posted 05 April 2002 00:03            

Anyone actually have experience using Arboretum's tools? And what exactly did you attempt to do with it?

posted 05 April 2002 06:10            

But I think they are also analizing the second signal and controling the filters by the differance between the two. This is the same principle as the corrected vocoder which I keep going on about (but they are using just two static EQs instead). This would stops it from being a covolution of both.

This is similar to a thread that was on this forum about a year ago.

I don't believe that the average of a sound of Barbara Streisand has anything to do with the fundemental timbre of her voice but more to do with which words and what notes she is singing. Some one else singing the same song would have an average timbre closer to this EQ than one where Barbara herself was singing a different song.

I suspect that if we did a test where we used these derived EQs on sound and then used random EQs of the similar variations , we wouldn't be able to say which is which. I think the emporers new clothes may be working here. I used to have a vocoder with a freeze function where you could sing a note and then press freeze and the EQ would remain there. It use to make sounds that put through it sound special(different). But once you try to use it as a mastering EQ you learn that the sound is best left alone or EQed the convetional way.

The Morph Control (one of the most missused words) seems to simply change the static EQ control from corrected to flat. In other words it changes the tone a bit.

With regards to getting the avarage of the timbre over time in kyma ,I have some module clusters that use feedback to accumulate the average of the a control signal But I think you may get better results by editing the analisis in the spectrum editor and doing "smooth amplitudes".

The corrected vocoder dosn't yet exist as far as I know, it's just a dream of mine but maybe one day?

If anyone has tried this and can correct my asumputions on what its doing or what it sounds like, I would love to hear.

posted 05 April 2002 13:02            

What do you mean by "corrected vocoder"? I tried searching for postings by you on this topic and came up dry.

I posted a sound the other day that performs amplitude modulation across 6 bands of sound in an attempt to get something akin to vocoding but not as intense. After playing around with the Kyma vocoder some more, I found I could achieve similar effects by narrowing the bandwidths and backing off on the voice input somewhat.

I was looking for subtle timbre changes in rhythm with spoken phrases. Vocoding generally provides too much. Most of all I wanted to avoid the intensified highs provides by sibillant speech. But the vocoder also provides a rolloff and passband filtering on the output that help here too.

posted 05 April 2002 13:25            

But I think you will find that even here, unless the person singing the two sections of sound are the same, their formants will be different and the apparent EQ difference will reflect this as well. I also think Pete is on to something when he states that the average has a lot to do with exactly what was spoken, since certain formants will be emphasized.

You know, I understand the apparent need for marketing hype, but I really tire of it. What is the Arboretum super duper special algorithm. Why can't they just be honest about it and state what they are really doing. Any scientist would be ridiculed off this planet for making such grandiose and unsubstantiated claims. Why do we put up with this in marketing? (end of rant...)

posted 05 April 2002 14:50            

Anyway

I think we talked about this before. With a normal vocoder you have two inputs. One has a bank of filters which are then rectifide and smoothd and turned into control signals. These control signals control a set of VCAs that are stuck on the outputs of another bank of filters which the second input signal is passed through.
This means that if the same signal were put into both inputs you would get the square of the spectrol envelope on the output.

The idea if the corrected vocoder is the same but you have both inputs turned into control signals and with a simple bit of math prosseing,the VCAs can be controlled in such a way that the output matches the spectrum of the first input signal regardless of the second inputs own spectrum.

I.E.if input one has level of 100mVpp at 1Khz and input two has a level of 50mVpp at 1Khz then the (1khz)VCA would be given a gain of 2 .But if the second input had a level of 150mVpp then that VCA would be given a gain of 2/3. This does of cause mean that if the second input signal had a level of Zero at 1khz , the maths would try to give the VCA a gain of infinate but we simple limit this or maybe even cut the gain compleatly if input twos signal is too low in any band.

With this vocoder you could put the same signal in both inputs and get the same signal out, but if the center frequency of one set of filters where then moved up or down in frequency you would get formant shifting. I dont know of another device that can do formant shifting on anything other than non polyphonic sounds. Of cause this won't be perfect but might give some very interesting new effects. I also think the vocoding effect when used with inputs signals that aren't the same, will be stronger and more in tune with the controling signal.

posted 05 April 2002 16:18            

Yes, I remember this discussion with you... In the back of my mind, I am thinking "thresholded control levels making 1 or 0 on the VCA". But that will probably sound too coarse. Maybe we need something akin to this, but with some rounding. How about N'th root of control signals. That pushes them mostly toward 1, but when they decay away the VCA control would go to zero.

Formant shifting sounds very interesting, but how do you separate formants (which are envelopes on the spectrum) from spectral lines caused by voiced sounds? I can see how a survey of singing over the duration of a song can yield information about these formant shapes and boundaries, but doing this instantaneously would seem very difficult. These formants are very nebulous sorts of things...

- DM

[Ahh... there is a way to do this separation of formants and spectral lines in "real time". It is called affectionately, Cepstral analysis. But this works by taking the complex logarithm of the Fourier transform of the Fourier spectrum, which implies not only logarithms as you know and love them, but also accurate phase unwrapping, since these are logarithms of complex valued numbers. That is a huge impediment to doing this accurately. Furthermore, every block of sound needs to go into one FFT to get the spectrum, then another FFT followed by complex log to get the Cepstrum. Then you can separate these multiplicative spectral components, modify them, and then go back with a complex exponentiation followed by an inverse FFT to get back to the modified spectrum, and then another inverse FFT to get back to the modified sound block. Whew!]

posted 07 April 2002 01:12            

I have been searching for days now for a method of imposing some kind of formant brightening on sounds but without the strong speech character of the modulation. Your RE Analysis with only 16 bands did the trick. I had to attenuate the paddish excitations by almost 50 - 60 dB, but this is what I have been searching for.

Kyma is so incredibly deep and wide that it is difficult to remember all the possibilities. Thanks!

posted 09 April 2002 12:11            

quote:

---

Originally posted by David McClain:
Anyone actually have experience using Arboretum's tools? And what exactly did you attempt to do with it?
- DM

---

I have used this feature, with mixed results. I found it more effective on final mixes, and only when "morphing" to another mix with similar instrumentation. (Which is exactly what Arboretum tells you to do.)

Their example uses Nine Inch Nails, and as you might expect, it works very well. My mileage varies. I find it most useful when trying to restore sound from very bad sources - e.g. old cassette recordings.

posted 09 April 2002 13:00            

What is it about Nine Inch Nails that would make this technique work very well?

posted 09 April 2002 23:09            

quote:

---

Originally posted by David McClain:
Hi Bill,

What is it about Nine Inch Nails that would make this technique work very well?

- DM

---