It’s been many many years since I posted anything here.
Curious and somewhat conflicted about AI vocalists, but with particular interest in the capabilities of “Solaria” I decided to get to know the technology and subsequently revisit some old “girl” songs - past co-writes, some actually dating back as far as 20 years, and put “her” to the test. Way back in the day I was often approached by varioius women from various online songwriting communities to put music to their words/poetry. This then is my 2023 re-creation of some of those distant past demos, some of which never actually saw the light of day at the time, but now, compiled in a new album I’ve produced in C12 and called “Soularia”, all sung by AI Solaria (lite) by Dreamtronics. Make of it what you will:
A follow-up “Soularia II” is currently a work in progress.
One thing to bear in mind is this is just the free ‘lite’ version of Solaria you’re hearing- I believe the full version has additional performance capabilities - including a VSTi plugin.
I was a little hesitant about sharing these given the conflicting and polarised views regarding AI generally - it seems to trigger a lot of often very heated debates. For now I see it as just another production tool - a useful one for anyone that doesn’t have access to a vocalist. In much the same way I’ve been using Jamstix VSTi regulary, an AI drummer substitute that’s been around for quite some years now. All I need now is an AI bassist, guitarist, keyboardist and perhaps a full orchestra of AI players as well - oh, and a complete AI replacement for myself would also be real handy.
It is a very interesting tool and your demonstration proves it. I don’t want to argue about the use of AI. For me, at least for now, AI-enabled VSTs are just tools like any other tools available to us.
You may have the best guitar, the best piano or any high-end instrument paid for at a high price, that does not mean that you will be able to get the most out of the sound available with such instruments.
It is the user and the creator who creates the result, it will be good, passable or bad.
If the AI allows me to obtain interesting results because I know how to use it well, that is what is important. The use you have made of it is really very interesting and well done.
Thanks! Glad you liked that one… written around 20-odd years ago!
As it happens, I’ve just finished (I think?) a second follow-up album, again old songs, some that were discarded, now resurrected and given a new lease of life:
Gotta admit, I don’t know how I feel about this. But I like the track anyway. I think we’re all trying to figure out what to do with AI. I mean, suppose I asked an AI to play a guitar track on one of my compositions, but it sounded better than my own. Well, I would definitely use my own, but I would think a lot about what I like better about the AI track. Also, I don’t think an AI will ever be able to write a piece like I’ve written, but that might actually be a good thing!
I only discovered AI ‘singers’ myself earlier this year, but I was curious so thought I’d give them a test drive. The first one I tried was Vocaloid - and I actually wrote a song specifically to test drive that one.
But I soon then discovered “Solaria” which I think is much better. She/it is programmed in an app called Synthesizer-V. I’m using the free version. I think the pro paid-for version actually has a VST plugin which would presumably make the whole process a little easier working within Cubase.
I think it’s fairly easy to use - there’s a learning curve of course, but plenty of Youtube tutorials etc out there to help. It didn’t take me too long to get reasonably familiar with it.
The way I approached it was to use variaudio on the original demo vocal. I
pitch-quantise that vocal, and then rendered a MIDI track from that which I then imported into Synthesizer-V. Cleaning up that MIDI, which invariably ends up with a large number of unwanted events and messed up note lengths etc actually takes significant amount of time! Overall, it does take quite some hours to get a decent vocal done and obviously a human could do it a whole lot faster! But, in that regard it’s very much like many other processes where we rely on manually programing up a performance in some virtual instrument, orchestral libraries etc etc - it all takes a lot of time to create a convincing performance.
So, if you don’t have access to a human singer then this AI session singer could be an alternative worth considering. As with most things, I’d ideally prefer the real thing of course, but like many here, we invariably have to rely heavily on VSTi’s etc so I guess AI vocalists are just another virtual “instrument” we now have at our disposal.
I didn’t know where to post this, so I’ll have to ask you Sherz and here as well. How do you mix/master with vocals? I had a bit of hard time, been to the master of the instrumentals, then a post master with both instrumentals and the vocals which I reduced the stereo out so it doesn’t clip, then a final master mixdown. Is this correct?
"I think Solaria was a good choice. Check out what I did with it."
Your link goes to an instrumental track. I assume you meant to refer to the 2nd track Valentine & Gaston.
Vocal sounds quite small, thin and rather swamped by the brash sounding instruments to me. For mixing tips and inspiration I suggest Youtube is your friend! It’s a BIG topic - too much to cover here!
