IPB

Welcome Guest ( Log In | Register )

 
Reply to this topicStart new topic
AAC quality vs MP3, at the same bitrate
AlphaQuam
post Aug 22 2003, 21:47
Post #1





Group: Members
Posts: 3
Joined: 6-January 03
Member No.: 4438



I'm currently using dbPowerAmp to convert my CDs to MP4. My library used to contain all 256kbps CBR MP3s. I'm encoding with -extreme VBR and when playing in WinAmp I get bitrates ranging from 192-256. I was wondering, will an MP4 sound better than an MP3 at the same bitrate? With my limited experience I would say yes, just wondering if there was an approximate scale... like 192 MP4 ~= 256 MP3.
Go to the top of the page
+Quote Post
NumLOCK
post Aug 22 2003, 22:25
Post #2


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



AAC was indeed made to be of higher quality than MP3.

However, don't forget that these formats were designed (and advertised !) from the beginning for ~96..160 kbps. When you go higher in bitrate, at some point the maturity (or tuning, cleverness) of the encoder become more important than the format itself.

In other words: at a given bitrate, with AAC you may get a "technically" better (ie: more accurate) sound, but there may be more spots where the encoder fails (ie: produces an artifact in the output file). LAME will have a bit less trouble cases.

Typically, at high bitrates, short and precise artifacts can be easier to spot than an overall less accurate signal (MP3).

Don't panic though: 160+ kbps AAC will sound very good wink.gif
If you want to spare bits, yet have low pre-echo, it will be a better choice.

This post has been edited by NumLOCK: Aug 22 2003, 22:41


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
JohnV
post Aug 22 2003, 22:43
Post #3





Group: Developer
Posts: 2797
Joined: 22-September 01
Member No.: 6



QUOTE (NumLOCK @ Aug 23 2003, 12:25 AM)
In other words:  at a given bitrate, with AAC you may get a "technically" better (ie: more accurate) sound, but there may be more spots where the encoder fails (ie: produces an artifact in the output file).  LAME will have a bit less trouble cases.

I don't quite understand what you are trying to say. AAC has all what is needed to be superior to mp3 in every case. It just depends on encoder tweaking really...
If you are talking specifically about Psytel encoder (which dbPoweramp uses iirc), that is a different thing, but in that case be more specific...


--------------------
Juha Laaksonheimo
Go to the top of the page
+Quote Post
danchr
post Aug 23 2003, 17:42
Post #4





Group: Members
Posts: 487
Joined: 6-April 03
From: Århus, Denmark
Member No.: 5861



My impressions from encoding QuickTime AAC is that 192kbps AAC is good enough for almost anything, and for the really good songs 224kbps is transperant. I'd compare 192kbps AAC to 256kbps MP3 and 224 to 320. The main reason I'm using AAC, though, is that it encodes much faster, so I'd rather throw too many bits at it than too few smile.gif

I don't have any data to back this; I'm a mac user so I can't ABX (and I'm too lazy to do it without software) wink.gif Some guy - who seemed to know what he was talking about - made a comparison between iTunes AAC and iTunes MP3, but I can't seem to find it...
Go to the top of the page
+Quote Post
Dologan
post Aug 23 2003, 19:03
Post #5





Group: Members (Donating)
Posts: 478
Joined: 22-November 01
From: United Kingdom
Member No.: 519



QUOTE (danchr @ Aug 23 2003, 10:42 AM)
[...] and for the really good songs 224kbps is transparent.

Good songs don't necessarily need more kbps to be transparent than bad ones. wink.gif Take classical music for example. It is consistently easy on bitrate and is generally infinitely better than some of the pop crap out there. Hmm... ok, let's avoid about musical preferences.

EDIT: OK, I've just realized what you really meant by that. Nevermind what I just said. tongue.gif

This post has been edited by dologan: Aug 23 2003, 19:08
Go to the top of the page
+Quote Post
danchr
post Aug 23 2003, 20:38
Post #6





Group: Members
Posts: 487
Joined: 6-April 03
From: Århus, Denmark
Member No.: 5861



As you probably guessed, what I meant was that I only notice minor artifacts on the songs I really like. You listen much more carefully to the good stuff (hopefully) wink.gif

Actually, I'd disagree that classical music is easy to compress. There can be so many high frequency tones and fine nuances in them that even LAME alt-preset standard doesn't cut it.
Go to the top of the page
+Quote Post
G-Force
post Aug 23 2003, 20:49
Post #7





