Voice manipulator

[Batty]

Hi again.
As I'm starting to do my own CGI Films now, I came upon an idea...or better a necessity and I wondered, if the many Melda Plugins can already do this - or if it is possible to invent this as a new module:

I know and have this effect already from the developers "Antares" and their plugin "AVox Throat evo".
But it still isn't developed very well (introduces artifacts and it doesn't go far enough in its possibilities).

I have a need to give my characters different voices - from an old man and/or woman to young ones and kids - and with the different characters of the human voicebox. The speech-dependent things are naturally not included (dialect, language and such :p).
Only the voicebox, it's formation, raspyness, narrow-wide throat, gums, thick or thin tongue...whatever forms the vocals and the path of the tone. And what ever makes the voice and gives it it's character. Okay MickeyMouse is an easy one...but I can't get AVox Throat to make a believable voice that sounds close to Joe Pesci's (not that I needed Joe's :p, please take it as an example ^^)...Avox distorts it with lots of dirty artefacts...it sounds like waves and Avox's range in this regard is very very limited.
And when it comes to a singing voice it's even more restricted as the change from Male to female vocal or to Child (MickeMouse) is pitch dependend and this feels simply not up to todays abilities of the many Melda plugins.
Is there a chance or a possibility for this as a module or can I use Mxxx for that?

[vectorwarrior]

I work in a similar field, doing a lot of monster sound design for a large project for the last three years using almost exclusively melda plugs.

There is nothing like avox throat on the market for subtle voice box and throat manipulation. It's a shame it's so frustrating to use. I was considering biting the bullet and getting it recently, but I don't want to. Another option is ircam Trax, which contains an effect that attempts to manipulate voice. You can specify what the input is and change the output to old\young, male\female\child. Plus alter raspiness, etc. It is okay, but too artifacty for my work and expensive.

Beyond that, if you're after more monstery effects melda has you covered with mtransformer, mmorph, mvocoder and others. I've been making some pretty insane mxxx setups and we've recieved alot of positive feedback about the uniqueness and quality of the monster sound, so I'm pretty happy. But for more subtle and natural voice alteration the above is going to be too extreme. For example, mtransformer is essentially a spectral frequency shifter (amongst other things) and as you probably know, frequency shifting makes things pretty wacky as it totally breaks the harmonics and sounds super unnatural.

There are some other melda effects that sound like they could help, mcharacter, mpolysaturator for example, but generally these things work with pitched voice, like singing and are pretty unhelpful for speech in my opinion.

Hope that helps.

[Tagirijus]

I probably cannot add that much more, but I would also mention MGranularMB, which I love to use when making special voice effects! There should also be a MXXX active preset I made, which is called "Monster Voice" or so ... I am not on my machine right now and I cannot remember the title of the preset.

Interesting to hear about the plugin "Throat" ... I did never hear of it and I feel like I am living on the moon again ... how can I not know about such plugin, while working in this kind of industry!? I personally still would not buy it. I would rather wait for Vojtech to come up with something similar or (more likely) wait for some user here, which comes up with a great melda-technique to achieve similar results, hehe.

Offtopic: making special voice effects is so much fun. A pitty that there never is a need of my clients to do such cool shit ... I really have the wrong kind of jobs!

[MeldaProduction]

Well, this is actually really complex... And I'd actually love to have time to play with that, but as you know there are some projects I need to attend to first, and there's so many of them, that I don't think I'll ever have time , but who knows...
Anyways the thing is, the classic mickey mouse / giant is produced by altering spectral envelope (which in effect changes formants). But the vocals are much more complex than that, one would somehow need to identify different parts of the spectrum of the sound and somehow alter them separately. I think a neural network approach could work with that, but that would require some solid development plus thousands of samples of various people, so...

[vectorwarrior]

I was thinking one potential approach which is almost possible, is to sort of remap the amplitude envelope of one frequency area to another. That could push the tonality of a person's voice from one range to a slightly different range.

You can almost do this with mxxx, by using a spectral dynamics, but use the side chain as the detector input, and put a transformer into the side chain before it goes to the spectral dynamics. You can then shape how you want the signal to be remapped. This works great except that spectral dynamics doesn't let you volume match. What you'd really need, and I've asked for this before, is a spectral phatik instead. That would let you match the signal how you want it remapped.

