[FFmpeg-devel] [PATCH] AAC decoder round 8
Mon Aug 18 16:20:29 CEST 2008
On Mon, Aug 18, 2008 at 02:03:53PM +0100, Robert Swain wrote:
> 2008/8/18 Michael Niedermayer <michaelni at gmx.at>:
> > On Mon, Aug 18, 2008 at 01:47:42PM +0100, Robert Swain wrote:
> >> 2008/8/18 Robert Swain <robert.swain at gmail.com>:
> >> > 2008/8/18 Robert Swain <robert.swain at gmail.com>:
> >> >> 2008/8/18 Robert Swain <robert.swain at gmail.com>:
> > [...]
> >> + for (m = 0; m < size; m++, start += inc)
> >> + for (i = 1; i <= FFMIN(m, order); i++)
> >> + coef[start] -= coef[start - i] * lpc[i];
> >> }
> > are you sure this is correct for both values of inc ?
> > i mean one will have coef[start - i] unfiltered and one filtered as far as
> > i can see ...
> You're right and I thought this was strange, but it's the same
> behaviour as before from what I see and so in that sense it is as
> correct as it was before. The spec says:
> tns_ar_filter( spectrum, size, inc, lpc, order )
> - Simple all-pole filter of order "order" defined by
> y(n) = x(n) - lpc*y(n-1) - ... - lpc[order]*y(n-order)
> - The state variables of the filter are initialized to zero every time
> - The output data is written over the input data ("in-place operation")
> - An input vector of "size" samples is processed and the index increment
> to the next data sample is given by "inc"
> I think the third point suggests that it's the intended approach.
but the first suggests it is not
isnt there some reference implementation or something around?
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
it is not once nor twice but times without number that the same ideas make
their appearance in the world. -- Aristotle
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 189 bytes
Desc: Digital signature
More information about the ffmpeg-devel