Group: Members
Posts: 56
Joined: 8-June 03
From: Gainesville, FL
Member No.: 7068



QUOTE (danchr @ Aug 23 2003, 08:42 AM)
I don't have any data to back this; I'm a mac user so I can't ABX (and I'm too lazy to do it without software) wink.gif

There's a mac version of ABX that you can run through the classic environment. Get it here.

http://www.pcabx.com/program/
Go to the top of the page
+Quote Post
Dologan
post Aug 23 2003, 21:07
Post #8





Group: Members (Donating)
Posts: 478
Joined: 22-November 01
From: United Kingdom
Member No.: 519



QUOTE (danchr @ Aug 23 2003, 01:38 PM)
Actually, I'd disagree that classical music is easy to compress. There can be so many high frequency tones and fine nuances in them that even LAME alt-preset standard doesn't cut it.

Well, usually I don't take the time to compare the encoded versions to the originals, but I have never found anything wrong with APS-encoded classical music, at least in the environment where I use MP3 (portable, mostly). Besides, I am not very sensitive to artifacts, anyway. wink.gif
What I am actually referring to is that bitrate tends to be consistently lower with classical music than with most other types of music (except for harpsichord music, of course). Whether it should be otherwise or not, I can't say, but it tends to be enough for me.
Go to the top of the page
+Quote Post
bidz
post Aug 23 2003, 22:34
Post #9





Group: Members
Posts: 351
Joined: 27-December 02
From: Norway
Member No.: 4258



what about Nero (6.0.0.11) MP4 (AAC) VBR versus MP3 (lame 3.90.3) ? for example, --alt-preset standard versus Nero's Normal::High preset?

I've been thinking about ditching mp3 totally, and go with mp4 instead.. but im unsure about the real quality.


--------------------
myspace.com/borgei - last.fm/user/borgei
Go to the top of the page
+Quote Post
guruboolez
post Aug 23 2003, 22:38
Post #10





Group: Members (Donating)
Posts: 3474
Joined: 7-November 01
From: Strasbourg (France)
Member No.: 420



[quote=dologan,Aug 23 2003, 09:07 PM] [QUOTE=danchr,Aug 23 2003, 01:38 PM]What I am actually referring to is that bitrate tends to be consistently lower with classical music than with most other types of music (except for harpsichord music, of course).
[/quote]
I partially disagree : lame VBR is very cool with classical, and unfortunately with harpsichord too. Vorbis and musepack have the opposite behaviour. Nero AAC encoder is similar too lame --aps : average bitrate is lower than what we can expect, and see with other kind of music, and there are problems (clear and very annoying for me). Ivan worked on the problem, but artifacts remains.
Go to the top of the page
+Quote Post
NumLOCK
post Aug 23 2003, 23:22
Post #11


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



QUOTE (JohnV @ Aug 22 2003, 10:43 PM)
QUOTE (NumLOCK @ Aug 23 2003, 12:25 AM)
In other words:  at a given bitrate, with AAC you may get a "technically" better (ie: more accurate) sound, but there may be more spots where the encoder fails (ie: produces an artifact in the output file).  LAME will have a bit less trouble cases.

I don't quite understand what you are trying to say. AAC has all what is needed to be superior to mp3 in every case. It just depends on encoder tweaking really...
If you are talking specifically about Psytel encoder (which dbPoweramp uses iirc), that is a different thing, but in that case be more specific...

Sorry for answering late.

What I meant was quite hard to state - so I'll try to re-express it:

With the superior format (AAC) comes higher overall quality (at equal bitrates), but there are risks associated to brand-new codecs (ie: slight psymodel "failures" from time to time).

Then I said, on some samples, a very consistent, albeit lower quality sound (LAME MP3) may be preferrable to a better sound which has more dropouts or glitches.

There is a simple testing case, which is not psymodel-related but still relevant:

At about ~80kbps, compare a Vorbis file to a crystal clear Musepack file.

As a whole Musepack could be better, but it raises a problem: many attacks sound (to me) like broken ice ("piou-piou"). In other words, even if the sound (as a whole) is more accurate, the artifacts kill the picture. Therefore the less accurate (but more consistent) file is (for me) a better choice.

