When making tracks I always get stuck against the fact that I can’t really sing and vocal sample packs are generally awful.
I’d like to find something to solve it.
One option would be to use a voice transformer plugin like Baby Audio Humanoid or Minimal Audio Evoke so I can either make my terrible voice or cheesy samples less awful. Another would be go full AI and get a subscription to something like Audimee which creates AI vocals for you.
I’d be really interested to know if anyone has gone down that road and what worked best for you?
Ace studio. It users machine learning ( I know, fuck ai) but its vocal models are ethically sourced (they work with vocalists for training data instead of stealing data from everywhere). You can sing or hum melodies (or record or import the MIDI) and type the lyrics (there’s a learning curve and you got to get used to typing phonemes). It’s pretty flexible. Gives you full control, none of that “prompting” a vocal like a talentless loser; you actually need to do the work to get something for your song. And on top of a bunch of different singers they also have some instruments, but I’ve never tried those.
My daughter loves Miku so she would be very interested if I could make some Vocaloid tracks. It’s also a one off purchase with a month free trial which is cool.
I had heard it was very hard and time consuming to use, but I don’t know if that’s true.
Ace studio looks very cool, but it’s $16 a month billed yearly so $200 which is a lot for a subscription to something I might not use all that much.
It reminded me of “Voice swap” which is the site Benn Jordan is affiliated with. I know they make a big thing about being ethical.
The pricing there is a bit more reasonable, but perhaps a little overcomplicated. It’s trying to be fair to artists, but in a way that is a bit confusing for the end user.
Theoretically you can, but the most basic license is $400 and given that it runs in the cloud it will stop working if the company goes bankrupt which given they are probably VC funded is more likely to happen than not.
You could just hire someone on fiver or post an ad locally to hire someone. This isn’t really a new problem, most of the old school producers didn’t have a dedicated singer they worked with.
I haven’t used any of these AI tools but they sound pretty bad (to me) in the demo videos on the sites. Maybe they will work for your sound or workflow though. They all seem really time consuming to get mid quality results out of.
I’m looking for a tool to help with songwriting. Having a feedback loop of having to send a track to an online rando and getting a result in a few days isn’t going to be a substitute for something I can quickly iterate on, but it might be a good option if i want to make a demo into a finished track.
I’m also planning to chop and heavily effect everything so it doesn’t really matter if it sounds 100% authentic.
So having done a bit of research I’ve come to the following conclusions:
I’m not super interested in most of the AI services that rely on the cloud . A lot of them want you to sign up for a year and given how quickly AI is moving they might be obsolete or out of business in a year.
The voice swap stuff is not actually that useful. Most of the examples are of someone who can sing really well turning into someone else who can sing really well, but if that’s not you then it’s really just changing the timbre of bad singing. It’s a bit like something that could make your Strat sound like a Les Paul. That’s cool,. but if you are a sucky guitar player it doesn’t help.
Much as I admire what Benn Jordan is trying to do, the terms and conditions of voice swap are so confusing as to be borderline hostile.
Evoke seems to be a lot more powerful than Humanoid. It was a fairly easy decision to go with that. I like it a lot, but it does seem to glitch quite easily if it doesn’t like something in the vocals.
Synthesizer V is incredibly impressive and not all that expensive compared to Vocaloid. This is probably the closest to what I’m looking for.
The only limitation is that it’s a bit of a DAW in a DAW which I think is fine for ARA compliant DAWs like Logic, Cubase, and Studio One, but if like me you use Bitwig or Ableton the integration is going to be a bit annoying.
There’s a fun option of running vocaloid type plugins into a voice changer service as well. Good vocals from the synth plus added flavor. I know the Auribus has a blend control and interesting features like choir type effects
I know, fuck ai) but its vocal models are ethically sourced (they work with vocalists for training data instead of stealing data from everywhere). You can sing or hum melodies (or record or import the MIDI) and type the lyrics (there’s a learning curve and you got to get used to typing phonemes). It’s pretty flexible. Gives you full control, none of that “prompting” a vocal like a talentless loser; you actually need to do the work to get something for your song. And on top of a bunch of different singers they also have some instruments, but I’ve never tried those.
