Well, let's get right into it.. Options

[YouriP]

I know, I know, not the greatest way to kick off a new forum, but I had to do something about the silence here.

[Dibrom]

Well that would be Psytel of course...

I hear alot about LQT but unfortunately its a pain in the ass to use IMO and its non ISO (the winamp plugin sucks ) not to mention Psytel offers quite a bit more functionality.

Just my 2 cents, though its nice to have a few choices

[YouriP]

I refrain from using anything that isn't CLI.

[JohnV]

Nice looking poll, except that Astrid/Quartex is not an AAC encoder. It was hacked PAC-encoder.

[Ivan Dimkovic]

Hehe... I have fresh Win32 build of ISO reference AAC code :)Anyone wants to try (hurt ears)?

[Dibrom]

Hehe.. actually I wouldn't mind trying that out, just to see how much things have progressed over the ISO code.

[Ivan Dimkovic]

Ok, but you are doing that at your own risk

Download the stuff from:

http://www.psytel-research.co.yu/temp/old_iso

This is complete build of FDIS (I think) reference code - this is October 99 source code, last MPEG4V1 reference code.

Unlike most popular 1998 code (used in homeboy, etc..) this one has block switching and other advanced stuff.. but nothing works well

[Nic]

So, Ivan :-) When can your adoring fans hope to get a peek at FastEnc ? Or at least some info on how the testing is going, speed of encoding, etc

Thought it would be a nice way to start off

Cheers,
-Nic

[Ivan Dimkovic]

Ok, Ok...

Fastenc (current build) is available for download from:

http://www.psytel-research.co.yu/temp

This is early early beta version! Many things do not work and it is not tuned yet. Speed is also not final, I expect at least 40-50% increase in the near future.

However, this version runs up to 3 times faster than AACEnc 1.2 and 3-5 times faster than FAAC, 2 times faster than FhG "slow" AAC and nearly 20 times faster than ISO reference AAC.

Happy testing

[YouriP]

Now that was unexpected. Hehe, it looks exactly like AACEnc.

I noticed there are no disable_ath and low_ath switches for FastEnc. Does this mean the ATH has not been implemented yet? Also, upon encoding, it prints out "MAIN: 0, LC: 1, SSR: 2, LTP: 3, LD: 4". Exactly what do you mean by this, as the numbers do not translate to the selectable profiles.

Regarding profiles; I was wondering why there is still a switch -lc, when it can be chosen with the -profile 1 setting. It's kind of confusing, and makes that people can get results they didn't expect (just like -nh overrides -adif, which I was unaware of at first, so I encoded all files without headers when I actually wanted ADIF headers - maybe a -header [0,1,2] setting would be more appropriate instead of seperate switches).

Also, are you still working on the alternative CBR mode? Because I noticed you have not included it in the current FastEnc release.

Speed is excellent though. At the very least 3 times times faster than AACEnc, even when using -production. Though I favor quality over speed, it is a nice added benefit.

[YouriP]

Also, could you drop support for the proprietary PsyTEL tags? I think it would be a much better idea to support ID3v1 tags (or ID3v2 if you need to have the tag at the beginning of the stream), since FAAD can handle these tags by nature, and the PsyTEL tags have already caused incompatibilities with FAAD in the past (resulting in the psystrip utility).

[Ivan Dimkovic]

Ok,

FastEnc is using alternative CBR mode That is the only CBR mode in fastenc...

FastEnc is also MPEG-2 AAC LC only, no other profiles are supported.

FastEnc is using low_ath by default.

Options included in latest aacenc v1.2 are not here because source code tree was already separated prior to inclusion of the new options.

Many other options don't work because they were removed, and frontend code is still not up-to-date

Regarding speed - it is fast enough, but here I have AAC encoder that runs 2 times faster than fastenc! So, further optimizations are possible.

And that 'LC:x MAIN:y' message was debug message - I said that it is early beta full of those debug msgs, etc...

[Nic]

Thanks very much for the preview, in your last post you stated you had an encoder which was twice as fast as FastEnc! Is this your own encoder? or some other institutes?

Thanx again....And keep us posted....(you know that any info is always more than appreciated) although,of course, feel free not to answer):
1) What is the difference between alt-cbr and your normal cbr?
2) Will the quality actually be better with FastEnc than ACCEnc or will the speed performance hurt the quality?
3) What is the main difference between the approach of fastenc & aacenc?
etc. etc. Please feel free to ignore the above because we all appreciate how busy you are (Actually I dont know how you produce at the speed you do! You've come a very long way, very quickly

Cheers,
-Nic

ps
Ill do a few tests with FastEnc, quality wise and post back....

[YouriP]

Can I add some more questions, if you have time?

FastEnc is using alternative CBR mode That is the only CBR mode in fastenc...

Will this mode replace the current CBR mode for AACEnc as well?

FastEnc is also MPEG-2 AAC LC only, no other profiles are supported.

Yup, I just noticed that. Do you have plans for supporting other profiles besides MPEG-2/4 LC for either FastEnc or AACEnc? The MPEG-4v2 spec defines some interesting new profiles, most notably MPEG-4 HQ.

