Aac+

Topic: Aac+ (Read 8010 times) previous topic - next topic

0 Members and 1 Guest are viewing this topic.

Aac+

2002-10-21 13:58:48

hi people

in a closed thread i saw a guy talking about AAC+. What is it exactly ? Was he talking about the next evolution of AAC ?

If so does that mean AAC to include SBR technique in a near future ?

Aac+

Reply #1 – 2002-10-21 15:18:20

AAC+ is like mp3pro but just with aac instead of mp3

mp3 + sbc = mp3pro
AAC + sbc = AAC+

SBC is a tecnology where you by analyse of the mainstream (AAC or mp3) plus som small aditionel info tries to recreate the higher freqencies.
as this recreation is not perfect, the technology is used for low bitrates where lowpass filters are really low anyway.

you will need a sbc compatibel player otherwires only the main data will show
that
mp3pro/aac+ can be playded byan mp3/aac only player. but you will miss the SBC data (higher freqencies.)

hope this could clear some up

Aac+

Reply #2 – 2002-10-21 18:35:04

Quote

mp3 + sbc = mp3pro
AAC + sbc = AAC+

It's SBR (spectral band replication), not SBC.

Aac+

Reply #3 – 2002-10-21 22:36:01

Quote

in a closed thread i saw a guy talking about AAC+. What is it exactly ? Was he talking about the next evolution of AAC ?

If so does that mean AAC to include SBR technique in a near future ?

Yes, it is already in use by two different digital radio concepts, XM Satellite Radio and Digital Radio Mondiale and will probably be standardized by the ISO/IEC in MPEG-4 version 3 in May 2003 according to their time schedule. See also the Wiki at http://www.audiocoding.com/wiki/ about "AAC" and "SBR".

Aac+

Reply #4 – 2002-10-22 11:12:13

Yes, AAC will soon have SBR as the official part of the standard.

I've been able to listen to various samples at 32, 40, 48 and 64 kbps and do side-by-side compare of AAC, AAC+, MP3 and WMA - and I must say that AAC+ is really superior to any other format at such low bit rates.

At 64 kbps, AAC+ comes very close to MP3 at 128 kbps - maybe with slightly higher pre-echo, but other things are very well. At 32 kbps (Stereo!!), AAC+ performs like MP3 @64-72 kbps, with frequency response of 15 kHz.

Becoming a part of the official MPEG-4 standard means that AAC+ player will begin to pop up in 2003.

Aac+

Reply #5 – 2002-10-22 11:46:14

Quote

I've been able to listen to various samples at 32, 40, 48 and 64 kbps and do side-by-side compare of AAC, AAC+, MP3 and WMA - and I must say that AAC+ is really superior to any other format at such low bit rates.

What about OGG?.

Aac+

Reply #6 – 2002-10-22 12:28:34

Well, at 64 kbps I tend to prefer AAC+, sounds more sharp and the stereo image is much better

Aac+

Reply #7 – 2002-10-22 13:31:59

Quote

Quote
mp3 + sbc = mp3pro
AAC + sbc = AAC+

It's SBR (spectral band replication), not SBC.

Damn I must have done to much DivX encoding these day

SBC= Single Bit Control = a way to optimize the use of bits with divx 3.11.

To Ivan
When are YOUR AAC+ comming ??

Damn you must be a hard pressed man. "make this" "do that" "where are XXX comming" "where are u updating psytell encoder " etc. etc.

Aac+

Reply #8 – 2002-10-22 13:34:19

Will the future AAC format (or specifically PsyTEL AAC) have the ability to switch between "normal" AAC and AAC+ on the fly when encoding/decoding depending on the complexity of the source material...?

Aac+

Reply #9 – 2002-10-22 13:41:30

Quote

Will the future AAC format (or specifically PsyTEL AAC) have the ability to switch between "normal" AAC and AAC+ on the fly when encoding/decoding depending on the complexity of the source material...?

