Skip to main content ↑ ↓

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: strange 48khz megamix encoding (Read 4998 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

strange 48khz megamix encoding

I'm a video producer. I want to make a shared sound library on our server so that our editing stations can access to it. All the sounds will be on a hard drive on the server.
-> Most of the sounds/musics will be ripped from audio CDs.
-> I choose to encode everything with ogg vorbis (for legal/ethical/quality reasons  ).
-> I want the best sound quality so I test the megamix encoder
-> The average bitrate should be around 256kbps so I encode @ -q7 (240kbps)

Because it's for a video purpose, the sounds must have a sample rate of 48000hz.
So I have two solutions:
- encode everything @44100hz (the editing software will have to resample AFTER decoding)
- ressample everything @48000hz BEFORE encoding (the editing software will only have to decode)

I guess the second solution is cleaner... 
I tried the both solutions, encoding from foobar2000, and the results are very strange :
Files encoded @48000 seems to sound slightly better (I have to do more ABX/listening tests to confirm this) and there are smaller than the same files encoded @44100 !!! 
Does vorbis prefer 48000hz sample rate ?

strange 48khz megamix encoding

Reply #1 –
Quote
Does vorbis prefer 48000hz sample rate ?

Not that I know of.  The main psychoacoustics values are the same for 44.1 kHz and 48 kHz though that doesn't really say much.  Have you tried using the standard Xiph.Org coder and see if it exhibits the same thing?

strange 48khz megamix encoding

Reply #2 –
I just tested with xiph.org/cvs encoder and the results are differents!

I do the test like this :

load a "track.wav" into foobar2000 (it's a rip from a CD made with EAC)
So I encode it @ -q6.9 (240kbps) with a vorbis external encoder (megamix for the test) two times :
first @ 44100 (without resampling) and the second @ 48000 (with dsp resampling).

So I get two files : track_44_mix.ogg and track_48_mix.ogg

and the track_48_mix.ogg file (13.4MB) is smaller than track_44_mix.ogg (13.5MB) 
average bitrate is 250 for track_48_mix.ogg and 251 for track_44_mix.ogg ...

the two files are not very easy to ABX, but I had a score of 6/10 (using foo_abx) may be it's the 48 vs 44.1 difference...

I've redone the test with the Xiph.org/CVS encoder and this time everything was normal :

track_44_cvs.ogg (12.2MB) is smaller than track_48_cvs.ogg (12.4MB).

what's the matter with Megamix ? 

strange 48khz megamix encoding

Reply #3 –
errhhh... an ABX score of 5/10 shows that you cannot tell the difference... that you are guessing.

And the difference in size is minimal.

strange 48khz megamix encoding

Reply #4 –
oops, this was a typo, my score is 6/10 .... corrected.

I can ear a very slight difference between the two files.  The difference seems to be in the HF noise. But may be the difference is only in my mind ...  6/10 is not enough. I should find time to do more ABX tests... but anyway, the difference is very small.

what is more strange is the files size ...

strange 48khz megamix encoding

Reply #5 –
This may be complete nonsense... but...

I think that a 48 kHz file may sometimes trigger less short-block switching as impulses are spread over more samples making the attack less "sudden". Those cases would be rare - but the bitrate difference reported is really small IMO...

If this is the case it might explain the different behavior of Megamix in comparision to Vorbis CVS as Megamix has a more sensitive preecho handling.

Just an idea...

strange 48khz megamix encoding

Reply #6 –
Quote
This may be complete nonsense... but...

I think that a 48 kHz file may sometimes trigger less short-block switching as impulses are spread over more samples making the attack less "sudden". Those cases would be rare - but the bitrate difference reported is really small IMO...

If this is the case it might explain the different behavior of Megamix in comparision to Vorbis CVS as Megamix has a more sensitive preecho handling.

Just an idea...

I've done more test and it doesn't append all the time ... only some files are smaller @ 48 than @ 44.1, not all of them    So I guess you're perfectly right 

mystery solved!

strange 48khz megamix encoding

Reply #7 –
Quote
mystery solved!

Not necessarily. The effects you observed could also be caused by the resampling process - which is not lossless and alters the waveform.

strange 48khz megamix encoding

Reply #8 –
hmm...in order to narrow this down further, could you do the filesize/bitrate testing using aoTuV beta 2, gt3b2, and aoTuV+gt3b2 and see if the 48 kHz files are smaller for them?

strange 48khz megamix encoding

Reply #9 –
Ok, I did the test with oaTuVb2 and Gt3b2. from the same track.wav file :

megamix_48 (13.4 MB) < megamix_44 (13.5 MB)
xiph-cvs_48 (12.4 MB) > xiph-cvs_44 (12.2 MB)
oatuvb2_48 (13.2 MB) < oatuvb2_44 (13.3 MB)
gt3b2_48 (14.7 MB) > gt3b2_44 (14.5 MB)

So the "problem" seems to be from aotuv ...
It's also interesting to notice that megamix (that means oaTuVb2+Gt3b2 here, since we are @ -q6.9) produce smaller files than Gt3b2 only.   
would the megamix minimize the GT3b2 bitrate inflation ? 
If it's confirmed, it's a good news! 

strange 48khz megamix encoding

Reply #10 –
Quote
Ok, I did the test with oaTuVb2 and Gt3b2. from the same track.wav file :

megamix_48 (13.4 MB) < megamix_44 (13.5 MB)
xiph-cvs_48 (12.4 MB) > xiph-cvs_44 (12.2 MB)
oatuvb2_48 (13.2 MB) < oatuvb2_44 (13.3 MB)
gt3b2_48 (14.7 MB) > gt3b2_44 (14.5 MB)

So the "problem" seems to be from aotuv ...
It's also interesting to notice that megamix (that means oaTuVb2+Gt3b2 here, since we are @ -q6.9) produce smaller files than Gt3b2 only.  
would the megamix minimize the GT3b2 bitrate inflation ?  
If it's confirmed, it's a good news! 

Thanks for the test.  That sort of confirms that the merging seems to be fine, and this observation is actually due to aoTuV itself.  I'm not sure whether it is "normal" behaviour for the 48 kHz version to be a smaller file than the 44 kHz one. 

strange 48khz megamix encoding

Reply #11 –
Quote
That sort of confirms that the merging seems to be fine

Really? So it may be time to make it the default encoder @ all quality levels

Why megamix produces smaller bitrates than Gt3b2 (@ -q7) ? 
 

strange 48khz megamix encoding

Reply #12 –
Quote
Why megamix produces smaller bitrates than Gt3b2 (@ -q7) ? 

I assume it's the aoTuV code that suppresses the bitrate.