Fft

whats so great about FFT modules?

 

You can do all sorts of cool things with them.

-Paws

 

Hahaha, it's all so clear now.

 

they analyze and do sound treatment in real time. they use convolution (if you want to do a search in google;-))

you can capture the response of a room an put it to other sound to create new type of reverb, you can do filters,etc...

 

As you might know, any sound wave is composed of a number of sine waves of varying pitch, amplitude and phase (time offset). Using FFT one can deconstruct a sound into these constituent sinewaves (as I understand) and use that information for all sorts of cool things

-Paws

 

release

wow, thats cool, how come reaktor doesn't have it? are the guys at NI planning on releasing this in the future?

 

The FFT is what enables you to convolve two signals in real time. Convolution doesn't enable fourier analysis.

 

That's not exactly true. Fourier's theorem states that any periodic waveform can be recreated using an infinite number of sinusoids. It doesn't make any assumptions as to what the root elements of sound are or anything like that. Just being nitpicky.

 

Re: release

Reaktor doesn't have it because it isn't so simple as slapping in a slew of FFT modules. You'll need a new signal type and such. Another possibility is that it would cannibalize sales of Specktral Delay. Personally I'd like to just see a simple A<->B convolution module. I think the best way to go would be with a series of modules that make use of the FFT internally, but don't expose the workings to the user, similar to how Reaktor deals with granular synthesis. There's no grain module that you use to do all those fun tricks with. Such higher level frequency domain processing modules would be easy and quick for everyone to use.

 

Well, what do I know

It does allow you to do cool stuff, however, right?

-Paws

 

Oh yeah.

 

Re: Re: release

echo...echo...echo

sowari

 

Re: Re: release

Don't know if you read my feature request on the builders list about vectors and buses. Anyway, I'll soon be posting an ensemble that comes pretty close to doing a spectral delay.

 

Re: Re: Re: release

If you can separate audio into 1024 frequency bands efficiently, you have my vote.

 

Re: Re: Re: Re: release

You'd be surprised what you can do with 256 filters.

 

Wow. I'm looking forward to this!

 

Re: Re: Re: Re: Re: release

You'd be surprised that there is no way in hell that will run on my Powerbook.

 

do a search about benchmark threads you 'll understand. *cry*:-(

 

Re: Re: Re: Re: Re: Re: release

I'm running it on an 866 12" Powerbook. You would be amazed at how efficient Reaktor can be when precessing lots of very simple voices.

What am I talking about? I really don't like to make preannouncements but it's a spectral delay/additive resynthesiser/vocoder thing that I've been working on for the past while. I'd have posted it long ago but each time I add a new feature it suggests another to me. I have a few extras still to add and I need to clean up the user interface. The thing is based on the very simple idea of using an event table to store a sonorgram (thanks to SolarX in laying the foundations with Metasyn64). I've added an analysiser which lets you use it as a "spectral sampler", or indeed as a spectral delay, or in fact as a mad realtime spectral resynthesiser thing, or as a strange evolving sound wash generator... This is why the thing isn't finished yet.

One thing is certain though - it's shown me the potential for using polyphonic signals to advantage, and it's the reason I'd like to see vectors (or something similar) in Reaktor. If we had vectors then an FFT module would be easy - it's output would just be a vector.

Ab

 

A couple things... First, how do you plan on doing the kind of brick-wall filters that you can perform via frequency domain processing using filters designed to simulate analog filters? Secondly, this term "vector" is getting thrown around a lot lately. A vector is a direction plus a magnitude of some sort, usually speed or direction.

"If we had vectors then an FFT module would be easy - it's output would just be a vector."

I don't really understand what that means. The reason we would need a new signal type is to handle the complex numbers. I'm not sure how vectors would aid us with implementing a fourier transform module. Can you explain what you mean?

I'm running it on an 866 12" Powerbook.

Yes, I'd not be surprised if you can run 256 bandpass filters at once on a newer machine. However, if you want to do anything useful like psuedo-convolve two signals, you'd need to use 512 filters (two input signals), and then multiply the results of all of them somehow, and then covert that back into a signal audio stream. That is what I don't think I can run.