One thing that has been amazing me recently is the current state of vocal synthesis. It seems to have improved rapidly in the space of just a few years, thanks to neural network models.
For example, here's a quick clip from a synth I've been using in recent weeks:
In light of what we learned with vocaloid in Japan, is this even meaningful? Singers don't make the song. You can have crowds fill stadiums to watch a hologram, it has already been proven.
To boot, isn't this the fidelity vs fun argument of videogames again, too?
Madonna's "Ray of Light" is still a good song without her; William Orbit is a master producer after all. I hope we see a refocusing away from personalities to producers out of this.
It depends on the audience, and the producer. Some people are content with listening to robotic-sounding Vocaloid type synths, others prefer something closer to an authentic singing voice with the more complex range of emotion that can be expressed.
Even just from a technology point of view, I find it fascinating to see how the envelope is being pushed via machine learning models.
>Some people are content with listening to robotic-sounding Vocaloid type synths, others prefer something closer to an authentic singing voice with the more complex range of emotion that can be expressed.
Let me just point out the nuance that many people, including myself, appreciate both types of music.
Thanks for sharing this. The Solaria AI video you linked to is impressive. Some of my favorite EDM tracks have guest vocals in them. As an electronic music producer and a terrible singer, it's mind blowing to think you can now produce an instrumental beat and attach realistic sounding AI generated vocals to it.
Prior to this, I'd been experimenting with pitch-shifting the output of Google's Text-to-Speech service (https://cloud.google.com/text-to-speech), which was quite interesting and somewhat pleasant sounding for spoken word vocals, but didn't really have the range or expressiveness I was looking for more generally.
For example, here's a quick clip from a synth I've been using in recent weeks:
https://vocaroo.com/125tAD40Gl6d
It sounds almost like a real voice, and that's just using the default settings, without any effort to tweak it further to improve the realism.