No, that's impossible, there would be frames with different samplerates (that could be worked around, but it would break backwards compatibilty with a normal AAC decoder (because then that would have different samplerates on frames)). And SBR introduces an extra delay, which makes it impossible too.

Menno

Aac+

Reply #10 – 2002-10-22 13:47:45

Well, I don't think switching SBR on and off per frame basis is a good idea (and quite hard to implement)- anyway SBR in AAC is very optimized, and decoding is at least 20x real-time on the P4 computer, so no worries about comlpexity

I am personally impressed with the quality at 64 kbps and less - with this feature, MPEG-4 will be certainly the best audio platform available today for all bit rates (from 16 to >512 kbps).

Aac+

Reply #11 – 2002-10-22 16:22:16

Is the AAC + SBR decoder easily implemented on a DSP / microProcessor that does not have a floating point unit?

wkw

Aac+

Reply #12 – 2002-10-22 19:50:13

Quote

IVAN:
At 64 kbps, AAC+ comes very close to MP3 at 128 kbps - maybe with slightly higher pre-echo, but other things are very well. At 32 kbps (Stereo!!), AAC+ performs like MP3 @64-72 kbps, with frequency response of 15 kHz.
Quote

time for some (premature) technical questions:

1.what will be the lowest bit-rate aac+ will encode to?if i'm correct,the lowest
SBR-achieved bitrate now is mp3pro @ 24 kbps.

2.will aac+ offer any low-complexity stereo support such as mp3PRO?
(..somewhat naive question, I know. I expect "of course it will,you dumbass'', but I want to confirm this.)

3.tell us more tell us more

actually,I considered aac+ to be far more delayed than a couple of months,
and I was surprised by all those good news.

Aac+

Reply #13 – 2002-10-22 23:51:27

Quote

1. what will be the lowest bit-rate aac+ will encode to? if i'm correct,the lowest SBR-achieved bitrate now is mp3pro @ 24 kbps.

This depends on the specific AAC version that will be implemented into AAC+. As far as I know, it will be AAC LD (= Low Delay), part of MPEG-4 version 2. I'm not sure if this codec is able to use only 16 kbps for 2 channels like the normal AAC codec can, because it has to sacrifice some efficiency for its low delay on bi-directional transmission lines. You can find more about the differences between AAC LC (= Low Complexity) and LD (and more on "beer and audiocoding") in a PDF file from Telos - Broadcast Studio Equipment that is linked on the following site (3.5 MB because of some graphic drawings):

http://www.telos-systems.com/?/techtalk/aa...r_2/default.htm

Quote

2. will aac+ offer any low-complexity stereo support such as mp3PRO?
(..somewhat naive question, I know. I expect "of course it will,you dumbass'', but I want to confirm this.)

Probably, because Coding Technologies (the developing company) also offers to use intensity stereo for mp3PRO encodings.

Here is a guide for the complexity of an AAC+ system, related to the question about hardware requirements of AAC+:

Table: Complexity comparison between AAC and AAC+SBR

AAC@24kHz stereo | SBR@48kHz stereo | AAC@24kHz + SBR@48kHz stereo | AAC@48kHz stereo:

MIPS: 10 17 27 20
RAM (24 bit words): 10k 10k 20k 10k
ROM (24 bit words): 7k 3k 10k 7k

Aac+

Reply #14 – 2002-10-23 08:44:42

AAC+ could theoretically use both AAC and AAC LD as the "core" bitstream. But for the consumer and high-quality applications AAC is the default. Only place where LD would be a better solution is the two-way hardware codec for live conversation.

At 44.1 kHz stereo, minimum (meaningful) bit rate for AAC+ is 32 kbps - can be lower for other sampling rates.

Quote

Is the AAC + SBR decoder easily implemented on a DSP / microProcessor that does not have a floating point unit?

I would have to look at the specs for this - I suppose it could be implemented much easier than AAC encoder ;-)

Quote

2.will aac+ offer any low-complexity stereo support such as mp3PRO?
(..somewhat naive question, I know. I expect "of course it will,you dumbass'', but I want to confirm this.)

