Ogg needs work (my little test)

[Doobie]

I took a 192Kb/s CBR MP3 (LAME).
I encoded it to 100% VBR MP3 (MusicMatch).
I encoded both MP3 files to Ogg 500Kb/s VBR (dBPowerAmp)

Results (average bitrate):
MP3 192Kb/s is 192Kb/s, converted to Ogg is 379Kb/s.
MP3 100% is 116Kb/s, converted to Ogg is 367Kb/s.

It seems to be that these two ogg files should be MUCH smaller. Conclusion, Ogg is being very inefficient.

And, why isn't the rehuff functionality built into Vorbis v1.0?

PS: I did this test because I converted several dozen 192Kb/s MP3 files and converted them to Ogg VBR 160Kb/s and found that every one of them came to within about +/- 10Kb/s of 160. I know that some of those files should have been much smaller. Could the problem be dBPowerAmp treating the VBR setting as ABR, or is this really an Ogg problem?

[canadian_fbi]

um...
why should they be much smaller?

and what do you mean by "500 kb/s"... are you not using the -q settings?

and it goes without saying that encoding from mp3 to ogg is going to result in files that are less than perfect. especially when you do it TWICE.

[Doobie]

dBPowerAmp uses one slider for CBR/ABR/VBR Ogg settings. The maximum quality setting corrisponds to the 500Kb/s position on the slider and the values on the slider appears to be very close to the actual average bitrate.

I expect to save bits when going from CBR to VBR (consider the LAME CBR to the FhG VBR). Whatever frequencies are lost by the MP3 encoder are gone for good and don't need to be encoded by Ogg. Besides, CBR uses bits when they're not needed, VBR isn't suppose to use those bits. Yet, out of dozens of MP3s I converted to Ogg, none of them showed any bit savings that I would expect.

[canadian_fbi]

at 192 kbps, you probably aren't losing any frequencies that the ogg vorbis encoder would otherwise use.

you think that the "100%" vbr mp3 is really 100% quality? re-encode it at --alt-preset extreme, the highest lame vbr preset. you'll see that there's a lot left to encode, if the codec can handle it.

basically, when you compress it to mp3, unless you run some sort of rather harsh lowpass setting to actually take them out, you aren't losing any frequencies. when you decode the mp3, it's still a full spectrum of sound, it just sounds slightly different now after it was reconstructed. so the ogg vorbis encoder is now taking this slightly different sound and encoding it at the highest bitrate possible.

besides which, ogg vorbis and mp3 are completely different codecs with different methods of compressing audio. it doesn't stand to reason that once you compress a decoded mp3 into vorbis that there are savings that would transfer from one to the other because it's been compressed twice along the way. when you transcode a 192 kbps stereo (dolby surround) AC3 there's not going to be a smaller bitrate than you'd get from an uncompressed stereo .wav file.

try this: do you have the original track the 192 kbps mp3 was encoded from? encode that using the 100% vbr profile. then decode the 192 kbps mp3 and encode that using the same profile. the difference should be very small.

oh, and when you're encoding at a "500 kbps" setting, you're not asking ogg vorbis to be very efficient. if you want efficiency, use the -q vbr settings. and if you want more efficiency, use them at a setting that'll produce a lower bitrate.

...and rereading your original post, you were complaining that ogg at the 160 kbps setting came within +/- 10 kbps of 160 kbps. why should they have been smaller? you were telling it to encode at 160 kbps - it did what it was supposed to do.

one more thing... transcoding mp3 to ogg will never increase quality. it will only make it worse. imagine making a high-quality copy of a garbled fax. it'll still look bad.

hope this helps :)

[Doobie]

I know that MusicMatch 100% VBR MP3 isn't really 100%.

Any 192Kb/s CBR MP3 *is* dumping part of the audio signal, even if "frequency" wasn't the best choice of words.

It's true that the reproduced signal isn't exactly the same as the original, but a MP3 encoder has to deal with this as well and I also used a different MP3 encoder (LAME vs. FhG) and they do things a little differently.

And, I only used Ogg in what the software claimed is VBR mode, yet the results did look more like ABR.

I don't expect Ogg to increase the quality of the source, but I do expect a Ogg VBR mode to use less bits than a less efficent codec in a less efficient (CBR) mode.

My conclusion is that either dBPowerAmp (the best transcoding app around) is buggy (by giving me ABR Ogg when I ask for VBR) or that the Ogg does a poor job with high bitrates.

[canadian_fbi]

well, i would suggest using oggmachine/besweet or oggenc with the -q quality settings. ABR is technically a form of VBR, so it wasn't being completely wrong, just misleading. you'll get more efficient results that way. if you set it to 500 kbps ABR mode, then likely you were maxing out how many bits the encoder could give the audio. try it at something between -q 4 and -q 6 and that should work better.

and i don't know if ogg vorbis is really "inefficient", but there are better codecs to use for higher bitrates. go to hydrogenaudio.org and read up on mpc.

[[Toff]]

How do you compress you video (if you do) ?

A) Original Video -> Compressed at 600 kbps-> Compressed at 1000 kbps
B) Original Video -> Compressed at 1000 kbps

I let you guess what give the best result.

Have you ever try to recompress one of your video ?
It's often very hard to obtain good quality.

It's the same thing with audio.