I hope this doesn't sound ridiculous wink.gif


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
Dologan
post Aug 23 2003, 23:33
Post #12





Group: Members (Donating)
Posts: 478
Joined: 22-November 01
From: United Kingdom
Member No.: 519



Heh. tongue.gif I guess you're right. I was kinda mixing up my experience with MP3 and other formats. I had a look at some harpsichord tracks encoded with Lame and indeed they have a lower than average bitrate, the same as most classical. Musepack OTOH has an average bitrate with classical similar to other types of music, with harpsichord boosting it ~20 kpbs more. I don't have harpsichord music encoded with Vorbis at the moment, but it seems to have the same behaviour as Lame with classical music, that is, lower than average bitrate. I don't have any experience encoding classical with AAC yet, so I can't say anything about it.

~Dologan

This post has been edited by dologan: Aug 24 2003, 00:31
Go to the top of the page
+Quote Post
NumLOCK
post Aug 24 2003, 00:21
Post #13


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



QUOTE (dologan @ Aug 23 2003, 11:33 PM)
(...) Musepack OTOH has an average bitrate with classical similar to other types of music, with harpsichord boosting it ~20 kpbs more. I don't have harpsichord music encoded with Vorbis at the moment, but it seems to have the same behaviour as Lame with classical music, that is, lower than average. (...)

I think it is the price to pay for next-to-none pre-echo. Harpsichord notes start very abruptly, look at them in a spectrogram (or simply at the waveform) wink.gif


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
guruboolez
post Aug 24 2003, 00:42
Post #14





Group: Members (Donating)
Posts: 3474
Joined: 7-November 01
From: Strasbourg (France)
Member No.: 420



QUOTE (NumLOCK @ Aug 24 2003, 12:21 AM)
I think it is the price to pay for next-to-none pre-echo.  Harpsichord notes start very abruptly, look at them in a spectrogram (or simply at the waveform)  wink.gif

Not true. Meditative harpsichord music, with few notes per minutes, need much more bitrate than presto, fast music, with more notes, and therefore more attacks. Mpc is very sensitive to tonal music : for exemple, a single violon is ~200 kbps with --standard profile, without any attack - compare it with any kind of metal music, rich in percussions (~175 kbps). Musepack abnormal behaviour with this instrument can't be explain with attacks only.
Attacks is maybe lowering the average bitrate. Just try to find a rack with a tonal instrument (wind, string), playing alone for a moment, and then in duo with another instrument (percussion). The fisrt part (tonal, no attacks) is much heavier than the second (tonal with attacks).


Can't explain why, I'm not an expert (something to do with perceptual masking ?).
Go to the top of the page
+Quote Post
NumLOCK
post Aug 24 2003, 09:59
Post #15


Neutrino G-RSA developer


Group: Developer
Posts: 852
Joined: 8-May 02
From: Geneva
Member No.: 2002



QUOTE (guruboolez @ Aug 24 2003, 12:42 AM)
Not true. Meditative harpsichord music, with few notes per minutes, need much more bitrate than presto, fast music, with more notes, and therefore more attacks. Mpc is very sensitive to tonal music : for exemple, a single violon is ~200 kbps with --standard profile, without any attack - compare it with  any kind of metal music, rich in percussions (~175 kbps). Musepack abnormal behaviour with this instrument can't be explain with attacks only.
Attacks is maybe lowering the average bitrate. Just try to find a rack with a tonal instrument (wind, string), playing alone for a moment, and then in duo with another instrument (percussion). The fisrt part (tonal, no attacks) is much heavier than the second (tonal with attacks).


Can't explain why, I'm not an expert (something to do with perceptual masking ?).

Oh. Okay, I stand corrected about meditative music vs presto. What we could do, though, is to cut/paste identical attacks to create artificial songs with various tempos, but known properties. With identical attacks, the cause for bitrate variations would be easier to find.

One could also playback take a midi file on a high-quality soundcard, and change the bpm.

About the relatively high bitrate on tonal sounds (with low harmonics), don't forget that all other codecs work in the tonal domain (1 coefficient = 1 cosine), while Mpc stays (essentially) in the time domain (1 coefficient = 1 subband sample).. this should explain their size advantage.

Metal music should offer various masking possibilities, because of so many harmonics. Plus it contains noisy components, for which frequency-domain encoders bring few benefits. Other than that I don't know..


