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: (AAC) advantages & disadvantages (Read 53823 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

(AAC) advantages & disadvantages

Reply #25 –
Quote from: rpp3po on
Quote from: Mike Giacomelli on
Since this doesn't appear to refute or even address the numerical counter-evidence I have provided to show that you are wrong,..


By numerical counter-evidence you mean the hydrogenaudio user polls, that are supposed to support your argument that 6 years of additional development basically don't have improved AAC over MP3??


You've edited my sentence to say something other then it originally did.  Please don't do that.  The original was perfectly clear before you removed its meaning.

Quote from: rpp3po on
The ISO example you cite, shows perfectly why spec compliant MP3 is basically fixed bitrate oriented in a crude way.


Quote from: rpp3po on
VBR support has never been officially included for MP3.


Quote from: rpp3po on
ISO/IEC 11172-3:1993 does not mention VBR. Only the bitrates, that I have cited.

What you claim is wrong and I can't cite what's not there. Ask some LAME devs if you don't believe it.


I'm curious, when you do this thing where you pretend you haven't been wrong and insulting the entire time, is it because it would hurt your ego to admit a mistake, or because you're just too big a jerk to care about being civil?

Quote from: rpp3po on
That's not VBR but VFBR, variable fixed bitrates.


haha

Quote from: rpp3po on
If you want to see it done right, look at AAC. For every frame there can be allocated as many bytes as needed.


MP3 too.

Quote from: rpp3po on
There's also generally less interdependency between frames, which makes it a lot easier for true VBR.


Nonsense.  AAC frames still have overlap because of the IMDCT.

Quote from: rpp3po on
That's also not true, btw...  Before I had become a member at HA you never had even mentioned AAC. The first time you commented on it was five days later. And doesn't that look that familiar?


You're really creepy.

(AAC) advantages & disadvantages

Reply #26 –
Quote from: rpp3po on
Well, I guess you didn't get that I can't because it's not in there. ISO/IEC 11172-3:1993 does not mention VBR. Only the bitrates, that I have cited.

What you claim is wrong and I can't cite what's not there. Ask some LAME devs if you don't believe it.


"In order to provide the smallest possible delay and complexity, the decoder is not required to support a continuously variable bitrate when in LayerI or II. Layer III supports variable bitrate by switching the bit_rate_index."
quote from ISO/IEC 11172-3.

(AAC) advantages & disadvantages

Reply #27 –
Ok, I obviously went too far, when I said it isn't mentioned. On the bitstream level each frame can have one of the 15 pre defined bitrates instead of an almost arbitrary frame size (between smallest and biggest allowed frame size) in AAC.

I initially claimed that players choke - the stream decoder is only one part of this. There is still no official spec for VBR file handling.* This manifests in the inability to both display a file's total playback time and provide skipping capability. The only possibility would be a full file scan before each playback. And tell me just one player that would do the latter, it's just far too ressource intensive for real world implementations.

In reality the lack of an official spec nowadays still results in players, that don't only fail to display a file's total playback time and provide skipping abilities, but choke completely when fed with VBR files. It breaks their implementation's assumptions (while their internal stream decoder might well be able to output a signal). Several onboard MP3 players in Audi's current lineup fall into that category. And since there is no official spec that would have been broken, you can't return the units for not delivering what had been promised. AFAIK the latter is no problem in AAC's (or the mp4 container format's) specification because there is an official definition instead of three different software developer created pseudo standards as in the case of MP3.

* And having to chose one of 15 predefined bitrates for each frame isn't optimal, either. That's why I called MP3 basically a CBR based design, where VBR operation is an extension (at least on a file level) and AAC a basically VBR based design where CBR and ABR operation are optional constraints.

(AAC) advantages & disadvantages

Reply #28 –
Quote from: rpp3po on
* And having to chose one of 15 predefined bitrates for each frame isn't optimal, either.


Those 15 framesizes are just for segmentation. The audio (or main) data belonging to that frame's side info can be spread over multiple frames or take up less than 1 frame. It allows for any byte size main data, just like in AAC, the segmentation is just much more complicated.

