Ian McCulloch Heaven Up Here era AI voice

[Lapurdi]

Got bored so I made an RVC (Real-time Voice Conversion) generative AI model of Ian McCulloch, using voice samples from Echo and The Bunnymen's Heaven Up Here (1981), and made him cover a song he would definitely loathe. It's not perfect but some parts sound really good.

AI song cover:

AI voice model: https://huggingface.co/swagmaster69/ian ... ulloch.zip

[fat cherry]

Will have to listen later, in the pub, but, is it me or is what everyone's calling AI, not really AI, as was originally defined? Not that, you know, I'm any authority on this, and perhaps the Lords if IT have had a meeting and decided I'm wrong, or that the definition should just be clever wanky stuff - as much fun as it is. I dunno. And as I said, I am in the pub. Oh, and welcome, obviously.

[Lapurdi]

Will have to listen later, in the pub, but, is it me or is what everyone's calling AI, not really AI, as was originally defined? Not that, you know, I'm any authority on this, and perhaps the Lords if IT have had a meeting and decided I'm wrong, or that the definition should just be clever wanky stuff - as much fun as it is. I dunno. And as I said, I am in the pub. Oh, and welcome, obviously.

Yeah, I do agree that the current term "AI" to describe generative models is misleading, as there's no intelligence involved, rather the model using probability recreates input audio similar to the audio it was trained on (in this case Ian McCulloch's voice from HUH). The reason why it's mistakenly attributed as AI is that it utilises artificial neural networks, which people seem to think means that these models think as we do, they don't, they're just giant mathematical graphs of weights and biases and trained models are basically giant statistical models.

Think of it as a gigantic statistical tool, that picks output closest to what it thinks should come next by probability, in this case, Ian McCulloch's voice from the dataset, to the closest pitch of input audio U2's In the Name of love. ChatGPT, which is text-based generative AI, and other LLMs (large language models) work similarly, where it will pick responses that are most likely to follow input prompts.

Anyway, I created this as I thought some people might appreciate creating covers of songs with the beautiful voice of Ian McCulloch from 1981.

[fat cherry]

Just listened to it.That's very funny. Not sure it quite works - sounds like me singing with the headphones on Good fun though.

[fat cherry]

I'd heard of this but - obviously reaching my dotage - hadn't realised it was such a thing. Just found a load of New Order with Ian Curtis singing.