The technology surrounding the Voyage project has got me thinking.
I wonder how you would feel about the prospect of being able to sample at a deep, detailed, micro, level - at the level of vowel sounds - Agnetha and Frida's singing voices, taken from their solo and ABBA recordings? Imagine that by doing this it would be possible to 'construct' a realistic-sounding duet 'performance' and recording of Agnetha and Frida 'singing', say, I Know Him So Well. Not to mention other possibilities!
Do you find this idea a turn-off, or would you welcome it? I view it with a mixture of being intrigued, excited, and feeling slightly disturbed.
richard , I would welcome it, but with some reservations.
What you're talking about is vocal synthesis - as for "deep, detailed, micro, level" I'll assume you're leaning toward the machine learning side... deepfake, and not the VOCALOID-esque iteration (that one with robotic inflection)
Within today's technology, it has yet to be perfected. It's possible but it'll have to be pushed out of the limitations to get to that.
Machine learning based vocal synthesis is mostly rough, but could be winged out with some fine-tuning, though not convincingly enough so far in this century.
A YouTube channel called Vocal Synthesis has attempted to grab as many vocal samples available to get a profile of Frank Sinatra's voice just to sing a handful of songs he'd be unlikely to sing. (Dancing Queen)
The result is mostly rough; not even fine-tuning by others could save it in this capability. That channel is also a proof of concept that it could handle conversational voices well (and we'll never run out of Frida and Agnetha saying "Hello California, we're here!" or "Bonjour Paris".)
There's possible scenarios that can reach to the point of uncanny valley. One of them involves a large number of vocal samples (good if it accounts for rising tone, falling tone, speaking Swedish, all of that), a better model built on it, and something groundbreaking yet proprietary (a better learning model or algorithm, I suppose), or the good old reliable open source.
Given the abundance of recent internet memes using models of vocal synthesis to make a joke come to life, one would assume that the benefits outweigh the risks. I fall into that camp where indeed it outweighs the risks - as long as there is solution to detect and distinguish synthesized voice from recorded, pure voice.
There's so many debates regarding this, and it's intriguing for a phenomenon that just took off years ago.
I would also like to comment that ABBA or Polar Music would make a smart move to protect their voices, if they do use this vocal synthesis technology. x
Thanks, Jonathan. Your response was far better than my cursorily set-out post deserved! You've covered it far more capably than I could. Of course, the technical challenges of getting a convincing artifact of their voices is difficult enough, let alone the copyright and ethical issues you've pointed out.
I feel that in the future anything will be possible. Do I think that means anything should be done? No. But my opinion doesn't and won't matter because it will happen anyway. I believe there's a cut-off point but who decides where that is? Not everybody is keen on the Abbatars. Many have quite anti views about it. But still it's here, the genie is out of the bottle. I'm all for the Abbatars but that doesn't mean I think it would be good to hear Agnetha and Frida sing I Know Him So Well using vocal trickery. It would sure be interesting but it sits uneasily with me.