[FFmpeg-devel] GSoC 2008 qualification task

[Kostya]

On Sun, Mar 30, 2008 at 01:01:53PM +0200, Tomasz ?ernicki wrote:
> Dear All,
> Currently I?m trying to figure out how does the transforms in Nellymoser Asao codec work. 
> Due to forthcoming deadline of GSoC 2008, I would like to know what are your expectations
> connected with "AAC-LC Encoder" task? 

My expectations are that I will do it :)

And more seriously, here is the list of requirements I think everybody will agree with.

General requirements:
 * encoder should be able to write data that AAC decoders can understand and play
 * sound should be recognizable and preferably not very distorted
 * encoder should perform fast (comparing with libfaac and other encoders)

Code requirements:
 * encoder should be clean and extensible (maybe somebody will want to add SBR or
   other HE-AAC features)
 * encoder parameters should be easy to manipulate

> What does exactly mean "flexible framework for a serious psychoacoustic implementation"?

Encoder should have some kind of interface to psychoacoustic models
implementations, so it can be used in this way (as I see it):

* some option(s) replated to psy model choosing and finetuning
* encoder init function calls ff_psy_init() and tells what psy model it wants,
  what kind of data to expect (bands, block size?) etc...
* encoding calls something like ff_psy_twiddle() to manipulate frequencies in
  desired way, then encodes modified data.

So it is not hardcorded into encoder ("let's throw away all high frequencies")
but can be tuned independently and reused in other encoders.

> What features would you like me to implement?
> 
> I would appreciate any help.
> 
> Greetings
> Tomasz ?ernicki