(AAC) advantages & disadvantages

Reply #29 –
Besides the technical details, sorry for being such a dick head this time, Mike Giacomelli.

(AAC) advantages & disadvantages

Reply #30 –
Quote from: rpp3po on
Besides the technical details, sorry for being such a dick head this time, Mike Giacomelli.


Accepted, but I still take issue with much of what you've said: 

Quote from: rpp3po on
I initially claimed that players choke - the stream decoder is only one part of this.


Of course decoders can have bugs.  No one disagreed.  Hell I even linked you to an AAC decoder with an equivalent bug.  This is so obvious it goes without saying.

Quote from: rpp3po on
There is still no official spec for VBR file handling.*


What you actually meant to say is that theres no official container that provides seeking information for MP3, and that is closer to the truth.  ISO/IEC 11172-3 only defines a bitstream, not the container.  AAC works the same way, with ISO/IEC 13818-7 only defining the bitstream and not the container as well.  There is actually an MPEG container for MP3 that provides everything you mention, I believe its defined in ISO/IEC 14496-1 and called MP4 (though its not used much in practice) as well as a number of containers that officially do support it and are used in practice (RA and ASF for instance).

The problem you're getting at is that people have decided in practice to use raw MP3 bitstreams, which is why seeking is tricky, particularly for old decoders.

Quote from: rpp3po on
Several onboard MP3 players in Audi's current lineup fall into that category. And since there is no official spec that would have been broken, you can't return the units for not delivering what had been promised.


No.  Any decoder that does not decode VBR fails the compliance tests and is therefore officially broken.  The spec is clear on this point.  You can take it back and complain but they're not going to care.  People walk all over the MPEG specs everyday.

Quote from: rpp3po on
AFAIK the latter is no problem in AAC's (or the mp4 container format's) specification because there is an official definition instead of three different software developer created pseudo standards as in the case of MP3.


I've actually done a lot of work on audio decoders, so I can confidently say you have no idea what you're talking about here.  Bugs creep into decoders all the time because they're complicated not because people are somehow unable to find documentation about how they work (well aside from the MS formats).  MPEG ones especially encounter problems due to complexity.  If your decoder doesn't handle a codec feature its required to, its a bug.  This happens to MP3 decoders (though I've never seen it in practice since MP3 is so old and they're essentially all debugged by now) and to AAC decoders (my old iPod).

Quote from: rpp3po on
* And having to chose one of 15 predefined bitrates for each frame isn't optimal, either. That's why I called MP3 basically a CBR based design, where VBR operation is an extension (at least on a file level) and AAC a basically VBR based design where CBR and ABR operation are optional constraints.


And you were wrong.  Seriously, read the spec, or at least the HA wiki entry on the bit res in MP3.  You need to understand how frames work before trying to explain them to us.

(AAC) advantages & disadvantages

Reply #31 –
In case anyone is interested in some published "official" listening tests results on MP3 vs. AAC, I suggest reading this:

G. Soulodre, et. al, "Subjective Evaluation of State-of-the-Art 2-Channel Audio Codecs," 104th AES Convention, May 1998.

If you want to skip the text, take a look at Figure 3 on page 19.

Some background knowledge:

By the time of this test, the tested AAC encoder was about 2 years old. The MP3 encoder was almost 10 years old.
The bass clarinet, harpsichord, and pitch pipe items in this test reveal a weakness of MP3. Most likely, these items will not be transparent for MP3 at any bitrate up to and including 320 kbps. And most likely, they will be transparent for AAC at 320 kbps or even less. If anyone is interested, I can do a blind test of these three items (since I have access to them) at 320 kbps AAC vs. MP3.

Chris
If I don't reply to your reply, it means I agree with you.

(AAC) advantages & disadvantages

Reply #32 –
Quote from: C.R.Helmrich on
In case anyone is interested in some published "official" listening tests results on MP3 vs. AAC, I suggest reading this:

G. Soulodre, et. al, "Subjective Evaluation of State-of-the-Art 2-Channel Audio Codecs," 104th AES Convention, May 1998.


Thats so old its basically irrelevant.  None of the modern MP3 or AAC codecs even existed when it was written.

