On Wed, Aug 18, 2010 at 04:35:34PM +0200, Stephan Assmus wrote: > Am 18.08.2010 15:02, schrieb pulkomandy: > >Hello, > >Some days ago I introduced a new resampler that uses interpolation in > >the media kit. I'm still not happy with it, so I'm going to write > >another resampler, this time using a Lanczos filter, which should give > >much better results. > > I would guess something is wrong with your filter, but some months > ago (years?) I also wrote an uncommited resampling version and was > also wondering why it didn't seem to have any affect on the quality. > I did not trust my implementation though, just curious that you also > seem to be unhappy with your resampling. In theory, it should > already have enough quality. Now that I know how to get the debug output, I can test it more easily. My current implementation does interpolate only in the range of samples given in a single call, samples at the begin and end of the generated output for each call of the Resample function are not interpolated, and may be completely wrong. Depending on the size of buffers passed to the function, this may have more or less audible consequences. I'm going to look at the debug output and try the proper Lanczos filtering, which removes much more noise. But in the case of upsampling interpolation should be good enough.