Vertigo vs Harmor vs Alchemy

[olepro]

i would not expect alchemy's additive import to do well in that example... because of the strumming, there isn't a single root note for it to use for import. spectral should be a lot better at the resynthesis (and it is), but perhaps not as good at the pitch shifting side.

harmor does sound really good here!

I do know that Alchemy's Additive synthesis is not for complex material.
I just included it for comparison and because when you use Spectral import in Alchemy you don't have all the synthesis options as in Additive mode, but you do in Harmor.
Another thing harmor does perfect is analysing single cycle waves and you don't have do anything special to get it working, you just drag and drop both in the image window or in the waveform window.

I could do a demo of Alchemys different settings under import (best time, best frequency...) but this is a Harmor discussion after all.

[Hunter]

You don't give up do you ?
Why do you think i don't know anything about how to import at sound into Alchemy ?

I'm just pointing out you are doing an invalid comparison, as you are not providing any information about how you have done your experiment in terms of settings, which means we don't know that you aren't using teh wrong settings for the job and thus providing an invalid comparison.

Could you provide the original sample as you used in the analysis, and the settings you used for your audio example?

[himalaya]

but this is a Harmor discussion after all.

Not really. Read the thread title again, and also see your own comparison which included *two* synths, not just one (Harmor). (not sure why I'm writing this as I will be shot down in flames again for being Capitan Obvious )

[jens]

Here, listen to this (sorry about the noise burst, I'm still exploring the Harmor demo). The audio example I posted below was made in less than 5 minutes. I start reaper, add a channel of Harmor, drag and drop a file, insert Edison, record the live performance, convert to mp3, post to the net. All this took less than 5 minutes. Now THAT's what I call creative power.

Harmor Kills

Incredible! Absolutely WANT!

I'd never have dreamt of saying this about a Gol-creation one day.

[luckystrike]

You don't give up do you ?
Why do you think i don't know anything about how to import at sound into Alchemy ?

I'm just pointing out you are doing an invalid comparison, as you are not providing any information about how you have done your experiment in terms of settings, which means we don't know that you aren't using teh wrong settings for the job and thus providing an invalid comparison.

Could you provide the original sample as you used in the analysis, and the settings you used for your audio example?

You heard his original sample then why don't you use it and show that the result is different in your system rather than being skeptical about others' testing method? If it's been wrong the sound itself would show the fact. Only then there will be able to doubt that the wrong method would have been taken in some of the test. It's more constructive and more polite.

[Nokenoku]

You would first try to retrieve specific characteristics of a sound, then mix those to get a new sound. But.. what are those characteristics, & is there any standard method out there?

No, you're right, there isn't.

I guess it's similar to picture morphing. If there's not a human, who marks points and stuff, some algorithms might be able to find forms/parts/clusters, which will then be moved, reformed, etc. and actually "kind of work".

Sorry, I'm really missing better words here. But I'm thinking about a technique, which could also be similar to what motion interpolation in TVs does.



Or maybe it would make sense to start by filtering a signal with the formant structure of the other signal (similar to a vocoder), leaving the time domain untouched first.
Or while doing that, applying the "pattern-type-morphing" I just mentioned in the time domain.

I guess there's a lot of trial and error, if you want to develop a really good audio-morphing algorithm.

[tony tony chopper]

Sorry, I'm really missing better words here. But I'm thinking about a technique, which could also be similar to what motion interpolation in TVs does.

yes, but it's not like motion interpolation really worked well, just like we haven't heard audio morphing that sounds really amazing so far.


To me this is tech of the future, and it must be really frustrating to work on something knowing that it won't be usable immediately, and will only get "better than nothing" in 20 years or more.
I even believe simple timestretching can still evolve for a century.

[olepro]

but this is a Harmor discussion after all.

Not really. Read the thread title again, and also see your own comparison which included *two* synths, not just one (Harmor). (not sure why I'm writing this as I will be shot down in flames again for being Capitan Obvious )

You are kind of right
I just didn't want to do an Alchemy compaired to Alchemy test when the thread is about three synths...

It would be nice if others would do some comparison demo's... if they dare

[Nokenoku]

Imo there are already some quite good working motion interpolation algorithms. Of course, they're not perfect, and usually they have problems with transparent stuff etc., but that's where the fun as a developer starts, isn't it?

To me this is tech of the future, and it must be really frustrating to work on something knowing that it won't be usable immediately, and will only get "better than nothing" in 20 years or more.

I don't agree with this.
If you're developing such algorithms there isn't "knowing that it won't be usable immediately, and will only get "better than nothing" in 20 years or more".
It's rather "Oh, my algorithm has problems with that particular thing. So I have to find a solution for this."

Most algorithms for motion interpolation for example seem to don't even consider/know transparent objects, so they totally fail at that point.
But it's not like that problem would only be solvable in xx years. It's solved as soon as you include a proper technique in your algorithm.