(AAC) advantages & disadvantages

Reply #33 –
Quote from: C.R.Helmrich on
If anyone is interested, I can do a blind test of these three items (since I have access to them) at 320 kbps AAC vs. MP3.

Chris


If you still have access to the lossless version, it would be great if you could share some <30sec samples. If your upload contingent is already full, I have some spare space available.

(AAC) advantages & disadvantages

Reply #34 –
Quote

What exactly is so irrelevant about it? The test compares Fraunhofer's new (at that time) AAC encoder against Fraunhofer's 10-year tuned reference MP3 encoder. Are you trying to say that the encoders used in that 1998 test basically "sucked" with regard to today's encoders? I think I clearly wrote that the test items for which MP3 did bad in this test are items which reveal a weakness of the MP3 codec structure (actually, they are also quite hard to encode transparently with AAC). That's still the case with today's codecs. I hope that, time permitting, I can support this with a blind test soon.

Quote from:  link=msg=637836 date=0
If you still have access to the lossless version, it would be great if you could share some <30sec samples. If your upload contingent is already full, I have some spare space available.

The bass clarinet and harpsichord items belong to the SQAM CD, which you can download here. Unfortunately, the pitch pipe is a copyrighted recording by Dolby, so I can't share it.

Chris
If I don't reply to your reply, it means I agree with you.

(AAC) advantages & disadvantages

Reply #36 –
Quote from: C.R.Helmrich on
The bass clarinet and harpsichord items belong to the SQAM CD, which you can download here.


Here is a ready-made set of above samples for LAME 3.98.2 MP3 and Quicktime 7.6 AAC at bitrates 128, 192, 320 (CBR) and V2, Q127 (VBR). V2 and Q127 both have an average target bitrate of ~192 kbit/s. The WAVs are also included. All other settings were left at their defaults.

28 files inside the archive, 10,2 MB: "rapidshare.com/files/238901987/ABX.zip" (prefix "http://", the board disallowed Rapidshare link provision).

Maybe somebody wants to add Nero encodes.

I'll try a round of ABX later this day.

(AAC) advantages & disadvantages

Reply #37 –
Quote from: rpp3po on
Maybe somebody wants to add Nero encodes.


Samples encoded by Nero AAC 1.3.3.0 at q 0.50, q 0.55, 128 + 192 + 320 CBR and 2 pass ABR.

Nero AAC encodes
"I never thought I'd see this much candy in one mission!"

(AAC) advantages & disadvantages

Reply #38 –
Important: do not ABX an encoded file vs. the original WAVE. Decode to WAVE and remove the codec delay at the beginning of the file. The waveforms need to be perfectly time-aligned, otherwise even transparent items may be ABX-able simply because when making loops, the start and end points of the waveforms will be slightly different.

For SQAM 17a and 40a, I uploaded decoded and delay-compensated LAME CBR encodes at 128 and 256 kbps in WAVE format here (ad-free). For SQAM 17, I removed 0.5 seconds of excess silence at the beginning of the file. Rpp3po, if possible, please upload this on your space. My download is only valid for two days.

For SQAM 40a, I could ABX LAME CBR at 256 kbps. I failed for 17a with LAME CBR 256 kbps (that already sounded relatively good at 128 kbps, actually). I will post my results for LAME 320 kbps here later.

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v0.9.6.7
2009/05/30 17:39:41

File A: C:\Users\Christian\Desktop\40_original.wav
File B: C:\Users\Christian\Desktop\40_lame256kb.wav

17:39:41 : Test started.
17:40:32 : 01/01  50.0%
17:41:03 : 02/02  25.0%
17:43:11 : 03/03  12.5%
17:43:48 : 04/04  6.3%
17:44:17 : 05/05  3.1%
17:44:55 : 06/06  1.6%
17:45:39 : 07/07  0.8%
17:46:17 : 08/08  0.4%
17:46:19 : Test finished.

