Simon Jaeger on Nostr: I stretched AI audio tools to their limits today. In 2011, Taylor Swift did a live ...
I stretched AI audio tools to their limits today.
In 2011, Taylor Swift did a live cover of Train's "Drops of Jupiter" (Tell Me) during her Speak Now world tour. It was never recorded in studio. So I decided to see whether I could turn a live recording of a guitar cover from a crowded stadium full of screaming fans into a clean and listenable studio vocal that could be mixed over the original instrumental. This required heavy use of Demucs and Goyo, and I also had to clean up the ambiance track, stick an aggressive high-pass filter on it, and mix it back in so I could resurrect some of the sibilants. And being a live guitar cover, it wasn't done at any particular BPM, just whichever one sounded good--which actually turned out to be quite a few, ranging from about 150 to 166. If I finish this, I'd rather find a good guitar instrumental, but for now, here's a 20-second sample of the source material plus the first verse and chorus. I think this is a pretty crazy example of what AI can do for audio in 2023. I can only really take credit for adding the stretch markers and doing some very basic mixing.
https://media.dragonscave.space/media_attachments/files/110/508/051/651/683/407/original/db32653bc3af13cd.mp3
In 2011, Taylor Swift did a live cover of Train's "Drops of Jupiter" (Tell Me) during her Speak Now world tour. It was never recorded in studio. So I decided to see whether I could turn a live recording of a guitar cover from a crowded stadium full of screaming fans into a clean and listenable studio vocal that could be mixed over the original instrumental. This required heavy use of Demucs and Goyo, and I also had to clean up the ambiance track, stick an aggressive high-pass filter on it, and mix it back in so I could resurrect some of the sibilants. And being a live guitar cover, it wasn't done at any particular BPM, just whichever one sounded good--which actually turned out to be quite a few, ranging from about 150 to 166. If I finish this, I'd rather find a good guitar instrumental, but for now, here's a 20-second sample of the source material plus the first verse and chorus. I think this is a pretty crazy example of what AI can do for audio in 2023. I can only really take credit for adding the stretch markers and doing some very basic mixing.
https://media.dragonscave.space/media_attachments/files/110/508/051/651/683/407/original/db32653bc3af13cd.mp3