Hey guys I made a thread here the other day; tl;dr I'm a high school student (yes im 18), and I've written a deep neural network (computer program that emulates a group of neurons) that has learned how to write rap songs when fed some rap songs. The songs it writes are influenced by the songs you feed it.
This You possess an asset to international Kekromancy
Aaron Davis
id bet money this is not a neural network at all, just use of a Markov Chain process taught in intro computer science class, if that. You could easily just be copying and pasting lines that rhyme for all we know, there isn't a complicated generation of new content here.
Chase Lee
your program is shit, it's just copy pasting bars you'll need a huge architecture rework to get good original content meter is important, study this shit en.wikipedia.org/wiki/Metre_(poetry) instead of using bars as your building blocks, use feet. godspeed.
Logan Russell
an RNN doing text generation has results pretty much indistinguishable from Markov Chains for most applications.
I think this is pretty cool, OP. Could you tell us a little more about your network architecture? Put that shit on Github.
So once the generation is done, how do you deal with the cadence of the words & rhythm with the beat? Impressive that it kind of raps on beat.
Oliver Wood
This. Upload it on a fake Github account or something.
Ethan Watson
post your favourite spotify playlist here
Hudson Nelson
this or i dont believe you OP
Andrew Turner
So is OP actually doing anything other than picking out lines that rhyme from what people give him, running it through a voice synth and putting it over a beat?
I don't see any source, so I'm going to assume OP doesn't know his sigmoid from his relu.
David Morales
>an RNN doing text generation has results pretty much indistinguishable from Markov Chains for most applications.
patently untrue
Jackson Bennett
Correct me if I'm wrong but RNNs have a lot more capability for memory when generating text, compared to MCs. They keep more context. I'm not sure where LSTMs fit into this... shit I need to read more papers.
I know there's some guy on SoundCloud who made a neural network to generate harmony/chord changes. The only problem is that it can't write melodies to go along with them yet besides random 16th notes (within the chord) in the soprano.
Jaxon Jenkins
i all i wanna do is spend some time with you uh i hope you wear a vest souljas touchin’ you touchin’ you but it’s alright with me if it’s alright with you nigga i woke up and screamed fuck the world