How to robotize your voice?

Post Reply
Darkhog
Posts: 250
Joined: Wed Apr 06, 2011 11:03 pm

How to robotize your voice?

Post by Darkhog »

Is there any tutorial how to make "robitzed" version of your voice in SunVox, kraftwerk style?
User avatar
NightRadio
Site Admin
Posts: 3944
Joined: Fri Jan 23, 2004 12:28 am
Location: Ekaterinburg. Russia
Contact:

Re: How to robotize your voice?

Post by NightRadio »

I'm not sure it is possible at all. You need some sort of vocoder, but this module is not implemented yet.
gilzad
Posts: 92
Joined: Wed Dec 19, 2012 6:46 am

Re: How to robotize your voice?

Post by gilzad »

I might be able to build a metamodule for this.
The compressor in Sunvox has this great side-chaining feature to make a foreign signal duck the audible sound. I haven't checked if the Metamodule passes a second signal the same way on to the compressor (so it can take input 2 to duck input 1). If this is possible, I could use about 16 self-made gates behind filters and modulate them with a voice track. So the formants of the voice track would open/close the 16 gates which would respectively change the color of a chord that you'd play with any instrument.

Another thing is: A generator can be modulated by another signal, right? I should test what can be achieved by doing this.

edit: It is possible. I'm on it. Modulating the generator won't give exactly the vocoder effect although it will sound familliar. I've built a 20-band Gate but need to refine it until you can understand the vocals. But it is possible in Sunvox.
User avatar
NightRadio
Site Admin
Posts: 3944
Joined: Fri Jan 23, 2004 12:28 am
Location: Ekaterinburg. Russia
Contact:

Re: How to robotize your voice?

Post by NightRadio »

I might be able to build a metamodule for this.
Wow. This will be great!
User avatar
samrai katt kovboy
Posts: 309
Joined: Wed Jun 13, 2012 11:07 am
Location: Spain
Contact:

Re: How to robotize your voice?

Post by samrai katt kovboy »

+1
SOOOOOOOOOOOOOOO MANY CIRCLES
gigi
Posts: 92
Joined: Sat Dec 15, 2012 3:33 pm

Re: How to robotize your voice?

Post by gigi »

+1
gilzad
Posts: 92
Joined: Wed Dec 19, 2012 6:46 am

Re: How to robotize your voice?

Post by gilzad »

I'm really on it, just give me a bit more time so it will become a useful module.
The things that I've faced so far in the multiband-gated version:

If the Gates have a threshold of zero and if each Gate's attack- & release-time is the same, then the vocals won't form the carrier (e.g. saw-chords) strong enough to have recognizable vocals. Sibilants - that are distant enough from our vowel range - come through well. But each vowel (A, E, I, O, U) has enough formants to open up all Gates at the same time. Thus the end signal makes it hard to distinguish which vowel is being spoken. So it will sound like a vocoder, it will give you robot sound but the quality will be rather poor. Now if I set the threshold of some Gates up, then the modulation of the carrier becomes clearer. Unfortunately this will be a very biased vocoder then, being sensible either more to A, E, I, O or U. And since every human creates different formants on each vowel, we'd run into problems sooner or later anyway.

In a second attempt (yes, I'm working on it as much as I can) I decided, I would need more filters in the vowels-range to have more dedicated Gates for each vowel. Didn't help. Again, if I chose to have equal Gating parameters (Attack, Threshold, Release), then the carrier wouldn't be modulated enough except for the sibilants. Actually, there were more attempts after that, like having a more aggressive slope but still the frequency gating is working like binary. Either a range comes through or not.

So! Where I'm standing right now is to use the Vocal filters. I knew they were at hand all the time but for some strange reason I thought I should go the classical way and build my own filters. That strange reason is again, that I wanted the frequencies to be biased as little as possible. That's why I started with 20 bands already. There are cool vocoders like Yamaha's PLG100-VH that sound very natural (like you) only with the notes you play. That's where I wanted to go. I'll leave this for a later attempt.

Now, while building my vowel-filtered-gate, I thought I could do some smart routing, so one could set the filter on the modulator and on the output at the same time. Theoretically this should work, because I'm seperating the stereo channels to have the left channel offer the modulator and to have the right channel offer carrier. But Sunvox is too smart here, that's where I ran into the feedback-loop-suppression. Well, I'm actually thankful that NightRadio has even thought of this. But now I need to ask for another feature for metamodules: It would be great if one parameter could be assigned to many controls. A module like the X/Y-Metadevice in Renoise would solve this easily. But that's just a suggestion for the later future.

Anyway, I just wanted to let you know that I haven't forgotten about this, nor have I given up. There will be a vocoder metamodule soon.
User avatar
samrai katt kovboy
Posts: 309
Joined: Wed Jun 13, 2012 11:07 am
Location: Spain
Contact:

Re: How to robotize your voice?

Post by samrai katt kovboy »

Amazing!
and even without success I will (and pretty sure a lot more people) be grateful for the time that you put into this project.
funnily never had the urge to robotize my voice until now.
keep up the good work!

going through the basement found some old sounds
threw them in the air and a g string struggled and found its way out
SOOOOOOOOOOOOOOO MANY CIRCLES
gilzad
Posts: 92
Joined: Wed Dec 19, 2012 6:46 am

Not done yet. Vocoder WIP

Post by gilzad »

No, no, there will be the Vocoder Metamodule! :)

I think I know now what the problem is and how to solve it. Some of my earlier assumptions were wrong.

My problem is the voiced sound in the modulating speech. Take a bass (e.g. Barry White ;) ) voice who's speaking "Eeny, meeny, miny, moe" into my WIP-vocoder. Then I have these gated filters and those around 2-4 KHz will open up for the "eee" (eeny, meeny), the ones around 0-1 KHz will open up for the "ou"(moe) and the ones around 1 - 2 KHz will open for "aaa" (miny). But the problem is that the word "Eeeny" spoken from a bass voice will still offer so many lower frequencies, that the "ou"-gate and the "aaa"-gate will also open up. This makes it hard to distinguish the spoken words from each other. What you'll hear from the vocoder will be something like "AANAA, MAAANAA, MAANAA, MAAA". I was wrong to assume that the Gates work like binary. They do controll the loudness relative to their side-chain-input as well. But maybe the reduced bandwidth requires us to isolate formants stronger from each other over time.

Again the side-chaining feature in Sunvox's Compressor is gold here. Using that, I'll try to make the ees duck the ous, make the ous duck the ees and same goes for the aas. And then I'll have to hope that the stronger vowel wins. Oh, well... :)

I'll show you a first and very basic test of this might work for one band behind the Gates. I might try to do the concurrent ducking before the Gates, too, just to compare the quality. The following examples are no Metamodules yet, only Sunvox files to show you the progress. Again, use the "COMPARE" DSP to hear how the vocals sound without the vocoder.

Vocoder (work in progress), 20-band

Another example making use of Vocal filters (only 10 of them) shows a simpler but yet effective approach.

Vocoder (work in progress), vowel-filters


Now, in both examples I've set differend thresholds for different formant ranges. This has a bunch of disadvantages. The most notables being the input volume of the voice. If the carrier is too quiet, then too few Gates will open up, so you won't hear the expected vowels. If the carrier volume is too high, all of the Gates will open up, making the vowels hard to distinguis from each other. That's why I desperately want to move away from any Threshold above zero.

If you guys have ideas how to optimize all this, I'll be glad for sure.

I'll keep you updated.

Edit: corrected bold-tags
Post Reply