If you want real details go to the hydrogenaudio forum and search for
reencoding or transcoding or recompress ...
http://www.hydrogenaudio.com

[DSPguru]

i didn't have the time to read the whole thread, but here's something you should know :
it's only natural that in a re-encoding process, mp3->mp3 would sound better than mp3->ogg.
try re-encoding ogg->ogg and ogg->mp3.
than you would probably conclude that mp3 sux... ;)

to summarize -
you could never measure quality of codec by comparing re-encodings.

[Head Hunter]

"dBPowerAmp (the best transcoding app around)"

I don't know what you're thinking, but that program sucks. The only reason anyone likes it is because it is "easy" to use. Try something like BeSweet with it's GUI. You'll appreciate the difference. When you actually have the command line, you don't need to worry about sliders that don't do what you want.

[Doobie]

OggDrop works exactly the same as dBPowerAmp. And, it appears to me that Ogg does NOT have quality settings per se. What it has is Average Bitrate settings that are passed off as quality settings. So, at a given quality setting instead of getting a larger file with a harder-to-encode audio track, you just get a lower quality file.

[canadian_fbi]

i can assure you that ogg vorbis DOES have quality settings. anyone here who has typed "-q" followed by a number in oggenc or besweet will say the same. try it out before you make such claims.

regarding hard-to-encode tracks, try this thread, about 3/4 way down:
http://www.hydrogenaudio.org/forums/showthread.php?s=&threadid=2651

[user]

http://www.hydrogenaudio.org/forums/showthread.php?s=&threadid=2985

[Doobie]

Reading it again, I noticed something that wasn't clear in what I said. I encoded the 192Kb/s file to 100% VBR MP3, not the original WAV. The Ogg file is 3 times bigger than the VBR MP3 even though the source was the same. That is point #1. Point #2 is that there doesn't seem to be much variation between file sizes relative to the quality setting.

Now, I'm sure in the most extreme conditions, such as pure tones vs. sounds of white water there would be a large variation in file size. But, until I see more evidence, my conclusion is that Ogg VBR doesn't work very efficiently.

Oh well, maybe this is something the Ogg Vorbis team will improve on in the future.

[canadian_fbi]

no, i understood that you encoded the 100% vbr mp3 from the 192 kbps one. that makes this all the more pointless.

the ogg file was three times bigger because you USED DIFFERENT SETTINGS. you told it to encode at 500 kbps. it only did 370 kbps. if anything, that's efficient. what did you expect, 64 kbps because the "100% vbr" mp3 was 116, thus proving that vorbis is more efficient?

that's point #1. as for #2.... how can you say that when you earlier stated that vorbis doesn't even have quality settings? you didn't even use them anyway, so your statement that they don't do much for the filesize is complete nonsense. furthermore, your "test" wouldn't have proven this even if you had done it correctly.

anyway, i'll clue the ogg vorbis team into this, and maybe they'll issue a recall of version 1.0 so that your startling observations are addressed. i'm sure they'll be alarmed to hear that their codec isn't performing efficiently, and in fact performs three times worse than mp3! just give me your name, address, home phone number and social security info, and i'll get them on the case.

[Emp3r0r]

:D

[buba king]

some people just amaze me ...:rolleyes:

[Doobie]

Canadian_fbi? You still don't get it. I didn't say that Ogg doesn't have quality settings only that it appears that the quality settings work more as average bitrate settings. That's not a difficult concept to grasp, is it?

Asking Ogg to encode at VBR 500Kb/s (that's q10) has nothing to do with efficiency. Efficiency is how small it can make the file at a given quality level. There's a difference. Can you see it?

And, I know it went above your head, but the point of encoding at 100% VBR MP3 and vs. encoding to Ogg was to provide a clue as to what file size is actually necessary to encode the audio data.

Besides, what's the deal with a 500Kb/s nominal or actual level anyway? I haven't heard complaints about MP3 lacking quality at 320Kb/s and isn't Ogg significantly better than MP3? You can get close to 320Kb/s with lossless compression, so unless Ogg isn't very efficient, what's the point of higher bitrate settings?

[Peters]

Doobie, is seems you don't want to understand.
Your test is a non-sense, codecs are not the same, psycho-models are different.

So to proove you are wrong, i made this test

Wave file 128Mo, 11 min 24 s

All encodings with besweet

Wav->Ogg q=6 Nominal bitrate 192 , average bitrate 144
(yes 144 kb/s, your theory on abr mode is false...)

Filesize=11,8 Mo

Ogg->mp3, preset alt extreme ->Filesize 18,2 Mo, average bitrate 222
Ogg->mp3, preset alt insane ->Filesize 26,1 Mo, average bitrate 320

So my conclusion, Ogg is really better than lame :)

Now maybe you understand this test is crazy

[oscarBravo]

Originally posted by Doobie
Canadian_fbi? You still don't get it. I didn't say that Ogg doesn't have quality settings only that it appears that the quality settings work more as average bitrate settings. That's not a difficult concept to grasp, is it?

Go do some reading up on how Vorbis works, as opposed to how it "appears" to work, and get back to us.

Oh yeah, and do your test with a clean source, not an MP3. Vorbis was never intended to be an efficient transcoder -- as you would already know if you had read up on the codec first.

Not that I'm at all clear on what you're trying to prove.