Author Topic: opus, aac or something else?  (Read 22647 times)

hiccup

  • Sr. Member
  • ****
  • Posts: 7798
Flac is an even clearer example.
You can set it to use different compression levels.
All resulting in the same bit-rate output, but with different file sizes.
(but also pretty much negligible in practice)
Well, FLAC is variable bit rate, which is not what we're talking about here, but even on its face this claim is decidedly not true.   The definition of compression is changing the bit rate.  

It looks like we are slightly talking past each other.
I was interested in differences in resulting storage file sizes.
For lossless flac, that will most certainly depend on the (0-8) setting. I am sure you didn't intend arguing that?

I have always been convinced that different lossy algorithms would perform differently in resulting file sizes depending on how sophisticated or advanced they are.

I did some quick testing with dBpoweramp, which has the option to refine any fixed cbr setting to 10 different quality levels.
I always assumed that higher quality settings would not only (substantially) increase the encoding time, but it also would result in larger file sizes, since more refined audio information would need to be stored.
It also has the option to select e.g. 'stereo' encoding vs. 'joint-stereo' encoding.
Again, I have always assumed and understood that 'stereo' would take up more storage space then encoding 'joint-stereo'.

But trying out these different settings (with the same 320cbr setting) indeed always results in the same file size.
That's really surprising to me. So I'll probably need to re-adjust my understanding on this matter.

But to return on-topic: I was responding to the notion that some users would use Opus also at higher bit-rates, because it would result in lower file-sizes than mp3, with comparable (or even better) sound-quality.
And on that, I have not found any reliable proof or abx testing.


frankz

  • Sr. Member
  • ****
  • Posts: 3840
It looks like we are slightly talking past each other.
I was interested in differences in resulting storage file sizes.
For lossless flac, that will most certainly depend on the (0-8) setting. I am sure you didn't intend arguing that?
No, that's exactly right.  FLAC 0 -> 8 adjusts the target bitrate on a downward scale as shown in my images, which affects the file size also on a downward scale.  
I have always been convinced that different lossy algorithms would perform differently in resulting file sizes depending on how sophisticated or advanced they are.
I did some quick testing with dBpoweramp, which has the option to refine any fixed cbr setting to 10 different quality levels.
I always assumed that higher quality settings would not only (substantially) increase the encoding time, but it also would result in larger file sizes, since more refined audio information would need to be stored.
It also has the option to select e.g. 'stereo' encoding vs. 'joint-stereo' encoding.
Again, I have always assumed and understood that 'stereo' would take up more storage space then encoding 'joint-stereo'.

But trying out these different settings (with the same 320cbr setting) indeed always results in the same file size.
That's really surprising to me. So I'll probably need to re-adjust my understanding on this matter.
I encode video and audio all day every day pretty much.  File Size will always equal bit rate per second times seconds.  Always. For CBR, it's easy to do the math, because each frame is encoded with the same bit rate (320 for example).  For VBR, there is no fixed bit rate because it's variable on each frame - silence in audio is encoded with 0 bits for example - so you can't ever know the exact size of the the resulting file, but you can estimate based on the target bit rate.

Quality settings in encoders have to do with how they get to the bit rate in question. They may futz this or that set of frequencies differently, rolling off above or below a certain frequency, or use one channel to cancel out what's in another channel so it's not encoded twice. But they can't affect the underlying math, which is always File Size = bits per second * seconds.  That is a mathematical constant.
But to return on-topic: I was responding to the notion that some users would use Opus also at higher bit-rates, because it would result in lower file-sizes than mp3, with comparable (or even better) sound-quality.
And on that, I have not found any reliable proof or abx testing.
I'm with you there. I've found no reason to use any lossy compression besides MP3.  They may get better quality at low bit rates, and that may please some people who value storage space over all else, or maybe if you're encoding audio books where you're looking at really low bit rates and not all that concerned about quality, but, for music, V0 MP3 will serve you well as it has served everyone else well for a very long time now.

hiccup

  • Sr. Member
  • ****
  • Posts: 7798
Thanks for enlightening me on this matter.
I now understand why my assumptions on mp3 cbr file sizes have been wrong.

