Hacker News new | past | comments | ask | show | jobs | submit login
A Teen Programmer Built a Tool to Generate Fake Lyrics for Your Favorite Artists (genius.com)
65 points by amasad on June 3, 2019 | hide | past | favorite | 48 comments

I'm a bit more impressed with this article. This person posted the project to reddit 14 hours ago [https://www.reddit.com/r/hiphopheads/comments/bw6xs7/lyricsr...] and already here's this professional looking write up generating additional clicks...really makes me wonder how much residual income a site like reddit creates. Reminds me of remora fish somehow..

This seems to generate mostly word salad. I haven't been able to find an artist (even one of the recommended artists like Drake) that it generates intelligible lyrics for. It seems to grab existing lyrics from other songs and mash them together, but it's not doing a great job of it.

The only reason this is on HN is because a teenager made it

and so that HN'ers could heap scorn upon it while claiming that they could bang something out like this in a weekend.

In this case though, I'd believe them.

Markov chains alone aren’t really the best solution for this, but you can find a few interesting snippets here and there in the output

Yeah, it just strings words together as far as I can tell.

He says he used Markov Chains - which is just stringing words together based upon probability.


Here's one, that uses OpenAI's GPT-2 to generate: http://billion.dev.losttech.software:2095/

Partial write-up: https://habr.com/post/453232/

Took about 2 weeks to make, most on the website part. So if you want to teach your kid programming and some AI, it's not so hard.

That was interesting, and if it had been an actual song (with a little bit of cleanup) would have made for interesting interpretation, depending how it was approached by music and singer.

Then you get something like this: http://billion.dev.losttech.software:2095/song/3971156712


I believe you can co-write music too. My original plan for expanding the project was to train on MIDI-karaoke files, and folks from OpenAI recenlty proved that actually works for MIDI with no text: https://openai.com/blog/musenet/

Having lived through the time when ‘official’ lyrics weren't widely available and having thus come upon a bunch of fan attempts at deciphering Kurt Cobain's songs, complete with advanced cases of Looking for Deep Meaning syndrome, I now see no attraction in a mechanistic approach to the phenomenon.

Nothing special. Just good ol' Markov chains.

Maybe nothing special, but pretty well executed by a teen!

I mean with the plethora of ressources available today, anything done by a teen today is not as impressive as something done by a teen say 10 or even 5 years ago.

As someone who didn't do much of note during his youth this is still impressive to me and I'm happy to encourage the author if I ever meet them.

I would disagree.

Whenever I see statements with absolute statements like anything, nothing, always, never, I question them.

We all stand on the shoulders of giants. So what we do is impressive for when we do it. What those in the past do is impressive for when they did it.

Every time you see those statements?

We ALL stand on the shoulders of giants?

That's a really sad way of looking at things

Why is that?

It's not a "way of looking at things", it's a statement. A statement cannot be sad, and even if a statement makes someone sad, that doesn't make it false.

edit: yeah, figures. When calling out ElsaGate, I got replies like "your argument is literally 'think of the children'". But in this thread it's all so heartwarming, or respectively turning statements that aren't dripping with soppy general statements that mean nothing into an "attitude" or "way of looking at things" (which remains BS, and reaching 500 karma and finding the downvote button doesn't make that false, either)

My current "flow" music is the Hamilton musical, so I threw in Lin-Manuel Miranda into it.

The entire result was hot garbage.

However, there were buried segments that were good, but I'd have to lay that on the tightness of Mr. Miranda's lyrics as we're dealing with Markov chains.

All in all, that was a few minutes of amusement.

This is awesome, very cool! You mentioned OpenAI in another post - it's not that hard to use the GPT-2 model. Check out https://github.com/minimaxir/gpt-2-simple for a quickstart. Would be curious to see how the music generation goes if you re-train your data set against gpt-2.

There's a version of that I posted just above in this thread: https://news.ycombinator.com/item?id=20086483

Brilliant. But doesn't quite work with older generation artists! A snippet from Elvis:

"Just you in the time for it's just a rabbit

Let's rock

Happiness will"

Some meta:

The cynic inside me thinks that people’s negative responses here are not really about corrupt media (tm). They are just reflections of the contrarian attitude that runs rampant on this site. The fact the HN guideline has to explicitly forbid 'shallow dismissal' tells you how much compulsive negativity is happening on this site.

(And yes, this post is also a display of my own negativity.)