The first time I tried out such a motion interpolation algorithm I was actually surprised at how good it worked. For many types of scenes the errors are already not detectable.
I guess for audio it might be even easier. Our brain is easy to trick here (for example with simple reverbs, which are nothing like a real room-response and still sound convincing).

Also some of the examples from here are already quite good. And I think I heard some good audio-morphing also somewhere else in the past.

It doesn't have to be perfect to be awsome.

[tony tony chopper]

If you're developing such algorithms there isn't "knowing that it won't be usable immediately, and will only get "better than nothing" in 20 years or more".

in the video above, would you rather use the good old image blending, or the frame interpolation (& it's supposed to be the best one out there) with those really weird artefacts?
There's another video with a skateboarding guy, I couldn't find it, but I remember it was pretty ugly.

For many types of scenes the errors are already not detectable.

the problem is that it's.. slow motion, so you have plenty of time to detect the artefacts, as the rest is hardly moving

I guess for audio it might be even easier. Our brain is easy to trick here

I think it's more acceptable for audio, which is why stretchers exist & are used, however.. this is still dumb stretching.
If you listen to a drumloop, and imagine the same drumloop "slower", you're not imagining a timestretched version of that drumloop, but the drummer playing that loop slower, and this is very different. Now imagine how many decades/centuries until an AI is capable of extracting
-vocals
-drums
-instruments
from a song, & timestretch these separately?
To start with, there is no convincing vocal extractor yet..

[Krakatau]

If you listen to a drumloop, and imagine the same drumloop "slower", you're not imagining a timestretched version of that drumloop, but the drummer playing that loop slower

so true !

This sounded obvious to me once listening the prosoniq demo of their time factory 2 with "extreme" stretching applied to a few notes on piano, at that degree the rendering of the piano sounded totally unnatural just for some psychoacoustic reasons, the stretching engine itself was not to blame at all !

[rj0]

I think having a quality morph of one sound to another is only about half the value in having morphs. The other (and perhaps larger) value is in those sounds that arise in-between. Sometimes, something really exceptional is found.

In many ways, I see it more important to have methods to morph between sounds (and the more methods the merrier) than whether those morphs are "perfect". What develops might really amaze.

[rj0]

I think having a quality morph of one sound to another is only about half the value in having morphs. The other (and perhaps larger) value is in those sounds that arise in-between. Sometimes, something really exceptional is found.

In many ways, I see it more important to have methods to morph between sounds (and the more methods the merrier) than whether those morphs are "perfect". What develops might really amaze.

To get the creative juices flowing:

Imagine, if you will, using the mixing methods from the two Timbre waves to create a sound (already exists, more mixing methods would be great).

Then, drag/copy the sound to a time on a blank Image Synthesis timeline!

Repeat, perhaps with a different mix of the Timbres (or even different waves). Drop to a subsequent spot on the timeline. Rinse, repeat, building the timeline (attack, sustain, etc.).

Now, with sound targets on the Image timeline, select a morph method.


-----------

P.S. I know, not the most exotic morphing (e.g. such as between full spectral parts), but ...

[Nokenoku]

If you listen to a drumloop, and imagine the same drumloop "slower", you're not imagining a timestretched version of that drumloop, but the drummer playing that loop slower, and this is very different.

As long as it's only the drumloop, I'd just use Slicex instead. ^^

I see your point, but I really think the morphing doesn't have to be perfect to be good. Also it's not like with timestretching or seperating stuff out of a mix, that you're aiming for "realistic" results.
There is no reallife-form of a morph between for example a kickdrum and the word "kick", so as long as the sound of the morphing "makes sense" it's cool.

[Lotuzia]

If you listen to a drumloop, and imagine the same drumloop "slower", you're not imagining a timestretched version of that drumloop, but the drummer playing that loop slower, and this is very different.

As long as it's only the drumloop, I'd just use Slicex instead. ^^

I see your point, but I really think the morphing doesn't have to be perfect to be good. Also it's not like with timestretching or seperating stuff out of a mix, that you're aiming for "realistic" results.
There is no reallife-form of a morph between for example a kickdrum and the word "kick", so as long as the sound of the morphing "makes sense" it's cool.

Very good example from Didier and simple explanation : If a drummer plays a groove much slower first it wont work that good because usually a drummer will not use the same groove/fills/etc but more than that :

If I hit a REAL snare at 140 BPM OR 70BPM the snare will have the same short decay/release, provided that I hit it in the same exact way ( angle of drumstick, fastness, same position on drumhead, and of course same strength ). Same for a cymbals it will have the same release time. ANd this is where all stretch algorythms fail : They will stretch the audio segments or grains, and the 70 BPM version will have huge snares and neverlasting cymbals, leading to a pretty surnatural sound. I personnaly enjoy this kind of unreal sounds ( often used in Bristol school for example ) but this is NOT natural.

Its due to the lack of analysis of the primary audio material. The algorythm doesnt understand what drums are, and how they are played, it just makes an anlysis of a "rendered audio material", and tries to stretch it, that's a BIG difference !

LtZ