On Wed, 15 Dec 2021 01:18:54 +0100 Ichthyostega <p...@ichthyostega.de> wrote:
>Am 14.12.21 um 09:48 schrieb Kristian Amlie: >> But that is interesting. Probably it is *not* related to randomness then, and >> it indeed could be because of float precision building up. PADsynth is >> heavily reliant on the FFT, I wonder if different versions of fftw3 could >> produce the differences. > >My suspicion is rather that some part of the previous spectrum can >somehow leak through the "apply()" call. This suspicion is based on >my observations done while we had that temporary bug in the last >development cycle. But actually I could not nail the problem, >because all these spectra are just such a huge amount of data. >Even a rather small change in the phase of a single bin in the >input spectrum can cause a dramatically different apparent waveform. > >OTOH, I'd rather doubt that rounding errors in the FFT alone >could produce such a strong effect. The smallest unit in the >float mantissa is 2^-23 after all (-138 dBFS) > >-- Hermann If this is the case, I think the problem must be in actually setting parameters accurately. The wavetable is completely rebuilt in exactly the same way each time. To test for this I suppose it would be possible to fake a change of parameters (i.e. just clear 'applied') and compare two supposedly identical wavetables. -- Will J Godfrey https://willgodfrey.bandcamp.com/ http://yoshimi.github.io Say you have a poem and I have a tune. Exchange them and we can both have a poem, a tune, and a song. _______________________________________________ Yoshimi-devel mailing list Yoshimi-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/yoshimi-devel
