Solaria

I was JUST looking at something like this — Emvoice.

Seems like it could be a useful demo/songwriting tool, and I’m sure there are situations where it might actually sound cool in a finished production.

Apart from that, I still want to hear real human voices. And I’m definitely not interested in AI-generated songs. Not that Solaria/Emvoice are intended for that purpose, but personally, I think AI songwriting is one of the corniest fucking thing I can think of.

And I think it's one of the very scariest I can think of. Seriously.

And above is further proof to me that we are destroying our civilization with AI, and that we will come to regret it. In the above example, it's not that the sweater song is good, or well written. It's that it looks more or less like what I'd expect from a young songwriter who's asked me to produce their first album. The day is coming, and it may not be very far off, where ChatBotAI writes the full lyric to a song that goes to #1 on the charts. The first time it happens, it will be an anomaly, maybe even sort of a gag, or an attempt to "go viral." But the day is coming when a music supervisor for a major show no longer listens to works submitted by songwriters and composers, but rather types (or speaks) into an app: "I need two minutes of moody, ambient music in a minor key featuring a slightly out-of-tune felted piano a la Nils Frahm and a cello." And the AIMusicApp will do it all. No actual musicians will make any money from this transaction. The revenue from this musical work will line the already-fat pockets of the C-suite at "AIArt, Inc.", none of whom have ever held an actual guitar in their hands before.

And the young songwriter who would've written the "sweater song" if the computer hadn't already done it? He never writes the 100 crappy songs he needed to write so that he could write his first really good song. He never gets to learn through creative exploration what it really means to be himself and to find his unique voice. And shit, man, it's just easier to get the AI to write the lyrics, anyway. Writing lyrics is hard. Telling the computer to do it will be easy.

EDIT: M57 : apologies, my friend. You were writing your post just above mine at the same time I was writing the above. I really appreciate you and your musicianship, so I hope you don't read my post as a lashing out at you personally. peace

At the current moment, it is humans destroying civilization. To effect that destruction, we are using a tool we call "AI". There may well come a day, however, when we lose control of the AI and it will no longer be humans destroying civilization. AI researchers are increasingly showing concern over this possibility.

I should also note that in this case, I am using the term "civilization" in a fairly narrow sense, not in a Matrix-like fever dream of robots harnessing human bodies for fuel.

In my scenario about the music supervisor on a major show bypassing human beings to get a cue (or at least moving the human labor from musicians to algorithm coders to save money), the economic model for creatives to be paid for their work disappears. It doesn't have to be malicious: it's just the production company saving money to get a higher return for shareholders.

For me, the distinction lies in the fact that with loops, virtual instruments, digital editing, samples, etc., it’s ultimately a human that determines the end result. And to be clear, I definitely think people already take that stuff too far as it is.

I’m sure some form of AI will eventually be able to generate incredible musical pieces, but I’m pretty much completely uninterested, because it removes the one thing that makes art meaningful to me personally, which is human creativity.

You bring up a good point here, of course. I'm not saying I know where to draw the line, but I do believe the line should be somewhere on this side of "no humans were involved in the making of this musical work."

Recently, it's been mastering, right? That's been a discussion point here for a few years now. Anyway ..the line has been drawn, and it's a slippery slope. Metaphorically speaking it's already engaged at 12db per octave. Prepare for AI to dial it in a lot more ..and pour a little extra grease on it for good measure. Hold on to your hats!  And thanks for putting up with my obstinance.


Getting back on topic


I'm actually thinking of picking up Solaria. I think that voice will blend pretty well with mine, especially if she's mostly doubling me. One of the things I'm paying close attention to are 'her' vowels, and I'm asking myself, are they like mine? I can't expect them to be perfect, so I'm curious to see if it's possible to alter formants  and other aspects of the voice.  It look like it's under $100.  Am I reading that right?

I'm older than you are, but I find myself at the opposite end of the issue.  I wonder that AI may offer the only way to save this fragile planet, and humanity along with it.  BTW, wasn't the 'singularity' predicted to hit around 2020?  Instead we got Covid.  Hmm.. Coincidence??  (I kid).

As for Solaria, Of course it's nothing more than a hyper-enhanced VI, but damn, it looks to be amazing, and the company already has 10 or so voices out there (only two of which sing in English, one male and one female).

So I just pulled the trigger and purchased Solaria and the upgraded V Studio Pro, which among other things is necessary in order to use multiple samples across a range of pitches. I'm looking forward to playing with it.. and I'll get back with my thoughts. 

I use the Kevin voice as a tool while making my songs. It doesn't appear in the final product. Still, useful tool.

I decided to give Solaria a whirl, mostly because I really want a female voice to supplement my background vocals.  I don't always like the sound of a chorus of me's. While I was at it, I picked up Kevin for an extra $70.  I mean, I bought the synth, so I might as well get a couple patches out of it, right?

It's quite intuitive to use. I pretty easily figured out about 80% of it watching youtube tutorials with subtitles.  After inputting notes and lyrics, right out of the blocks it sounds quite good but rather stylized, with lots of scooping for instance.  I had to edit it quite a bit to get it where it needed to be, and that included a fair amount of drawing pitch curves.  

Here's my first go-round with it. I covered a song by a songwriter I know from a different forum. You can hear S and K at the two bridge/chorus sections (@1:40 and 2:40) where I sang four parts and doubled two of them with Solaria, and two with Kevin, for a total of 8 tracks.

https%3A//soundcloud.com/m57/utopia-4-2

My first thoughts are it's ok.  Solaria's style is up my alley, Kevin's not so much. His bass note are laughably out of tune (hes's definitely a tenor) and he's a bit too broadway for my needs, but I was able to tame him enough to be usable for the project.  You actually have a surprising amount of control over a number of qualities of each voice.  Intensity, breathiness, vibrato, etc, and you can edit them either globally or by note.  You can type lyrics in phonetically, but it can be hit or miss. if you want precision, it's best to learn and use the ARPABET.

In the future I can see myself using this VI, but subtly, and probably never in exposed settings except maybe for effect.  I listened to demos of the other two female singers who sing in English and I don't think I'll getting them.

Getting in under the hood. Some of the editing capabilities are pretty impressive. There are at least a half dozen variable parameters that change the quality of the voice ..like gender, how breathy the singer, softness, tension etc, and they differ from voice to voice. you can set them globally or change them from note to note. The ability to re-draw pitch curves lets you finely control musical nuances like scooping and vibrato.

Perhaps one of the most surprising set of editing features involves the capability to edit the timing, duration and strength of phonemes on a note by note basis. To give an example, in the project I posted above I was finding that Solaria's default 'r' is harder (in a mid-western way) than mine when she sings the word "for." She is doubling the part I put down originally, and I wanted her to blend with me - not the other way around. When I typed the word "for" in on the note, the player automatically converted it into the phonemes (f ao r). To get her to blend with me I lengthened the duration of 'ao' and shortened the duration of 'r' while also bringing down the strength of the r. The sliders that do this are labelled with percentages, and the GUI superimposes the delineations between the phonemes on the waveform in the playing field. The only thing that could make it better would be if you could drag the delineation lines on the playing field.  From there, and just to get things perfect, I made a slight adjustment to the pitch curve, making it line up with new placement of the phoneme attacks, though now I'm wondering if the program did this automatically when it re-drew the note.  Honestly, I'm becoming more and more impressed with the range of stylistic adjustments you can make and how subtly you can tweak them.  Like I said before, it's not quite good enough for really exposed parts, but this technology is only going to get better.   I don't think anyone could tell with confidence that half of the singers on that recording are not real.

Yup, the music is really cheesy, but I have to say that for a synthesized vocal, that's extremely impressive. When they keep it in a normal range, it's hard to tell.....