[FFmpeg-soc] [PATCH] AMR-WB Decoder

[Marcelo Galvão Póvoa]

2010/9/4 Marcelo Galvão Póvoa <marspeoplester at gmail.com>:
> On 4 September 2010 15:16, Vitor Sessak <vitor1001 at gmail.com> wrote:
>> On 09/04/2010 05:22 PM, Marcelo Galvão Póvoa wrote:
>>>
>>> On 4 September 2010 06:02, Vitor Sessak<vitor1001 at gmail.com>  wrote:
>>>
>>>>
>>>> On 09/03/2010 11:50 PM, Marcelo Galvão Póvoa wrote:
>>>>
>>>>>
>>>>> Also, whether with this patch or not I guess my output to the first
>>>>> sample above (attached) is not good compared to the reference decoder.
>>>>> The are some sharp peaks that can be seen with Audacity, one of them
>>>>> also with clipping. The stddev I've got seems ugly:
>>>>> stddev:  884.86 PSNR: 37.39 MAXDIFF:23298 bytes:   473600/   473600
>>>>>
>>>>>
>>>>
>>>> Indeed, the stddev is not particularly good, but don't trust only it to
>>>> do
>>>> comparisons. A very small phase shift can completely ruin the stddev.
>>>>
>>>> Do you see sharp peaks in your output that are not there in the reference
>>>> decoder? Or the sharp peaks shows up when you invert and superpose the
>>>> waves? Do these sharp peaks gives audible artifacts?
>>>>
>>>>
>>>
>>> Sorry, I've forgot to mention that I was comparing the reference
>>> decoder output waveform with mine. In general, they look mostly the
>>> same but there are some 3 or 4 points in which my decoder gives a
>>> short peak. I couldn't notice audible artifacts at these moments.
>>>
>>
>> It does looks like a bug :(
>>
>
> I've just tested dumping all the voice_factor() values from the
> reference decoder (I suspected this function was problematic) and the
> output difference was not too much:
>
> all_man.awb stddev:    0.38 PSNR:104.69 MAXDIFF:    4 bytes:   473600/   473600
>
> The sharp peaks and the clipping are still there, so I guess we can
> discard checking this function.
>

I've forgot to mention that the high-band part isn't guilty too.

Looking at the waveforms below it's strange because most of the time
they look quite much the same but only at these few parts they start
to diverge greatly. I'll try to investigate what happens just before
the clipping (starting at +- 9.650 seconds)

http://www.students.ic.unicamp.br/~ra082115/all_men_ref.wav

http://www.students.ic.unicamp.br/~ra082115/am_vfOK.wav

-- 
Marcelo