[FFmpeg-devel] Error in ATRAC1 decoder?

[Michael Niedermayer]

On Tue, Nov 17, 2015 at 04:08:29AM +0300, Даниил Чередник wrote:
> Thank you for answer.
> 
> If I understood available ATRAC1 docs, the purpose of this delay line is
> just to compensate delay of 1st QMF because for 1st and 2nd band we have
> two QMF but for 3rd band just one.
> 
> About test, yes for some reason this patch brakes it. I will try to find
> out. Sorry, I should have done it before.

np
if the changed / new version is more correct then please also provide
a new reference file for the fate samples with a new filename and
update the fate test so it uses the new file

thanks

> 
> I tried to decode "The Four Seasons, Concerto Alla Rustica in G - Winter-
> Largo.aea" sample, the results are:
> without patch: https://yadi.sk/i/PCFplrafkWEFC
> with patch: https://yadi.sk/i/iLUR188gkWEFn
> On zoomed spectrogram it is possible to find aliasing if decode without
> patch. I can share decoded wav file if needed.
> 
> 
> 
> On Mon, Nov 16, 2015 at 3:45 AM, Michael Niedermayer <michael at niedermayer.cc
> > wrote:
> 
> > On Sat, Nov 14, 2015 at 02:59:30AM +0300, Даниил Чередник wrote:
> > > Hello!
> > >
> > > I have noticed if decode
> > >
> > https://samples.ffmpeg.org/A-codecs/ATRAC1/Test%20tones%20disc%20-%20Chirp.aea
> > > file by ffmpeg we got aliasing near 11025Hz. Screenshots:
> > > https://yadi.sk/i/r-95jZkKkSnbu https://yadi.sk/i/PV92LNESkSnby
> > >
> > > I was tried to solve it:
> > >
> > > ATRAC1 - hybrid codec, has two stacked QMF and splits the signal into 3
> > > band before MDCT. Thereby we need to compensate delay of one QMF to
> > > achieve reconstruction.
> > > There is a delay line but it looks like delay for 23 sample is not
> > correct.
> > > I have done some experiments and got 39 should be right delay. Results of
> > > decoding with patch: https://yadi.sk/i/yooaIQrmkSncB
> > > https://yadi.sk/i/CUXsH7-CkSncK
> > >
> >
> > > To be honest I am not a prof in math and it would be great if someone
> > > recheck it from math perspective.
> >
> > i dont think math can help here
> > whatever a format requires that has to be done, this could even be
> > wrong from a math point of view of a ideal transform
> >
> >
> > >
> > > There is a patch in attach.
> >
> > the patch also affects the atrac1 fate test
> > have you checked if the new output for that test is better than the
> > old ?
> >
> > make V=2 fate-atrac1
> >
> > also have you looked at other files, is this improving all files ?
> > any that get worse ?
> >
> >
> > [...]
> >
> > --
> > Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
> >
> > Many that live deserve death. And some that die deserve life. Can you give
> > it to them? Then do not be too eager to deal out death in judgement. For
> > even the very wise cannot see all ends. -- Gandalf
> >
> > _______________________________________________
> > ffmpeg-devel mailing list
> > ffmpeg-devel at ffmpeg.org
> > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> >
> >
> 
> 
> -- 
> Daniil Cherednik
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

The greatest way to live with honor in this world is to be what we pretend
to be. -- Socrates
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20151118/306ff456/attachment.sig>