Using AI for vocals?

[avatar]

Thought I'd share an interesting experience that I had this week. I'd posted one of my 50 90 lyrics (Circus of Ghosts) on another music collaboration site, and a musician picked it up and created some nice music for it, which left us needing to find a vocalist. I got a bit of a surprise when a male musician uploaded a very nice-sounding female vocal. It turned up that he'd used synthesized vocals, something called "Synthesizer-V Solaria".

I guess all the vocalists on here may be horrified, but I was surprised that the technical capability exists to paste in lyrics and get a reasonable-sounding vocal out the other end. Wondered if anyone else has come across that before?

P.S. I suppose that I really shouldn't have been too surprised. Because in my line of work, digital communications, we have found that promotional messages generated by an AI software nearly always outperform the messages created by humans! Which is definitely scary

[avatar]

😮 @headfirstonly Have you already stumbled over "Solaria" and "Kevin"? Maybe something for your Twitch show? Solaria had a contest over summer to let people show off what can be done with it: https://www.eclipsedsounds.com/solaria-original-contest. There are also some people who show these off on YouTube by covering known songs, such as https://www.youtube.com/watch?v=rhGDgrhdrZg and https://www.youtube.com/watch?v=EQIbsz69o_4. This topic still is pretty niche, though. Maybe because their website https://www.eclipsedsounds.com/shop looks like a relic from the 1990s? 😂

[avatar]

Sounds interesting - I'll take a look, thanks!

[avatar]

Thanks @florianhoffmann . Those videos are very interesting showing the way that a musician would get the vocals to match to the music

[avatar]
@brisk

Hey, that's amazing, @florianhoffmann! Thanks for drawing my attention to this. Because I was curious, I downloaded Synthesiser V Studio Basic, and the accompanying "Tsurumaki Maki Lite" vocal library (also free) this evening and made this little 1 minute demo thing to play around with it. I don't really get how to use its features or anything but it was relatively intuitive for me to come up with this, and I love how it sounds. https://yeslessness.bandcamp.com/track/emptiness

[avatar]
@yam655

Thanks for mentioning this @lyricslinger! I played with it using the MIDI I made for my cover of "All Aboard for Podunk." (A hit song in the 1890s.)

It was really easy to use to get a basic sound, though I'm sure I could fiddle with it if I wanted something fancier.

The question now isn't whether I'll be buying it. (It's easy to use and affordable, after all.) The question is whether I'll actually use it in February.

[avatar]

I want to say @cola has used digitized voices in her music before (including a digitized voice she created from her own voice!)

[avatar]

@lyricslinger @brisk @yam655 Goodness. This thing is amazing. And scary. I tried out their free demo and was thrilled. So yesterday bought their package of the synthesizer and the Solaria database. Today I put together a first song: https://soundcloud.com/florian-hoffmann-music/nothing/s-UVwK5qFUsZi This is a revised version of https://fiftyninety.fawm.org/songs/11407, with lyrics and vocals. I entered the melody line and let it sing la-la-las. Then I started making up lyrics while it sang to me, experimenting with the words and their sound. I didn't touch any of the advanced features, tons of timbre and vibrato sliders. Only words that span multiple notes need some tweaking to distirbute the syllables correctly over the melody.

[avatar]

Hey all, just popping in to say there's another free one that I use a lot - https://www.plogue.com/products/alter-ego.html
The free characters don't sound as good as the ones you all are using, and you gotta put in a lot of effort to pronounce certain stuff. But I'm quite fond of Bones and Marie Orc as they were my "singers" before I had the nerve to ask humans to sing songs for me lol

[avatar]
@yam655

@florianhoffmann Using a + (plus) will cause the system to automatically split a word over notes while still sounding natural. Using a - (minus) will extend a syllable over multiple notes. ("goodbye" in one note and "+" in the next will split the word as expected, while "goodbye" in one note and "-" should extend the "bye" syllable.)

I had trouble finding a phoneme list so I could tweak the pronunciations. Eventually, I found: https://pdfcoffee.com/synth-v-english-phonemes-list-pdf-free.html

[avatar]

@yam655 Wow, cool! Much appreciated, thanks! I had tried splitting words with hyphens in the usual places. Didn’t think it would split automatically when putting them at the end. Most phonemes can also be found interactively by entering a similar word. Started experimenting with the sounds, and it’s impressive how malleable it is. The voice is obviously designed for “grand” stuff, like musicals, or epicore, https://soundcloud.com/florian-hoffmann-music/caraval/s-AnkmXYK4UTf but it works in a large variety of styles.

[avatar]

@lyricslinger YOU SAID: "AI software nearly always outperform the messages created by humans! Which is definitely scary" - terrifying and totally incomprehensible to me. I hate those informative videos on YouTube that use an AI generated voice - it sounds so soulless.

@brisk That "Emptiness" is spooky. I am guessing that we don't instantly recognise it as a non-human voice because we are all so used to hearing "autotune" used on human voices. The music industry is going to love this. A whole AI generated band and vocalists mean 100% profits.

@florianhoffmann Congratulations on your song "Nothing" - though the implications of its success terrify me. I asked my wife what she thought of the vocalist and she said. "She's OK. But nothing special!" - when I revealed she wasn't human, her jaw dropped.

[avatar]
@mikeb

I downloaded the (free) Solaria stuff the other day, thanks for the tips on the + and -