How do you mix spectra?

Question

I'm trying to mix spectra and I'm getting odd results and I'm not sure what I'm doing wrong.

I've got this Sound where I'm taking an analysis file and putting some delays on the amplitude (left) channel before joining the channels back together for the oscillator bank. 

 

And this works great!

Then I tried mixing in another analysis signal. I know I haven't synchronized the frames yet but the weird thing is even with the level of the second signal faded down to zero, the sound is completely messed up and really high frequency.

 

It's as if just the presence of the 3 Sounds highlighted somehow messes up the amplitudes of the "spect" spectrum. It looks the same on the oscilloscope. If I delete the highlighted Sounds it works fine again. Is the mixer adding some kind of delay that is making the Freqs and Amps go out of sync? Does the mixer get optimised away if there's no other input going in to it?

I tried adding a delay after "spect" but never found a setting that made it sound normal again. 

What am I doing wrong?

 

Comments

Is your SpectrumInRAM linear (or log)? I suspect that you might be mixing a linear spectrum (the LiveSpectralAnalysis) with a log spectrum?

---

Also, do you have the frame lengths of LiveSpectralAnalysis and spect set to the same value? (The frame length is a function of the analysis frequency).

---

Alan, could you please upload or email the Sound so I can check the log/linear and spectrum period?

Answer 1

There is no way you can mix spectral analyses the way you mix audio files. Imagine feeding a 100 hz sine and a 110 hz sine: Depending on your analysis settings they most likely end up being both on the same track. Adding the amps will work, but how would you deal with the freqs? Taking the 100 hz freq part with the mixed amps will produce a louder 100 hz sine, the 110 hz are gone. Imagine now mixing arbitrary analyses: You might add some amp signal to a freq signal which contains garbage even.

You can mix FFTs though by adding the real and the imag signals and get the same result as when mixing the time domain signals.

What you are doing here is more of a cross-synthesis than a mix. I hope this helps?

Comments

I was just mixing the left, amplitude, channel. That's a good point about the frequency bins being different. I was expecting the results to be a bit chaotic (this is a good thing).

But what's happening is if I mix the analysis signal (Amps) with a Const (set to 1) through a Level (set to 0), the sound is breaking completely. I would expect that mixing the Amps channel of a single analysis with zero shouldn't break anything, but it does.