Hi everybody,
I just completed a kind of patchwork of some Conformance
input documents to try to give a picture of SA Conformance in MPEG.
It is downloadable as a pdf file at the SAOG Forum pages.
I am not so pleased to say that just in assembling this file I already
discovered two errors, for which I'll try to ask a corrigendum as
soon as possible (i.e., in two weeks).
The first one is not so bad; the table of the profiles does not include
ttsi (text-to-speech interface) in the Synthesis profile, while it does,
correctly, in the main Audio standard.
The second one is much worse, in the sense that it concerns the SA
filters. You find in the document:
"Transition band is the 15% of Fn" (cutoff frequency). In my intention
this should have been the inverse, Fn is the 15% of Fs (stop band)
that corresponds roughly to a slope of 20 dB/octave.
It is incredible how an editing (hectic, I'd say) of different people
can be so independent from the technical content they are editing :-((
There are references to Appendix B. It is essentially contained in the
third paper again at the SAOG Forum page.
At 07:56 PM 1/20/2000, John Lazzaro wrote:
>-- Since the standard (and by your comment, I assume the conformance
>tests as well) leaves the slope to the implementor, a sound designer
>using filters in a way where the filter slope is sonically import
>needs to either craft their own filter from scratch (either using
>delay() opcodes or the biquad() or the fir/iir/firt/iirt filters),
>or use instrs or opcodes written by others that are normative with respect
> to cutoff slope.
I have included in the file, for convenience, also the Systems part about
SR conversion. To answer the above point: in MPEG we had lot of discussions
about these conformance points. For SR conversion it was hard to convince
vrml-oriented guys to delete the "nice" 12 dB they put in there for SNR; definitely
SA lives in a standard where signal processing is well-made and important.
During the last phone calls I had about that, it was pointed out, particularly
by the Audio chairman, that it would have not been acceptable to have a
32-bit standard with byte-exact conformance points and with "open to
implementers" points, at least where objective evaluations could be made.
Only one of these "faint points" still remains in my opinion, i.e. non linear
interpolation test.
Under this light, it is true that a particular filter can be crafted
when needed, but it is not too much in the MPEG philosophy to have the
"acceptable" as standard and the "better" as option. In this case we leave
the acceptable as option (i.e. you can design by fir/iir your weak slope if
it is enough for your purposes).
This said, these constraints are not in the main standard, but in the Conformance
standard. This means that saolc and sfront are perfectly normative to MPEG-4
SA, simply they should be refined a little to be normative to MPEG-4 in general.
I hope this help.
Best regards,
Giorgio
__________________________________________________________________
Giorgio ZOIA
Integrated Systems Laboratory - DE/LSI - EPFL
CH-1015 Lausanne - SWITZERLAND
Phone: + 41 21 693 69 79 E-mail: Giorgio.Zoia@epfl.ch
Fax: +41 21 693 46 63
__________________________________________________________________
This archive was generated by hypermail 2b29 : Mon Jan 28 2002 - 12:03:52 EST