SBR extension does not care if the baseline stream has intensity stereo, PNS or any other tool - I am personally against Intensity Stereo, and I also think that PNS is useless in combination with SBR.

Also, SBR decoding could be done in quite different ways, with various complexity levels.

Quote

3.tell us more tell us more

Unfortunately, I can't Except that I am impressed with the quality so far It is still really too early to talk about AAC+ as a consumer encoder - other than technical things.

Aac+

Reply #15 – 2002-10-23 10:37:57

I don't think it is too difficult to write a fixed point SBR implementation, from what I know of it now.

Menno

Aac+

Reply #16 – 2002-10-23 14:26:17

Ivan,thanks for your quick reply.In you we trust!
Hans, pretty good pdf!

Aac+

Reply #17 – 2002-10-23 15:27:26

Quote

I don't think it is too difficult to write a fixed point SBR implementation, from what I know of it now.

Indeed, as for as I know, Coding Technologies already has a fixed-point SBR implementation, & it think it is already used in Digital Radio Mondiale receptors...

I also believe that in MPEG-4 the AAC+SBR solution will be based on AAC LC, not AAC LD... (but as this MPEG-4 standard is not finalized, we'll see).

BTW, has anybody got some news from the present MPEG meeting in Shangai? I think the audio group is working on SBR...

Aac+

Reply #18 – 2002-10-23 15:48:09

Yes, there is an ad-hoc group for SBR (MPEG-4 bandwidth extensions, or BWE), and as far as I know at this point the bitstream format is being finalized.

SBR could be used on LD AAC, too - but except telco applications I can't see any benefit in consumer applications (like music archivers, etc..)

Aac+

Reply #19 – 2002-10-24 07:05:14

Quote

AAC+ could theoretically use both AAC and AAC LD as the "core" bitstream. But for the consumer and high-quality applications AAC is the default. Only place where LD would be a better solution is the two-way hardware codec for live conversation.

(taken from "ISO/IEC - Workplan for MPEG-4 Audio Extension 1 Core Experiments" from March 2002, so it may be outdated already):
--------------------------------------------------------------------------------
3.2.2 Core Experiments
[...]
With respect to potential CEs targeting at low delay functionality, the following principles are to be followed:
* All developments must be extensions of the current RMx BWE tool.
* AAC-LD must be used as the base coder.
* The performance targets agreed upon for the BWE work item apply, that is the coder/BWE system must perform at least as good as the coder without BWE at a 25% higher bitrate. Due to the difference in efficiency between AAC and AAC-LD, this translates into a comparison between AAC-LD/BWE at 32 kbps/ch and AAC-LD at 40 kbps/ch.
* Proposals for LD strongly deviating from RMx need to be significantly better than LD proposals more closely related to RMx to justify their existence.
--------------------------------------------------------------------------------

So this could also mean that they only wanted to simplify the development of AAC+ with LD functionality and not limit the use of SBR to Low Delay implementations of AAC - which I would prefer too, of course...

Quote

At 44.1 kHz stereo, minimum (meaningful) bit rate for AAC+ is 32 kbps - can be lower for other sampling rates.

By the way, where would you state the minimum meaningful bitrate for PsyTEL AACEnc? I'm asking this, because I have spent the last few weeks trying to come close to the sound of the FhG AAC (evaluation build from August, 23th) that was used in the recent c't listening test at 64 kbps. In my mind (or ears) PsyTEL has to use some more bits than 64 kbps (15-20%) to compete with FhG, WMA9 or mp3PRO at that bitrate. So the best command line I could come up with was -radio -resample 32000, which resulted in 73 kbps with the c't reference.wav.

Quote

Quote

Is the AAC + SBR decoder easily implemented on a DSP / microProcessor that does not have a floating point unit?