I always had the idea that each different mp3 encoding algorithm -besides from doing 'it's own magic' in encoding to lossy audio- would also do it's own magic in the way it saves the encoded file.
(similar to .zip, .winrar, 7zip etc.)

Technically that surely would be possible, and that would result in different file sizes, but now I understand they just cannot do that, since they simply have to adhere to mpeg (playback) standards.

GoodGuy

  • Newbie
  • *
  • Posts: 5
  • Full time webmaster without hobbies ;)
I think some of you understood my previous post because someone started to talk about file size (Opus vs MP3) but where compressing them at the same bitrate...


The thing is that Opus is able to maintain a better sound quality at lower bitrates which will make the file size smaller.
And that was the whole point.


Example:
Try a 64 kbps OPUS files vs a 64 kbps MP3 file and convert them from the same noloss source.
Both files will be about the same size but Opus will sound way better. I would even say that Opus at 32kbps vs MP3 at 64kbps , Opus will still win.

But of course you cant assume you'll get CD-quality Opus files at 64 kbps, I have no idea what bitrate the people behind Opus would compare it with.

Incase you want to try it out:
LameXP is a pretty good GUI for LAME Enc and Opuslib and a couple of other codecs if you want to play around. :)
http://lamexp.sourceforge.net/

frankz

  • Sr. Member
  • ****
  • Posts: 3840
I didn't misunderstand your post. I stated that that's opus folks' claim more than once.

And it may be entirely correct.  But who wants to listen to anything at either 32kbps or 64kbps anyway except, as I said, for audio books maybe?  The whole exercise is academic rather than practical.  

As a practical matter, when you get to bitrates people actually use for music in real life, you're better off going with what is now and will for the foreseeable future be the most compatible.

Ansem

  • Jr. Member
  • **
  • Posts: 99
It was a really interesting post!
So the thing is that still there is nothing relevanty better than mp3 at 320kbps right? I was interested in opus also because it can have an higher bitrate than 320kbps.

NickIst

  • Jr. Member
  • **
  • Posts: 101
It was a really interesting post!
So the thing is that still there is nothing relevanty better than mp3 at 320kbps right? I was interested in opus also because it can have an higher bitrate than 320kbps.
To my mind AAC is better, it provides the same or maybe even better quality at 256 VBR. Plus MP3 is dying not because it won't be supported but because it wont'be developped further. For example the last version of LAME (3.100) has some minor improvements and adjustments which are not connecnted with the encodinng process at all. Concerning the support of the formats other than mp3, only some very conservative markets don't add another formats, but even in factory car head units AAC support has been added recently, before that usually only two formats were widely supported: MP3 and WMA. And that support will be enlarged because AAC is widely used in Apple systems, that's one of the reasons why car makers added it in their head units. And android devices, HD players, Hi-Fi players always have wider choice of formats of playback. And the last one mp3 has more distortions than other lossy formats.

https://audiophilesoft.ru/publ/my/encoders_test/11-1-0-53
It's on Russian but authomatic translation will help, maybe ::)

P.S. As for me I'll soon start to replace all my mp3 to aac. For me it's the best option concerning compatibility. Opus is very good and really efficient format but it won't be playbacked in my car head unit. Maybe later I'll use it in my phone.
Last Edit: February 26, 2018, 12:39:13 PM by NickIst

frankz

  • Sr. Member
  • ****
  • Posts: 3840
It was a really interesting post!
So the thing is that still there is nothing relevanty better than mp3 at 320kbps right? I was interested in opus also because it can have an higher bitrate than 320kbps.
To my mind AAC is better, it provides the same or maybe even better quality at 256 VBR. Plus MP3 is dying not because it won't be supported but because it wont'be developped further. For example the last version of LAME (3.100) has some minor improvements and adjustments which are not connecnted with the encodinng process at all.
If it already provides transparent encoding, what further encoding development is required?  Audio can't get more transparent.

