[FFmpeg-devel] [PATCH] QCELP postfilter

[Alexander Strange]

On Apr 19, 2010, at 7:09 PM, Ronald S. Bultje wrote:

> Hi,
> 
> On Sat, Apr 17, 2010 at 9:19 PM, Ronald S. Bultje <rsbultje at gmail.com> wrote:
>> On Thu, Apr 15, 2010 at 12:29 PM, Ronald S. Bultje <rsbultje at gmail.com> wrote:
>>> On Tue, Apr 13, 2010 at 2:31 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
>>>>> Maybe I should try Quicktime Pro instead?
>>>> 
>>>> a sample from another encoder is a good idea too
>>> 
>>> OK, same place, resampled* is the input WAV (from the FLAC, resampled
>>> using ffmpeg) and the output .3g2 file from Quicktime, and the decoder
>>> output from Quicktime.
>>> 
>>> Then r1-3.wav is the blind files with no, my or reference postfilter.
>>> 
>>> http://people.gnome.org/~rbultje/ff/
>> 
>> Ping, for any other audiophile (particularly Michael, although he's
>> more an any-media-phile), I'm hoping the result of this blind test
>> will be sufficient to commit this patch version. It's also attached so
>> you can look at it.
> 
> Michael, that was sort of for you, ping again? :-). If you find no
> obvious difference between r1-3.wav, I'd like to apply this. If you
> do, I'd like to know also, so we can decide which is better, I can
> tell you if that was good or not for me, and then maybe apply.

A = mplayer -endpos 3 r1.wav
B = mplayer -endpos 3 r2.wav

  votes: ABABABBBBABBBABAAABB
 actual: BBBBBABBABBBAAAABAAA
correct: 9/20 (45%)

probability (p) of result being the same as random guesses = .7480806

A = mplayer -endpos 3 r1.wav
B = mplayer -endpos 3 r3.wav

  votes: BABABBBBABBAABBBA
 actual: AAABBABABBBAABABAABB
correct: 10/17 (59%)

probability (p) of result being the same as random guesses = .3144816

In a noisy environment on iPod earbuds so not the best conditions, but they still all sound the same to me.