Options included in latest aacenc v1.2 are not here because source code tree was already separated prior to inclusion of the new options.

Many other options don't work because they were removed, and frontend code is still not up-to-date

When you are going to update the frontend, will you drop the -lc option and replace -nh/-adif with -header [0,1,2]? Pretty please?

Btw, though you have not said anything about support for ID3 tags (not really priority, seeing as FAAD can add these as well), it does mean the PsyTEL tags will be removed later on, right?

Thanks again for the insight.

[Ivan Dimkovic]

Ok, some answers:


1. Details about how fastenc optimization was done are specified in: http://www.psytel-research.co.yu/papers/di090201.pdf

2. "alternative cbr mode" is using different bit-allocation strategy and different bit reservoir management than old CBR mode, resulting in "softer" artifacts and less annoying dropouts.

3. Fastenc is optimized for speed, not for maximum quality. However, I tried to keep quality as high as possible.

4. Improvements in fastenc won't be included in AACEnc - instead, FastEnc will be upgraded with some advanced tools from AACEnc - it is easier because fastenc changed too much to be integrated back in aacenc.

5. Frontend will be updated in the near future

6. The encoder I have that runs two times faster than FastEnc is not mine, of course

[Nic]

Thanx for the info (and sorry I should look at your site for info rather than bother you....my apolgies)

So, in a very general way (ignoring my lack of knowledge, hopefully) FastEnc could be said to of been optimised (apart from general optimisation) in the following ways:

(LC does not use prediction)

Greater Audio Bandwidth Reduction (also speeds Psychoacoustic)

MDCT instead of FFT (higher quality for noise/mask)
-tonality classfication using Spectral Flatness Measure (rather
than ISO suggestion)

Short or Long blocks determined early (rather than complex block size decision)

Quantizer Optimisation (Therefore the fewest number of loops needed to maintain highest quality)
-Limit to number of unsuccessful loops (instead of unlikely
search for better)

(Noise estimation & Spectrum Scaling optimised)

Huffman over estimation of bits rather than fully accurate (no more Greedy-merge search) (extra bits goto bit reservoir for future use)

As I said, I dont know where you get time for all this. The documentation was excellent

Cheers for the info (& get some sleep! Im sure you cannot rest

-Nic

ps
Can you tell us what the speedy encoder was? Or is that the same one you said you couldn't talk about in r3mix (i.e. Dolby ? )

[Ivan Dimkovic]

Yep - your post is accurately summing techniques used in speed optimization of the AAC encoder

There are much more small improvements in the code, but I find them too small to comment them. Now, the problem is that each change in some important module requires complete codec tune-up - in psychoacoustic codecs everything is based on some variable thresholds, and the best choice is always kept private

Ah, the speedy encoder - this encoder is NOT Dolby one (and I believe that Dolby encoders are actually based on FhG code). I have signed non-disclosure agreement with the company, and I can't tell anything more about the encoder. First impressions are - encoder is amazingly fast, but the quality is not perfect, but not bad when speed is taken into account.

[YouriP]

I wish you could sign an NDA with FhG for their encoder and then accidentally "leak" it to the public. Nah, I wouldn't wanna see any lawyers put your excellent work to waste. Anyways, thanks for the insight and the answers. I hope that when you drop AACEnc and integrate its features into FastEnc you will keep some options to improve the quality (basically everything where Nic said speed optimizations were possible) so that users will have a choice between quality or speed. In any event, you're doing fine work. Can't wait for the next beta.

[Nic]

It does sound (from the pdf) that once the features are added it shouldn't be too much of a pain for Ivan to remove the quality limitations to gain better quality. (then again I dont know how difficult it will be to put back the greedy-merge search

Cheers.
-Nic

[Ivan Dimkovic]

Greedy Merge won't be a problem... there will be problems with block switching algorithm, however...

[t_vitaly]

where i can dowload in_aac plugin wor winamp
with support tag editing?

[layer3maniac]

vitaly, you probably won't remember this but you hooked me up with an aac plugin a long time ago. Thanks. I have an aac plugin (recent faac beta) which DOES support id3 tag editing. If you still need it, e-mail me at and I'll send it right out to ya. Offer valid to anyone else interested too. :deal:

[layer3maniac]

While here in the aac forum, I just wanted to say:
Ivan, you're a GOD!!!

[Nic]

Ivan/Deity

I told you ages ago that AACENC crashed my system on exit -

It still does, but it does this (im pretty sure) because my system overheats!I do loads of realtime video capture (directly to MPEG-4) and play far too many games & nothing overheats my computer other than AACENC Its weird, I go check my temperature afterwards and its sky high (800Mhz Duron)

Anyway since I was in the forum, I thought id let you know.

Cheers,
-Nic

[t_vitaly]

Originally posted by layer3maniac
vitaly, you probably won't remember this but you hooked me up with an aac plugin a long time ago. 

But that plugin was not able to edit tags (I wonder why?)
Ivan sent me yesterday the new version of plugin.
It realy work!!! Thank you Ivan.