I would have to look at the specs for this - I suppose it could be implemented much easier than AAC encoder

I should have cited this paragraph from the same document before (where I also found the table with the CPU MIPS):
--------------------------------------------------------------------------------
Decoder complexity of Coding Technologies RM0 proposal for bandwidth extension (SBR). The numbers given below are based on an actual 24 bit fix-point DSP implementation for MPEG-2 AAC LC and MPEG-2 AAC LC+SBR. They are given for information purpose and might differ for other implementations.

Quote

Quote

2.will aac+ offer any low-complexity stereo support such as mp3PRO?

SBR extension does not care if the baseline stream has intensity stereo, PNS or any other tool - I am personally against Intensity Stereo, and I also think that PNS is useless in combination with SBR.

Yes, I remember your rant in alt.music.mp3 about the different stereo methods in AAC. But of course I couldn't help to try out intensity stereo with your encoder in my listening tests and found absolutely no difference at these low bitrates (this could be related to the lack of a superb stereo image in the reference file). Does the -is switch work at all or is it disabled in the code? The display at least told me that I would be using intensity stereo then...

Quote

Quote

3.tell us more tell us more

Unfortunately, I can't Except that I am impressed with the quality so far It is still really too early to talk about AAC+ as a consumer encoder - other than technical things.

The best place to look for important new informations probably is the official website of the MPEG Audio Subgroup that also lists all recent press releases. So maybe there is already one from the last meeting on October, 21th in Shanghai:

http://www.tnt.uni-hannover.de/project/mpe...g/audio/public/

Aac+

Reply #20 – 2002-10-24 10:25:47

Regarding comparing FhG/Dolby AAC and PsyTEL 2.15 - psytel encoder v2.15 had serious CBR bug that affected very low bit rate coding, like that @64 kbps.

This bug has been corrected now, and new products (I can't speak about them, of course ;-) will have this improved AAC encoder inside.

Aac+

Reply #21 – 2002-10-24 12:11:52

Quote

Regarding comparing FhG/Dolby AAC and PsyTEL 2.15 - psytel encoder v2.15 had serious CBR bug that affected very low bit rate coding, like that @64 kbps.

This bug has been corrected now, and new products (I can't speak about them, of course ;-) will have this improved AAC encoder inside.

Very interesting, did this bug also affect VBR encodings like with -tape or -radio? Talking of bugs, did you perhaps read my other messages in this board, e.g. about the wrong frame count with resampling?

By the way, "...and the winner is..."

http://www.codingtechnologies.com/news/ass...ebutest_eng.htm

Mountain View, CA, October 9, 2002 - Coding Technologies, the industry's leading provider of audio compression technologies for the mobile, digital broadcasting and Internet markets worldwide, announced today that the European Broadcasting Union (EBU) Subjective Listening Tests on Low Bitrate Audio Codecs shows aacPlus [tm] from Coding Technologies to be the highest quality codec for streaming audio applications. Calling aacPlus "a clear winner," the report states that it is now possible to provide excellent quality stereo audio programming at 48Kbps. The EBU report also credits Coding Technologies' Spectral Band Replication (SBR) [tm] technology with the significant improvement in codec performance compared to the previous report completed in 2000.

This older report is available at EBU's website:

http://www.ebu.ch/trev_283-kozamernik.pdf

Aac+

Reply #22 – 2002-10-24 15:54:39

Quote

Very interesting, did this bug also affect VBR encodings like with -tape or -radio? Talking of bugs, did you perhaps read my other messages in this board, e.g. about the wrong frame count with resampling?

No, bug was related to short block inter-frame and inter-channel bit allocation. Because of it at critical bit rates there was a huge NMR disturbance over the time.

But, there were some other (tonality) bugs that affected VBR, too But only under some circumstances.

Resampling issue... yes - but this won't be a problem anymore since resampling will be done outside of the MPEG-4 codec

Regarding AAC+SBR - yes, at 64 kbps and below it has been regarded as the best codec by EBU. SBR will become a very important tool in the near future.

Aac+

Reply #23 – 2002-10-24 20:52:45

Quote

Quote
Very interesting, did this bug also affect VBR encodings like with -tape or -radio?