NickIst

  • Jr. Member
  • **
  • Posts: 101
But the so-called transparency is acheived on certain bit rates. And each codec can have or has it's own threshold of transparency. It's considered that mp3 has pre-echo and noise shaping at ~200 kbps VBR. And when we speak about lossy codecs we don't talk only about quality but also about efficiency. That's where for example AAC is better because it uses less data to provide the same or better quality. Plus according to technical data AAC is more advanced like better encoding of 16kHz, more flexible joint stereo, better transients encoding etc. And in the link mentioned above it was shown that aac, ogg etc have less distortion and closer to the original source.
Although everyone decides by himself what to use, because for some mp3 128 CBR is more than ok.

Quote
If it already provides transparent encoding, what further encoding development is required?  Audio can't get more transparent.
Original audio can't, but encoded sound (a lossy one) can with certain reservations.
 Plus that update of LAME which was mentioned above has been long awaited. If I'm not mistaken for 5 years, so for some it was a bit disappointing

P.S. sorry for posting russian sites https://audiophilesoft.ru/publ/my/encoding_recommendations/11-1-0-83
There is a small comparison of several not mp3 encoders
Last Edit: February 26, 2018, 05:53:03 PM by NickIst

frankz

  • Sr. Member
  • ****
  • Posts: 3840
Having spent the past two days with every AAC encoder I could get my hands on (QAAC, ffmpeg with native aac, ffmpeg with fdk_aac, iTunes) set to high bit rates (QAAC went up as high as 353 at times!), I can say without a moment's hesitation that there is no audible increase in quality between any of them and LAME at -V0.  None.  Zero.  FFMPEG's native AAC sounds worse than LAME.  That's the only difference I heard.

There may be a laboratory-measurable difference in peak levels, but machine-sensed differences in the articles cited are not audible differences. 

For someone to reencode their entire library from mp3 at -V0 to AAC would be a waste of time IMO.  There's no harm as far as I can tell in switching over to using the higher quality AAC encoders, but there's no audible benefit to it either.

hiccup

  • Sr. Member
  • ****
  • Posts: 7798
Having spent the past two days with every AAC encoder I could get my hands on (QAAC, ffmpeg with native aac, ffmpeg with fdk_aac, iTunes) set to high bit rates (QAAC went up as high as 353 at times!), I can say without a moment's hesitation that there is no audible increase in quality between any of them and LAME at -V0.

Thanks for sharing this.
I have made the choice for using either mp3 or flac a very long time ago, and never looked back, but after reading this topic, and investigating aac a bit more, learning about specific technical (theoretical) improvements on aac vs. mp3, I started to doubt if I shouldn't open the drawbridge for aac.
Besides that I wondered if the gains would be sensible for a lossy format anyway, another reason for ignoring aac was the difference in it handling metadata, and the fact that MusicBrainz' Picard doesn't support aac.

Your testimony reassures me I shouldn't really worry about it any further.

Just curious: would you mind to share the equipment you used for comparing this?
(dac/headphones/speakers)

frankz

  • Sr. Member
  • ****
  • Posts: 3840
Dragonfly Black DAC/AMP + Senneiser HD598 headphones. Decent consumer grade stuff.   Not earbuds, not audiophile grade.

hiccup

  • Sr. Member
  • ****
  • Posts: 7798
Dragonfly Black DAC/AMP + Senneiser HD598 headphones. Decent consumer grade stuff.   Not earbuds, not audiophile grade.

Thnx. That's certainly more than good enough to judge lossy formats.

frankz

  • Sr. Member
  • ****
  • Posts: 3840
Best way to judge is to grab qaac (they even tell you how to set it up in MB) and the most recent LAME and do a shoot out.  I'm very interested in getting the best sound for myself and would have switched over to aac if it had proved to be an improvement.  It just wasn't.

NickIst

  • Jr. Member
  • **
  • Posts: 101
For someone to reencode their entire library from mp3 at -V0 to AAC would be a waste of time IMO.  There's no harm as far as I can tell in switching over to using the higher quality AAC encoders, but there's no audible benefit to it either.
Yes I agree that there won't be audible benefit especially on the ordinary (not Hi-Fi or smth) equipment. But for me another there is another benefit of using AAC - saving disk space without any loss in quality comparing to mp3.
Last Edit: February 28, 2018, 11:14:49 PM by NickIst