 ---------- 
Total: 8/8 (0.4%)

Update: The 320 kbps LAME CBR decoded and delay-compensated WAVE files for the two above SQAM items are available here. Again I was able to ABX SQAM 40 at LAME 320 kbps. It is still not as "crispy" as the original.

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v0.9.6.7
2009/05/30 19:04:37

File A: C:\Users\Christian\Desktop\40_original.wav
File B: C:\Users\Christian\Desktop\40_lame320kb.wav

19:04:37 : Test started.
19:05:32 : 01/01  50.0%
19:05:49 : 02/02  25.0%
19:06:23 : 03/03  12.5%
19:06:58 : 04/04  6.3%
19:07:44 : 05/05  3.1%
19:08:05 : 06/06  1.6%
19:08:41 : 07/07  0.8%
19:09:10 : 08/08  0.4%
19:09:12 : Test finished.

 ---------- 
Total: 8/8 (0.4%)

Chris
If I don't reply to your reply, it means I agree with you.

(AAC) advantages & disadvantages

Reply #39 –
Foobar is able interpret gapless information from lossy files. Doesn't its ABX component do the time synching implicitly, when it converts the lossy files to WAV prior to a round of testing?

(AAC) advantages & disadvantages

Reply #40 –
Quote from: rpp3po on
Foobar is able interpret gapless information from lossy files. Doesn't its ABX component do the time synching implicitly, when it converts the lossy files to WAV prior to a round of testing?

Yes. If someone manually tweaks foobar's decodes afterwards he will only break time sync.

(AAC) advantages & disadvantages

Reply #41 –
Quote from: rpp3po on
Foobar is able interpret gapless information from lossy files. Doesn't its ABX component do the time synching implicitly, when it converts the lossy files to WAV prior to a round of testing?

Maybe for MP3, but not for the QuickTime MP4 files in your upload. When I ABXed that at 256 kbps, I could hear a difference due to delay. But again, this might not be the case for nero MP4 files (nero decodes seem to be delayless). But to avoid the possibility of delay in the first place, one should decode to the same format as the reference file and do "manual" delay compensation.

Chris
If I don't reply to your reply, it means I agree with you.

(AAC) advantages & disadvantages

Reply #42 –
Ok, sounds sensible to use pre-converted wavs, then. Have you removed the delays by hand or do you know a tool to do this automatically?

(AAC) advantages & disadvantages

Reply #43 –
rpp3po, I did this by hand. Don't know of any publicly available tool. Opened the LAME MP3s in Audition, removed the first x samples (where x = 2257 for LAME and 1088 for your QuickTime files), saved as 16-bit WAVE file. To open MP4 files in Audition, I used the FAAC/FAAD plug-in.

Chris
If I don't reply to your reply, it means I agree with you.

(AAC) advantages & disadvantages

Reply #44 –
So, here are my results for SQAM 40 with nero 192 kbps CBR. As you can see, I got the first three right, but it went downhill from there, so ultimately, I failed. To eliminate fatigue as the reason (it's midnight in Germany), I'll do this again with fresh ears tomorrow. Anyway, nero 192 kbps to my ears sound at least as good as LAME (or MP3 in general) at 320 kbps.

Code: [Select]
foo_abx 1.3.4 report
foobar2000 v0.9.6.7
2009/05/30 23:59:25

File A: C:\Users\Christian\Desktop\40_original.wav
File B: C:\Users\Christian\Desktop\40_nero192kb.wav

23:59:25 : Test started.
00:00:06 : 01/01  50.0%
00:00:41 : 02/02  25.0%
00:00:59 : 03/03  12.5%
00:01:28 : 03/04  31.3%
00:02:06 : 04/05  18.8%
00:03:04 : 04/06  34.4%
00:03:34 : 04/07  50.0%
00:06:35 : 05/08  36.3%
00:07:52 : 05/09  50.0%
00:08:54 : 06/10  37.7%
00:09:00 : Test finished.