Anyway, Vojtech, have a think about it. Aside from that... those crazy other ideas you had sound wonderful... but i thought the above idea might be considerably easier to implement.

[jmg8]

+1 for spectral Phatik!

[MeldaProduction]

I'm actually not sure what you mean by "spectral phatik". But anyways remapping a spectral envelope could be an interesting experiment.. But I'm quite worried that it wouldn't get you what you actually want, because the human vocals contain resonant points, which need to be identified and treated separately, which is kind of the main problem. Nevertheless it could be worth experimenting. I think it could be easy to add to MTransformer, will check.

[pekbro]

Cant you just use Melodyne? Everyone must have a copy of it by now.

[vectorwarrior]

I'm actually not sure what you mean by "spectral phatik". But anyways remapping a spectral envelope could be an interesting experiment.. But I'm quite worried that it wouldn't get you what you actually want, because the human vocals contain resonant points, which need to be identified and treated separately, which is kind of the main problem. Nevertheless it could be worth experimenting. I think it could be easy to add to MTransformer, will check.

So... Spectral phatik. The problem with Spectral Dynamics is that when feeding a signal into the side chain, there is no way to say 'make the main input MATCH the dynamics of the side chain'... currently, you can use the side chain input as the detector, but SpecDyn uses a static threshold/ratio compression/expansion, rather than doing a comparison between the input and the side chain as saying 'ah, this frequency needs to raise by 3dB, this frequency needs to be reduced by 7dB', etc across the whole spectrum.

In the Melda world, MPhatik is like a single band version of what i'm describing above... instead of doing, for example, a static 3:1 ratio reduction above the threshold, like a compressor does, it COMPARES the input and side chain, and determines what to do. What we're talking about is doing that across the whole spectrum, just like SpecDyn does with compression. Now, the key thing is that the side chain signal could just be the input, but transformed in the spectrum, using MTransformer... that's the twist.

Anyway, yes, you're right in that it's more like a 'spectral envelope remap'. Which... is a pretty cool name

I know what you're saying about the resonant frequencies, but that's where the awesome melda frequency graph is so helpful, it would be super easy to customise the distribution curve to avoid any weird behaviour.

Sorry if I'm rambling or not making sense, I've been in sessions all day and my brain is fried.

[Batty]

Wow I didn't know that it would be that complicated
Good that I asked :p

I really thought vocoder-like approach with about 1000 Bands and a good tone generator that reacts to the vocal input pitch adjusting the genarator to the source vocal's pitch would do.
Or something like MCharacter/MMorph perhaps.

Sadly I don't need monsters :p have some at home already ^^
Well for my film I am in need of 2 old males about 75 in Years of age - both should sound distinctive...Vocalizer doesn't do this as far as I know.
Avox does some weird wavy artefacts while pushing those vocals.

Vojtech, I hope you lean more on the ...but who knows... part (I love it when you do)
Anyways I'm really glad that Melda exists even without MSpectralVocals (...sounds like ghosts).

[MeldaProduction]

vectorwarrior: Aaahaom but that would need you to generate the actual envelope first, which would be rather difficult. The MTransformer approach should be much easier. But good point about MSD, I missed it myself a few times...
EDIT: Btw. what you could do is actually use MMorph probably...

Batty: Hehe yeah, monsters are everywhere!

[jmg8]

I'm interested to see this new addition to MTransformer!

[vectorwarrior]

vectorwarrior:
EDIT: Btw. what you could do is actually use MMorph probably...

Yeah, i use MMorph A LOT, it's a central part of my monster processing approach... but there's some reasons why it's not perfect for this, which I'd love to demonstrate if i had more time, related to how it attempts to 'match' signals... but I'm in the middle of another 13 hour day and this is my half hour break!

Now, that said, i did just try really quickly morphing using a transformed side chain of the original... and you know... it's pretty darn interesting actually. Definitely worth trying out if people are interested in this. If i get time later i'll investigate more (probably next week).

[MeldaProduction]

So, basic spectral envelope transformation implemented and works surprisingly well! Just transformed by ex-singer's voice to much more jazzy character, which she needed .

[vectorwarrior]

So, basic spectral envelope transformation implemented and works surprisingly well! Just transformed by ex-singer's voice to much more jazzy character, which she needed .

Woah, awesome! Very excited to try this out!