We're calling markov chains AI now?<p>Only the lyrics have been generated. Everything else you hear is just the musician, trying their best to sound like nirvana. No AI wizardry involved.<p>The lyrics sound less like an unreleased nirvana song, and more like a cut up of their previous songs. Which is to be expected when you feed a relatively small corpus of lyrics into a markov generator.<p>I get that this is just supposed to be fun. My point is that the title is misleading.
I was hoping the whole song/audio was generated. Looks more like fragments of existing nirvana lyrics re-arranged into a semi-coherent ordering. I recognize almost all the lyrics from existing songs.
It’s only the lyrics that are generated (and he’s just using markov chains?) , the music seems to be written by the guy doing them based on the article.<p>So barely anything to do with AI?
You can listen to actual AI-generated songs in the style of Nirvana with the OpenAI Jukebox: <a href="https://jukebox.openai.com/?song=807313705" rel="nofollow">https://jukebox.openai.com/?song=807313705</a>
I didn't like this as much as some other AI songs, notably Metallica [1].
It seems this model wasn't trained long enough as this song is the In Bloom/Breed riff with some lyrics and the general structure of Smells Like Teen Spirit.<p>[1] <a href="https://www.youtube.com/watch?v=zTG1Nb1hND0" rel="nofollow">https://www.youtube.com/watch?v=zTG1Nb1hND0</a>
Does anyone remember Microsoft Songsmith?
<a href="https://www.youtube.com/watch?v=wHduATM-o7M" rel="nofollow">https://www.youtube.com/watch?v=wHduATM-o7M</a>
Hey if you're interested in human voice synthetized and mimicked by AI check out this awesome project <a href="https://www.descript.com/lyrebird-ai" rel="nofollow">https://www.descript.com/lyrebird-ai</a><p>Overdub allows you to replace recorded words and phrases with synthesized speech that's tonally blended with the surrounding audio.