[FFmpeg-devel] [PATCH] QCELP decoder

[Kenan Gillet]

On Oct 30, 2008, at 11:49 AM, Michael Niedermayer wrote:

> On Tue, Oct 28, 2008 at 04:26:36PM -0700, Kenan Gillet wrote:
>> On Oct 24, 2008, at 6:00 PM, Michael Niedermayer wrote:
>>
>>> On Fri, Oct 17, 2008 at 09:46:31AM -0700, Kenan Gillet wrote:
> [...]
>>> [...]
>>>> Index: libavcodec/qcelp_lsp.c
>>>> ===================================================================
>>>> --- libavcodec/qcelp_lsp.c	(revision 0)
>>>> +++ libavcodec/qcelp_lsp.c	(revision 0)
>>>> [...]
>>>
>>> we have some float lsp code in ffmpeg already (in wma&vorbis), i  
>>> would
>>> prefer to avoid adding more duplicated code.
>>> If there are problems with merging the different implementations
>>> then this
>>> should be discussed, maybe there is a problem that would leave us
>>> no choice than adding this but until such problem is found this is
>>> not good.
>>
>> After looking at the vorbis code, and the vorbis specification, at
>> http://www.xiph.org/vorbis/doc/Vorbis_I_spec.html#vorbis-spec-floor0-synth
>> and comparing it to the QCELP specification (TIA/EIA/IS-733  
>> 2.4.3.3.5),
>> I don't see any way of merging the vorbis/wma code with the qcelp  
>> code.
>> But again I am no audio guru, and any help would be appreciated to
>> confirm.
>>
>> I've reworked the qcelp_lsp.c though to extract a more general
>> convolve function.
>
> Well at least this should be optimized so it does not multiply by 1  
> and
> does not do calculations twice due to symmetries.

should it still be in its own file ?

Kenan