 ---------- 
Total: 6/10 (37.7%)

Chris
If I don't reply to your reply, it means I agree with you.

(AAC) advantages & disadvantages

Reply #45 –
Here are my first preliminary results. For now I have skipped sample 17a, because I could not identify any characteristic artifacts in the first run. I continued with sample 40a:

LAME CBR 128:

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.2
2009/05/30 23:31:11

File A: C:\Dokumente und Einstellungen\Administrator\Desktop\ABX\LCBR128-tec_sqam_40a_bwf_tcm6-12548.mp3
File B: C:\Dokumente und Einstellungen\Administrator\Desktop\ABX\tec_sqam_40a_bwf_tcm6-12548.wav

23:31:11 : Test started.
23:32:53 : 01/01  50.0%
23:33:01 : 02/02  25.0%
23:33:08 : 03/03  12.5%
23:33:18 : 04/04  6.3%
23:33:29 : 05/05  3.1%
23:33:41 : 06/06  1.6%
23:33:50 : 07/07  0.8%
23:33:55 : Test finished.

 ---------- 
Total: 7/7 (0.8%)

Notes: Laughably easy! Background totally wobbly, like a chorus effect or early internet radio. If I hadn't known it's 128kbit/s straight from a WAV I would have put money on that it is a transcoding.

LAME V2:

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.2
2009/05/30 23:37:37

File A: C:\Dokumente und Einstellungen\Administrator\Desktop\ABX\LV2-tec_sqam_40a_bwf_tcm6-12548.mp3
File B: C:\Dokumente und Einstellungen\Administrator\Desktop\ABX\tec_sqam_40a_bwf_tcm6-12548.wav

23:37:37 : Test started.
23:38:51 : 01/01  50.0%
23:39:44 : 02/02  25.0%
23:41:28 : 03/03  12.5%
23:41:48 : 04/04  6.3%
23:42:37 : 05/05  3.1%
23:43:02 : 06/06  1.6%
23:43:18 : 07/07  0.8%
23:43:48 : 08/08  0.4%
23:43:53 : Test finished.

 ---------- 
Total: 8/8 (0.4%)

Notes: Already harder. Focused onto range 0:00.4-0:01.6. Original is identifiable as having a more defined 'punch' at the first low note.

LAME CBR 320:

Code: [Select]
foo_abx 1.3.3 report
foobar2000 v0.9.6.2
2009/05/30 23:44:32

File A: C:\Dokumente und Einstellungen\Administrator\Desktop\ABX\LCBR320-tec_sqam_40a_bwf_tcm6-12548.mp3
File B: C:\Dokumente und Einstellungen\Administrator\Desktop\ABX\tec_sqam_40a_bwf_tcm6-12548.wav

23:44:32 : Test started.
23:45:13 : 00/01  100.0%
23:46:02 : 01/02  75.0%
23:46:25 : 02/03  50.0%
23:46:41 : 03/04  31.3%
23:48:17 : 04/05  18.8%
23:48:31 : 04/06  34.4%
23:51:09 : 05/07  22.7%
23:51:48 : 06/08  14.5%
23:52:08 : 07/09  9.0%
23:53:22 : 07/10  17.2%
23:59:13 : 08/11  11.3%
00:00:04 : 09/12  7.3%
00:06:46 : 10/13  4.6%
00:09:01 : 11/14  2.9%
00:11:50 : 12/15  1.8%
00:12:24 : 12/16  3.8%
00:15:55 : 13/17  2.5%
00:16:29 : 14/18  1.5%
00:20:42 : 15/19  1.0%
00:23:26 : 15/20  2.1%
00:24:24 : 16/21  1.3%
00:25:15 : 17/22  0.8%
00:25:18 : Test finished.

 ---------- 
Total: 17/22 (0.8%)

Notes: Terribly hard, but beatable. Again focused onto 0:00.4-0:01.6. First try already false. One sample seems to have time smearing in its highs compared against the other. Several mistakes, pauses needed due to fatigue. Focussing your ear strictly at the first punch, without letting it get 'carried away' by the sound that instantly follows, reveals ringing highs.


