Wednesday 11 September 2013

Kawai ROM switch update

So this custom ROMs business is pretty slow progress I am afraid, but be patient I am working on it as fast as I can but this is a spare time project and I haven't much spare time right now.  My first 808/909 kit worked but was had so much wrong with it that it was faster to start again from scratch.  So I moved on to a sequential circuits pack and I am much happier with the results.

I am using soundforge to do all my editing.  I take the original ROM sample, zoom in as far as you can go and place markers at the zero crossings closest to the beginning of each sound.  Each new sound has to be the same length or less than the original sound.  The samples need to be conditioned before you put them in - they need to be of a similar output level, monoized, sample rate/bit depth converted, pitched up, normalised then overwritten in place of the original sample before being converted back into a BIN file for burning.

Why pitched up?  well the R-50 and R-100 both allow you to repitch the sample within the drum machine, but the sample is stored on the EPROM at its highest pitch/fastest rate.  So 0/midpoint is actually the sample pitched down by half an octave.  And so the Kawai specs of 12 bit/32K sample rate is actually a bit of a lie, it would only be 32K when all samples were at their highest pitch.  But it is probably this sample rate reduction that gives the Kawai R5/100 so much punch!  It is also this part that is currently knocking me down.

You see when I repitch things they sound wrong, there is a metallic sound which I am putting down to aliasing.  Now when I am doing my sample rate conversion within soundforge I apply an anti-aliasing filter, but this is for 32K.  What I THINK I need to do is to resample it at a (yet to be determined but presumably 16K if the tuning range is an octave) lower rate first, to filter out the offending frequencies that cause the aliasing when the samples are pitched down within the drum machine.

At the moment pitched stuff like kicks and toms sound pretty good while crashes and rides sound awful - I am putting this down to the frequency content.  Barely anything can get aliased in the kick drum because the offending frequencies found in the crashes simply aren't there.  If the 16K theory doesn't work then I think I can figure it out with FFT or spectrogram of the original samples.

Annoyingly though this is only guess work.  I may not figure it out at all, this is all very new to me.  It means I am not yet able to offer custom ROMs.  If I do crack it then I will charge £25 per custom ROM or £15 if it is one I have already made.  I will do some kind of discount if you buy the ROM switch and ROMs together, but I have not yet figured the price out for that yet.  If you are interested in buying any of the factory ROMs now, email me at analogmonster at live dot com

No comments:

Post a Comment