--------------------
Try Leeloo Chat at http://leeloo.webhop.net
Go to the top of the page
+Quote Post
Ivan Dimkovic
post Aug 25 2003, 17:35
Post #16


Nero MPEG4 developer


Group: Developer
Posts: 1466
Joined: 22-September 01
Member No.: 8



Harpsichord poses problems to codecs which use derivation of ISO/Psymodel II with tonality estimation based on prediction of the magnitude and phase of the FFT spectrum.

Problem lies because of the short-term harmonic structure harpsichord trips tonality detector which classifies that sound as more noise like than tonal (in reality - the signal is perceived as tone in human ear), and because noise-like signals have much higher masking power the final result is a undercoded signal.

Different psymodels do not calculate measure of "tonality" at all, but derive masking power from the temporal uncertanity of the signal envelope, or use non-linear additivity of the individual maskers, or use some kind of peak filter to determine is signal tonal or noise like.

Problem with these other methods is that for some signals they overcode, giving higher bit rate than really needed and therefore making them very hard (but possible) to tune for CBR or low bit rate VBR encodings.


== edit

Pure tonal signals require high bit rate because pure tonal maskers have less masking power than the same amount (in dB) of noise-like maskers (sometimes more than 15 dB!). So, adding few attacks in the trumpet + violin sections might give lower bit rate because of much higher masking power of the resulting signal.

This post has been edited by Ivan Dimkovic: Aug 25 2003, 17:39
Go to the top of the page
+Quote Post
sld
post Aug 25 2003, 17:51
Post #17





Group: Members
Posts: 1015
Joined: 4-March 03
From: Singapore
Member No.: 5312



Would it then be possible to make coding pure tones in mpc more efficient? Or is this precisely a limitation of mpc as a subband codec?
Go to the top of the page
+Quote Post
2Bdecided
post Aug 26 2003, 14:53
Post #18


ReplayGain developer


Group: Developer
Posts: 4945
Joined: 5-November 01
From: Yorkshire, UK
Member No.: 409



QUOTE (guruboolez @ Aug 23 2003, 11:42 PM)
QUOTE (NumLOCK @ Aug 24 2003, 12:21 AM)
I think it is the price to pay for next-to-none pre-echo.  Harpsichord notes start very abruptly, look at them in a spectrogram (or simply at the waveform)  wink.gif

Not true. Meditative harpsichord music, with few notes per minutes, need much more bitrate than presto, fast music, with more notes, and therefore more attacks. Mpc is very sensitive to tonal music : for exemple, a single violon is ~200 kbps with --standard profile, without any attack - compare it with any kind of metal music, rich in percussions (~175 kbps). Musepack abnormal behaviour with this instrument can't be explain with attacks only.
Attacks is maybe lowering the average bitrate. Just try to find a rack with a tonal instrument (wind, string), playing alone for a moment, and then in duo with another instrument (percussion). The fisrt part (tonal, no attacks) is much heavier than the second (tonal with attacks).


Can't explain why, I'm not an expert (something to do with perceptual masking ?).

Ivan has probably explained it properly, but I always (maybe wrongly) thought that is was the "clear voice detection" at work - see original Musepack home page.

This appears to increase the bitrate where the short-term spectrum looks noise-like, but the human perception is tone-like - similar to what Ivan is suggesting I think.

Cheers,
David.
Go to the top of the page
+Quote Post
sony666
post Aug 28 2003, 20:14
Post #19





Group: Members
Posts: 573
Joined: 22-February 02
Member No.: 1375



QUOTE (bidz @ Aug 23 2003, 11:34 PM)
what about Nero (6.0.0.11) MP4 (AAC) VBR versus MP3 (lame 3.90.3) ? for example, --alt-preset standard versus Nero's Normal::High preset?

yes, this is where it becomes interesting smile.gif
Go to the top of the page
+Quote Post
guruboolez
post Aug 29 2003, 22:55
Post #20





Group: Members (Donating)
Posts: 3474
Joined: 7-November 01
From: Strasbourg (France)
Member No.: 420



QUOTE (Ivan Dimkovic @ Aug 25 2003, 05:35 PM)
Harpsichord poses problems to codecs (...)