I really like the samples. These are excellent recordings. Thanks for the reference!

40a is a natural instrument, not a synthetic test sample and I'm not impressed with LAME's performance! The VBR algorithm also does not get the gravity of the situation and fails to scale up higher than 196 kbit/s average.

Let's see how the AAC contenders compare to that. CBR 320 was quite tiring. I may try it later tonight or tomorrow.

Quote from: C.R.Helmrich on
For SQAM 17a and 40a, I uploaded decoded and delay-compensated LAME CBR encodes at 128 and 256 kbps in WAVE format here (ad-free). For SQAM 17, I removed 0.5 seconds of excess silence at the beginning of the file. Rpp3po, if possible, please upload this on your space. My download is only valid for two days.

Here you go: rapidshare.com/files/239055214/sqam_17_40_lame_cbr_128_256.zip

(AAC) advantages & disadvantages

Reply #46 –
Not to be a pedantic prick, as I not questioning what is claimed to be heard, but whatever happened to the idea of doing a fixed number of trials?  This pattern of constantly varying the number of ABX trials caries the taint of shopping for answers, be it justified or not.
Creature of habit.

(AAC) advantages & disadvantages

Reply #47 –
Quote from: C.R.Helmrich on
Quote

What exactly is so irrelevant about it?


The results do not reflect tests of anything relevant. 

Quote from: C.R.Helmrich on
Are you trying to say that the encoders used in that 1998 test basically "sucked" with regard to today's encoders?


I'm saying they're not today's encoders.

Quote from: C.R.Helmrich on
I think I clearly wrote that the test items for which MP3 did bad in this test are items which reveal a weakness of the MP3 codec structure (actually, they are also quite hard to encode transparently with AAC). That's still the case with today's codecs. I hope that, time permitting, I can support this with a blind test soon.


What you wrote clearly was your hypothesis, which you then proposed to test.  Thats very good.  But you've already explained that those results don't support or even test your hypothesis, so why did you recommend I read them exactly?

(AAC) advantages & disadvantages

Reply #48 –
Quote from: Soap on
Not to be a pedantic prick, as I not questioning what is claimed to be heard, but whatever happened to the idea of doing a fixed number of trials?  This pattern of constantly varying the number of ABX trials caries the taint of shopping for answers, be it justified or not.


I don't agree. Sometimes, as in the first two cases, the difference is so laughably easy, that it would be a boring waste of time to continue testing far below the 1% probability of guessing threshold. I'm doing this in my free time. So I wouldn't want a minimum number of trials above 8. It's also a question of stamina, you want to save this for higher bitrates.

But you can't take 8 as a fixed number, because that's not suited for harder samples at the edge of audibility. For that you need more. In my opinion as long as a subject delivers <1% results and you get to see the whole track record, that's fine. It will show wether someone just oscillated around 50/50 and somewhen hit stop after 50 tries or if there was a significant trend. In my 3rd run I don't see any indication that this looks like the results of random oscillation.

(AAC) advantages & disadvantages

Reply #49 –
Quote from: rpp3po on
I don't agree. <snip>
But you can't take 8 as a fixed number, because that's not suited for harder samples at the edge of audibility. For that you need more. In my opinion as long as a subject delivers <1% results and you get to see the whole track record, that's fine. It will show wether someone just oscillated around 50/50 and somewhen hit stop after 50 tries or if there was a significant trend. In my 3rd run I don't see any indication that this looks like the results of random oscillation.

According to what you've stated, you're watching the results as they come in, which is also invalid procedure.
Quote
3. The p values given in the table linked above are valid only if the two following conditions are fulfilled :
-The listener must not know his results before the end of the test, exept if the number of trials is decided before the test.
...otherwise, the listener would just have to look at his score after every answer, and decide to stop the test when, by chance, the p value goes low enough for him.
-The test is run for the first time. And if it is not the case, all previous results must be summed up in order to get the result.


I wasn't calling anyone out by name - for you are not the only one, rpp3po, who has been doing this as of late.
Creature of habit.