Re: [LAU] Explodomatica - explosion sound effect generator

Yeah, I know there's a bit of clipping. I think it's because I'm summing what amount to a bunch of normalized random data -- so some of those sums end up outside -1 < x < +1, though I do normalizing steps between sums, so... I don't know exactly where the clipping is coming from -- I think it's coming from my "poor man's reverb" effect, which, that thing should probably be replaced with something better and faster, but... like I said, I don't know JACK about DSP, I'm just playing around with my computer and my compiler and my soundcard. Best thing ever would be to send me a patch! -- steve ________________________________ From: Jeremy Salwen &lt;jeremysalwen(a)gmail.com&gt; To: Stephen Cameron &lt;smcameron(a)yahoo.com&gt; Cc: &quot;linux-audio-user(a)lists.linuxaudio.org&quot; &lt;linux-audio-user(a)lists.linuxaudio.org&gt; Sent: Friday, October 28, 2011 9:09 PM Subject: Re: [LAU] Explodomatica - explosion sound effect generator Very cool.  I was playing around with it.  I do have one suggestion though: It seems you are introducing unwanted artifacts due to clipping. First, you probably don't want clipping.  However, it is showing up, even in your demo samples. Second, you might want clipping (Probably we've ingrained digital clipping in our perception of explosion sounds), but if you do, you're not clipping in the right way.  I haven't looked at your code, but because I had the exact same symptoms while using libsndfile, I'm guessing that what you are doing is feeding libsndfile floats with values >1 or <-1.  If you want proper clipping, these samples would be pulled to the closest value within the range [-1,1].  However, due to performance reasons (I'm guessing), this is not what libsndfile does, but instead it wraps around.  So 1.1 will be interpreted as the amplitude of -0.9.  Here's what it looks like:  http://i.imgur.com/sselM.png  That image is from demo0.wav.  This causes the "clicking" artifacts you hear in the file.  In order to fix this, just go through and find any samples > 1 and replace them with 1, and similarly with -1, right before you write the data out with libsndfile.  This should give you proper "chopped peaks" clipping.  Or, if you don't want clipping at all, just calculate the max amplitude and divide every sample by it before outputting. Good luck and thanks, Jeremy On Fri, Oct 28, 2011 at 2:05 PM, Stephen Cameron &lt;smcameron(a)yahoo.com&gt; wrote: