a different MP3 PSY !! ...

Hi shadowking, very interesting.
Which version and setting did you use for 160k ABR, Helix and FhG?


LAME 3.98b4 --abr 260, Helix 5.1 - 128k cbr / 160k cbr, FHG v1.3 surround encoder 128k cbr.
wavpack -b4x4s1c

  • halb27
  • [*][*][*][*][*]
Short re-encoding blind listening test
Reply #33
Thanks.
lame3995o -Q1

  • shadowking
  • [*][*][*][*][*]
Short re-encoding blind listening test
Reply #34
Does poor transcoding plague a hybrid encoder at 'low' bitrate too ?
I tried a short test:

Wavpack lossy 256k > wavpack lossy 256k

I encoded an entire song and then a short guitar intro. Quality is excellent. This is the equivalent for 128k transcoding for transform coders, but hybrid encoder shows none of the violent aggressive artifacts produced with the other coders. Only some extra noise was there on guitar pluck, not annoying, otherwise hard to pickup. Hybrids are either not suffering from this bad interaction and if they are it sounds much different.
  • Last Edit: 24 July, 2007, 11:20:04 AM by shadowking
wavpack -b4x4s1c

  • pdq
  • [*][*][*][*][*]
Short re-encoding blind listening test
Reply #35
This may simply be my lack of understanding, but doesn't wavpack lossy (or other hybrid) work by fudging the low order bits to make the data easier to compress? And as such shouldn't converting wavpack lossy to wavpack non-lossy give the same result as wavpack lossy to wavpack lossy?

  • singaiya
  • [*][*][*][*]
Short re-encoding blind listening test
Reply #36
256 AAC> 128k AAC: Very good in general. Most people will be pleased. Only slight artifact in some places.


I'm wondering if the slight artifact in some places would also appear in these samples when you go CD/lossless > 128 k AAC. In other words, are the artificats due to the transcoding?

  • buktore
  • [*][*][*][*][*]
Short re-encoding blind listening test
Reply #37
I just tried ABX aotuv b5 at q.8 > 128 cbr lame 3.97 VS lossless > 128 cbr lame 3.97

the result is pretty good. but i can still ABX them. slight artifact at hi-hat.

I want to try Nero at q 0.66 for comparison too. but i'm too tired and don't have much time.

So. shadowking did you already tried aotuv q.8? if so, what the result compare to nero at 256? if not, could you try them out? thanks.
  • Last Edit: 29 July, 2007, 10:48:38 AM by buktore

  • shadowking
  • [*][*][*][*][*]
Short re-encoding blind listening test
Reply #38

256 AAC> 128k AAC: Very good in general. Most people will be pleased. Only slight artifact in some places.


I'm wondering if the slight artifact in some places would also appear in these samples when you go CD/lossless > 128 k AAC. In other words, are the artificats due to the transcoding?


Most of the time the original has some distorsion and transcoding might amplify it to some extent. Seems it can happen at random places even at very high bitrate. I am certain that 128k-192 is inferior to 260k as a transcoding source, yet I am less sure if quality scales past 260-320k.

I just tried ABX aotuv b5 at q.8 > 128 cbr lame 3.97 VS lossless > 128 cbr lame 3.97

the result is pretty good. but i can still ABX them. slight artifact at hi-hat.

I want to try Nero at q 0.66 for comparison too. but i'm too tired and don't have much time.

So. shadowking did you already tried aotuv q.8? if so, what the result compare to nero at 256? if not, could you try them out? thanks.


Thanks for testing. I haven't tried vorbis much but I expect similar findings to yours - reasonably high quality but not always perfect.
wavpack -b4x4s1c