I am not defending the media in any way. This is low quality journalism (on a site that does not seem journalistic anyway). The achievement of the 16-yo in the story is not outstanding in this field. HOWEVER, instead of reading yet another comment saying 'meh Markov chain', I guess I’ll prefer something like more advanced materials on text synthesis, some related projects, or just some advice for the author would be nice.

He basically reimplemented Dissociated Press.

Credit where it's due, that's more than most 16-year-olds ever manage (though well within the power of a 16-year-old with a strong interest in programming).

"Weird Al" Yankovic ...

"DOCTYPE HTML / if lt IE 7 / html class no js IE6 oldie!"

That polka lyric jams.

Pop music is already pretty algorithic

there is nothing new in that , even a simple rnn can recite lyrics that good. But given that it was built by a "teen" it is good. media people are obsessed with highlighting teenagers anyway.

There should be a life rule, if the word "teen" or "child" appears in any headline about some new "invention" or "app" you should immediately ignore that article as it's going to be some breathless gushing bullshit.

Ok, but this isn't any better, and when it's upvoted to the top of a thread—which indignation usually does—then it just sits there steaming and giving off nasty fumes.


Thanks for that encouragement /s

I think it's great to encourage students like Samarth and I by sharing and talking about what we're building. You can offer specific and helpful feedback which I would really appreciate. But I don't appreciate your calling my work bullshit. And I think this attitude chases away less privileged students from pursuing things like software further. There's value in making something that people get joy from, even if it's not the cutting edge shit that OpenAI blogs about. And I'm confident in what I made here.

They didn't call your work bullshit, they said the article would be bullshit.

Weeeell... strictly, yes, but actually I'm inclined to think not. The tone was sweeping, not to mention disrespectful, enough that I think a broader interpretation is warranted. By claiming such articles are only worthy of being ignored the implication is clearly that the object(s) of such articles are also unworthy of interest.

The comment wasn't about you, it was about articles written by people who want to try very hard to Be Encouraging, an affliction which inhibits adults from seeing children and, in advanced cases, minorities, as people, instead of Things Which Need Encouragement. The end result is articles which are incapable of talking about the achievements, and instead talk about the achievers, as if "encouragement" consists of having your life story plastered on the nearest piece of media.

From my perspective, it's the article calling your work bullshit: if the author thought it was interesting on its own merits, they wouldn't have to lead with the age of the creator.

I would bet that OP was once themselves the recipient of similar "breathless gushing" when they were young, and lost a lot of the wind in their sails when formerly impressive achievements became simply expected. Better not to depend on that kind of validation, it stops early and never comes back.

And worse, someday you might read articles like this and feel discouraged, because it seems nobody will care that you just discovered how much fun you can have with Markov chains in your 30s.

My specific and helpful feedback to you would be to not immediately feel attacked if someone criticizes something tangentially related to your work. OP never said your work was bullshit. They said the article/coverage of your work is probably 'breathless gushing bullshit'.

OP was actually wrong with their hueristic in this case - the article doesn't really go into your age beyond a passing mention, but that doesn't change the fact that OPs critique was with article writers, not programmers-who-have-their-work-written about.

Point taken. Thanks for noting.

Still feels like HN generally has a counterproductive attitude to stories of this type.

Oh, 100%. Unfortunately the best strategy I think is to just ignore the mindless comments and pick and choose who you are going to have a conversation with.

Personally, I think your project is getting coverage not at all because of your age, but because it produces interesting results that people want to see, regardless of whether the algorithm used is considered cutting edge or not.

It's getting coverage [on the website of the company that provided the API that was used for the app] _because_ of the age , that's why the age is in the headline, it's a gift-wrapped marketing win.

This isn't to denigrate the work of either of the two kids involved: the program produced is interesting and fun, the encouragement is awesome, and they've lucked out with this publicity and should milk it for all it's worth.

The fact that you have the courage to share your work speaks volumes about your ability to take risks and receive feedback.

Some feedback that will help you as you grow older: grow a thicker skin and stop taking everything so seriously.

Yeah I was expecting something actually interesting - but nope, markov chains.

Just because you learnt them long ago doesn't mean it isn't the most interesting thing in the world for someone just learning it anew

Yes, but doing tutorials doesn't really generate news headlines.

Except if they're well-heeled American dropouts at an elite school. Then the world is their oyster.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact