posted 30 November 2007 22:45            

I have a singing bowl sample which I want to do spectral analysis on so that I can mangle it - stretch etc during performance - I want to start by playing the sound in a life like way - ie. it sounds the same as the bowl and then slowly manipulating it based on the number of times I hit the real bowl. I want to use FFT so that I have independent control over time and pitch, BUT I can not for the life of me actually get a good analysis of the bowl.

The sample is here http://download.yousendit.com/4CBAEBB91F9AFF13

Just wondering if others would like to try and give me some advice on getting a good reproduction from a spectral analysis?

posted 01 December 2007 07:18            

bowl1.kym

The first thing you need to do is re record the bowl without the rustling clothes and the breathing in a the end.

We humans are very good at listening to sound as a whole and separating objects into the component parts (computers can't even separate the instruments in a mix yet) so we don't even notice that the rustling clothes or the breath are effecting (and at some frequencies even drowning out) the bowl sound as we've already separated them in our head and filled in the missing bits. But an analysis looks at each partial as it's own separate entity and in many of the partials the contribution of the rustling and the breath are louder than the bowl its self.

In the attach live analysis I've added a pitch smoothing and an amplitude smoothing. If either of the gets to value one it becomes a freeze forever. If you put both of the controls to almost one (say 0.99) the rustling and breath become un noticeable as they are both fast changes But of cause the initial attack of the bowl gets smoothed out so that it is unrecognizable.

You could control these by an upside down ADSR that was triggered at the same time as the sample is triggered to satisfy both requirements.

but better still is to record the bowl on laboratory type conditions.

hope it helps

Pete

posted 01 December 2007 16:51            

Thanks heaps - this forum rocks - the quality of advice and support is unbelievable - thanks heaps.

I agree the recording is poor - it is actually from a studio session - the engineer extracted a single hit and placed it on YouSendIt for me - I tried recording it at home again but the bowl is so soft and the tail so long that it seemed impossible to get a clean recording, so as you suggest, I shall try again on the studio. Having said that, I also though there must be a way to filter or to post-process the FFT - ie. in AudioSculpt, one could hand clean.

Thanks again for your support and the patch - I will look at it in detail

Cheers

posted 02 December 2007 01:51            

or if that space is dicey on ftp, try http://www.greyotter.com/z-mp3s/Bowl24_192_s32
and take out the underscore before using

don't know if that's any better than what you've already got, though...

posted 02 December 2007 02:43            

nuts those links don't work at all

If you'd like to look at the spectrum file I made, please send an email to bluefire@greyotter.com, and I'll attach the file in reply. It's 1MB, so it can't be attached on the forum.

bf

posted 02 December 2007 12:54            

Low Freq = 136 hz
High Freq= 140 Hz
Thresh = 2828 Hz

Then save the Tau file and click the Gallery button. There are some fun manipulations of the bowl in there. Also, playing with the Formant envelope in the Tau Editor gives some surprising results(!)

posted 03 December 2007 13:27            

Re the Pre Filtering:-
You could remove frequencies below the fundamental pitch but this won't really help that much as the main problem is the effect of the unwanted sounds is having in the common frequency bands.

Re post processing:-
Well thats in effect what I am doing with my sound but I'm doing it live. You could just as easily make a file with the spectral analysis tool and do similar smoothing in the spectral editor. Here you can do it selectively in both time and harmonic region and hand clean the file just as you suggested.

Pete

posted 10 December 2007 03:15            

I analyzed it from Kyma regular (not TAU) using the 6th level for fundamentals. Even though the machine sees a fundamental of 566 Hz (or so, I'm writing this from memory), and the 6th level fundamental is slightly above that.

My logic in resaving the bell.aif file to 24-192 was that it would interpolate values and give the analyzer more to work with, even though the aural product would be equivalent.

That reasoning may be nonsense, and this just a lucky shot LOL.