From: jy <jy...@gm...> - 2011-01-30 14:03:59
|
Hi Cal Well I'm not sure this could be an issue for a feast reasons : 1 - both double and float types have enough resolution to describe fourier coefficients for our tasks, meaning that switching these from double to float might lead to a detuning but it would be so small that we probably wouldn't be able to hear it. For yoshimi, I guess a precision to the third to fifth decimal is sufficient. 2 - if switching from double to float lead to such huge differences, this would be a major issue for libfft, imho. And there could indeed be a bug in libfft. Unlikely... 3 - last but not least, unless there's a build flag I missed somewhere allowing to forbid the use of floats for these operations, we would all be affected. But hey, I can be wrong! Le 30 janv. 2011 01:40, "cal" <ca...@gr...> a écrit : > I still don't claim to understand anything much about fast fourier transforms, > but this thought keeps coming to mind. In original zyn and yoshi 0.058.1, the > basic buffer data type for the fft processing was double. Beyond 0.058.1, I > switched to using the fftwf_ function variants, which are float based instead > of double. Both original zyn and yoshi 0.058.1 do _not_ suffer the pitch > weirdness. With no sound theoretical basic that I understand, maybe it'd be > worth trying a revert to doubles through the fft processing? > > That's easy enough for me to put together, but I'd like to hear thoughts on > whether some kind of weirdness through the transforms might actually result in > the problem we're addressing. > > cheers, Cal > > ------------------------------------------------------------------------------ > Special Offer-- Download ArcSight Logger for FREE (a $49 USD value)! > Finally, a world-class log management solution at an even better price-free! > Download using promo code Free_Logger_4_Dev2Dev. Offer expires > February 28th, so secure your free ArcSight Logger TODAY! > http://p.sf.net/sfu/arcsight-sfd2d > _______________________________________________ > yoshimi-user mailing list > yos...@li... > https://lists.sourceforge.net/lists/listinfo/yoshimi-user |