No, bug was related to short block inter-frame and inter-channel bit allocation. Because of it at critical bit rates there was a huge NMR disturbance over the time.

Meaning that the quantization noise would become too noticable in these cases? What I heard at 64 kbps and below with CBR was either a rather muffled sound with the standard cut-off at 10 kHz or too obvious "flanging" on sharp treble attacks with a raised cut-off, accompanied by a thinner "empty" harmonic content on samples that were supposed to sound "rich" (e.g. Kylie Minogue's harmonized voice).

Quote

But, there were some other (tonality) bugs that affected VBR, too But only under some circumstances.

These artifacts became less annoying with -qvbr 17 -resample 32000, which resulted in a 64 kbps VBR file, but still could not compete with mp3PRO that I would also describe as a little bit calm or "conservative" in the high-frequency range. I rated it third best in the c't listening test because of that tendency. Only with -radio -resample 32000 PsyTEL had a chance to be on a par with mp3PRO, but still not equal to FhG AAC or WMA9 (according only to my taste, of course).

Quote

Quote
Talking of bugs, did you perhaps read my other messages in this board, e.g. about the wrong frame count with resampling?

Resampling issue... yes - but this won't be a problem anymore since resampling will be done outside of the MPEG-4 codec

And then it will be able to use SSR (Scalable Sampling Rate) so that I could also try 28 kHz as a valid value? I'm asking this because I noticed that the harshness/flanging or "rectangle" treble sounds on cymbals and so forth become smoother, if I match cut-off and resample rate closer together than -c 13000 and -resample 32000.

By the way, this is also true for LAME that sounded better than the FhG MP3 in the test with --alt-preset 64 --lowpass 12 --resample 24, but not enough to overtake RealAudio8. When using a cut-off at 13 or 14 kHz and a resampling frequency of 32 kHz, everything becomes very "flangy" and artificial sounding with LAME.

This leads me to another question... Is it correct that AACEnc couples the cut-off frequency to most of its VBR settings, especially -qvbr with a step of 80 Hz upwards when increasing this quality-based VBR? The only way to use a self-defined cut-off with VBR is -vbrhi (and -vr), but this setting seems to be a bit buggy when combined with a base bitrate equal to (Winamp freezes while trying to play such a file) or lower than -br 32 (AACEnc crashes right at the start of the encoding).

And last but least: do LTP (-profile 2) and intensity stereo (-is) switches work at all? I could not hear any differences in the resulting sound when using them. Opposed to that Main (-profile 1) and PNS switches seem to work, but result in a much worser sound than without them.

Quote

Regarding AAC+SBR - yes, at 64 kbps and below it has been regarded as the best codec by EBU. SBR will become a very important tool in the near future.

That's what I think, too, and I'm looking forward to encode my demo tapes with AAC+ at 48 or 56 kbps for modem visitors of my homepage. But I'm pointing out this issue with FhG AAC now and again, because it also seems to be possible to sound quite good at 64 kbps without SBR. Have you heard something about a release date of this evaluation build, perhaps?

Sorry, this message was written in "verbose mode" again...

Aac+

Reply #24 – 2002-10-27 14:07:13

Quote

This older report is available at EBU's website:

http://www.ebu.ch/trev_283-kozamernik.pdf

By the way, these Technical Reviews from the European Broadcasting Union are an excellent source of publically available information, because they contain articles that also have been reprinted in publications like the Journal of the Audio Engineering Society and others which are not available on the internet. Another example for this is the report about CT-aacPlus from Martin Dietz and Stefan Meltzer from July 2002 (7 pages), providing a good overview on AAC+ and a short insight into the results of the latest 2002 EBU listening test (page 5) mentioned in the press release from Coding Technologies:

http://www.ebu.ch/trev_291-dietz.pdf

An index of all Technical Reviews of the EBU with search function can be found here:

http://www.ebu.ch/trev_frameset-index.html

Notice