[MPlayer-dev-eng] Resampling

Arpi arpi at thot.banki.hu
Mon Oct 29 15:28:01 CET 2001


Hi,

> I get to flame my self. I found a bug (kind of) in the re-sampling
> example I submitted. The code can only do up-sampling - I have started
> to work on a version that can do both up and down sampling. I will
> publish it when it works, so I guess the ones stuck with old sound
> cards that only can do 44.1kHz will have to wait... sorry. The bug is
> mathematical- the code actually does a pitch shift when it does
> up-sampling (it also creates a lot of aliasing), kind of funny I
> guess.

yes, i've also seen a strange thing, possibel cause of this.
you only increment source data pointer by 1, but for >2x downsampling
it requires increment by 2 or more.

> I have also started to look at different methods to design the filter
> parameters. If you have been playing around with my example you might
> have noticed that it reduces the amplitude by 2. This is necessary in
> order to deal with impulsive sounds (they cause overshoot). I will try
> to design filters that don't have this drawback. Fixing this will also
> reduce the quantisation error caused by the fix-point implementation,
> and hence the noise in the filters. I will also try to optimise the
> filter length such that the output frequency comes closer to it's
> optimum for the most common input-output frequency pairs.

ok. i think it can be easily converted to MMX, so speed won't be problem.


A'rpi / Astral & ESP-team

--
mailto:arpi at thot.banki.hu
http://esp-team.scene.hu



More information about the MPlayer-dev-eng mailing list