Project 3 – Craig Fahner – Ten Million Love Songs
For my project, I decided to generate love songs. I started with a model – the song “Their Hearts Were Full of Spring” by The Four Freshmen:
[youtube=http://www.youtube.com/watch?v=djDm2JVMm9Y]
Using the syllabic structure of this song, and using an aggregated database of love song lyrics, I created a program that synthesizes love songs.
————————–
agree with you. Being sung back would definitely make this seem more finished, especially if you could do some tune/melody markoving to go along with it? It might be tricky to get the syllabls/pentameter to match up, but hey, more fun computer science problems!!
Or you could just sing all of them and put out an album. That would also work.
lovely Rita, Metermaid, where would I be without you?
HAH. It really did change it once you sang it aloud. Your point of getting it performed really is more clear.
I definitely think having it sung makes a huge difference, otherwise it’s just more computer-generated poetry. I wonder if you could also generate the melody using a Markov chain.
This project is very cool! love the idea and implementation of it!
love it. That PDF file? I TAKE IT. MINE.
“And then he kissed me” seemingly appears in every song. good chorus (or title)
interesting, i always think about generating these cheesey songs as well, it would be cool to have an automated jukebox that plays them
You can use the Mac OSX speech synth as a singer: https://developer.apple.com/library/mac/#documentation/UserExperience/Conceptual/SpeechSynthesisProgrammingGuide/FineTuning/FineTuning.html#//apple_ref/doc/uid/TP40004365-CH5-SW3
+1 nice. Then its more familiar to audience — well, it’s more about using an existing, useful tool
————————-
So next step, make a song? AUTOTUNE!
Singing FTW! Golan should make it a requirement.
^ Phase two: Generate melody
A brief note about Markovizing melody: You want to have a high n-gram size to avoid sharp changes, but without an extremely high sample base, you’ll get exact replicas of sections from your sample.
Autotune ALL the lyrics! (the possibilities are ENDLESS)
10,000,000 Love Songs
— Wants suggestions on where to go with these experiments.
I wish you’d brought a guitar an performed one.
Do you know about Jonathan Feinberg’s Twitter-based “Found Haiku” project?
http://mrfeinberg.com/twaiku/
I like the analytical breakdown of the song’s structure.
I would love a MassageStart. Upday! Best day for Massage Start!
Still a little confused as to how the lyrics were separated out / the process by which you did so. Seems like there was a lot of thought going into it.
Really cool library! Now I want to go check it out
Riley Harmon has some links to extremely high-quality speech synthesis API’s (Ivona, etc – http://www.ivona.com/en/) and I have a 1982 Roland Vocoder upstairs (SVC-350, http://www.youtube.com/watch?v=cOO6xTXTeiA) — Golan
I think this is a good application of text generation / Markov models because I don’t know love songs have the same thing over and over.
Mechanical Turk: Get them to sing them back to you! It would be hilarious.+1++
Your voice is like a MassageStart on an Upday! You could be the 5th Freshman+1
It seems like your project can be applied as an useful application. I little hoped to see some aesthetics in the proejct though.
————————-
The lyrics are interesting. I think that the fact that they are not particularly improbable is great. I love the idea of these being sung by a computer voice. It could be very funny.
I am confused- did you play these generated songs with a voice? see, you should sing it. hahahah
I think it’d be great to hear these with the rythm, but reading them and following the “narrative” that way is also really interesting. So that leads to, possibly..karaoke?? If you want to go that direction.
i dare you to sing all of them!
Was worth it, just to hear you sing dude.+1
Definitely have these sung by Mechanical Turkers, then correct their pitches with Autotune or a vocoder! Awesome!
————————-