Nice explanation ! Thanks a lot for the time spent to write it smile.gif
Go to the top of the page
+Quote Post
wkwai
post Aug 30 2003, 10:07
Post #21


MPEG4 AAC developer


Group: Developer
Posts: 398
Joined: 1-June 03
Member No.: 6943



QUOTE (NumLOCK @ Aug 24 2003, 12:59 AM)
The fisrt part (tonal, no attacks) is much heavier than the second (tonal with attacks).


Can't explain why, I'm not an expert (something to do with perceptual masking ?).

I believe it had something to do with Post Masking effect?? I noticed that Psytel AACEnc has a "temporal masking" settings.. I have not implemented any post masking compensation in my psychoacoustice models and I found for both Model I and Model II, the penalties on bitrates for switching to SHORT BLOCK during attacks is very severe.. More that 2 times the bitrates of LONG BLOCK..
As a result, the bit reservoir can be rapidly depleted in sample clips that require a lot block switching..


wkwai

This post has been edited by wkwai: Aug 30 2003, 10:08
Go to the top of the page
+Quote Post
danchr
post Aug 30 2003, 17:01
Post #22





Group: Members
Posts: 487
Joined: 6-April 03
From: Århus, Denmark
Member No.: 5861



My own, subjective experience with QuickTime AAC is:

128kbps AAC ~ 160kbps MP3: This is Good Enough™ for Joe User, but an experienced listener will notice artifacts.
160kbps AAC ~ 192kbps MP3: The sound is definitely tolerable, but it's not transparent and would probably fail on numerous problem samples.
192kbps AAC ~ 256kbps MP3: "Virtually transparent" - if such a thing exists. It sounds quite good, but may fail now and then, especially on classical music, IMO.
224kbps AAC ~ 320kbps MP3: This is transparent to me. If it isn't, it's either my mind playing tricks with me, or me being too picky wink.gif

Given that QuickTime AAC is a CBR codec, you can just throw as many bits at it as you want, and hopefully improve quality. I don't think that ABX makes any sense if you want to determine your preferred quality setting. With ABX testing, you heighten your awareness for artifacts, and that's counterproductive if you want to determine how low a bitrate satisfies you. I haven't seen any data that contradicts this yet smile.gif

I use 192kbps AAC for normal music, and 224kbps AAC for classical music (or the really good stuff). I used LAME aps and LAME insane before I switched to AAC.
Go to the top of the page
+Quote Post
blessingx
post Sep 5 2003, 21:38
Post #23





Group: Members
Posts: 256
Joined: 22-April 03
From: just west of san fran
Member No.: 6131



Danchr, are you using the iTunes FhG encoder for those numbers or the mentioned Lame? If the former and you feel like testing it there's the iTunes-LAME plugin for the LAME encoder (though you mentioned --aps/--api). I'm actually using Lame --apfs and QT AAC 160-192 for my iPod (w/Sony V6's). Which sounds better to my ears depends on the trade-offs of "clarity" versus "warmth" for specific music.

BTW, I think QT AAC is technically not CBR (with its variable bit reservoir), though it certainly performs like it (try recording silence).

This post has been edited by blessingx: Sep 5 2003, 21:44
Go to the top of the page
+Quote Post
danchr
post Sep 6 2003, 11:06
Post #24





Group: Members
Posts: 487
Joined: 6-April 03
From: Århus, Denmark
Member No.: 5861



QUOTE (blessingx @ Sep 5 2003, 10:38 PM)
Danchr, are you using the iTunes FhG encoder for those numbers or the mentioned Lame?

Yes, I used LAME. Those numbers aren't really based on anything but a subjective feeling, though. The main reason for switching was that LAME is horribly slow on a 400MHz G3; ripping 0,6-0,7x speed is unacceptable in the long run. I can't tell the difference between 192kbps AAC and the source, so I'm pretty satisfied with it. 192kbps AAC would probably fail on problem samples, but I haven't noticed anything yet.

QUOTE
BTW, I think QT AAC is technically not CBR (with its variable bit reservoir), though it certainly performs like it (try recording silence).

I'm pretty certain that MP3 CBR behaves the same way. IIRC they both have a bit reservoir. CBR is maximising it's usage, VBR is minimising and varying the bit rate instead.

Anyway, I found the link to pretty much the only test on this field; it's in this thread.

EDIT: There's a thread about bit reservour in MP3 right here smile.gif

This post has been edited by danchr: Sep 6 2003, 11:12
Go to the top of the page
+Quote Post

Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 19th April 2014 - 11:05