How to Generate Music - Intro to Deep Learning #9

We're going to build a music generating neural network trained on jazz songs in Keras. I'll go over the history of algorithmic generation, then we'll walk step by step through the process of how LSTM networks help us generate music.
Coding Challenge for this video:
github.com/llSourcell/How-to-...
Vishal's Winning Code:
github.com/erilyth/DeepLearni...
Michael's Runner up code:
github.com/michalpelka/How-to...
More Learning Resources:
medium.com/@shiyan/understand...
mourafiq.com/2016/05/15/predic...
magenta.tensorflow.org/2016/0...
deeplearning.net/tutorial/rnnr...
maraoz.com/2016/02/02/abc-rnn/
www.cs.cmu.edu/~music//cmsip/s...
www.hexahedria.com/2015/08/03/...
/ algorithmicmusic
Please Subscribe! And like. And comment. That's what keeps me going.
Join us in the Wizards Slack channel:
wizards.herokuapp.com/
And please support me on Patreon:
www.patreon.com/user?u=3191693
Thanks Ji-Sung Kim for the example code:
deepjazz.io
Follow me:
Twitter: / sirajraval
Facebook: / sirajology Instagram: / sirajraval Instagram: / sirajraval
Signup for my newsletter for exciting updates in the field of AI:
goo.gl/FZzJ5w
Hit the Join button above to sign up to become a member of my channel for access to exclusive content! Join my AI community: chatgptschool.io/ Sign up for my AI Sports betting Bot, WagerGPT! (500 spots available):
www.wagergpt.co

[@jackglendenning8509]

Siraj you're videos are getting better and you're starting to explain concepts so that a wider audience can understand! Just started second year of comp-sci degree but been doing ML on the side for 6months, your videos always inspire! I appreciate your work.

[@SirajRaval]

Thanks Jack! Keep it up!

[@KapilYadav-90]

As i can infer from your videos you are multi talented. you sings, you dance, you learn, you read and write, you make youtube and watch all the movies at the same time. My question to you is: DO YOU EVEN SLEEP, IF YES HOW MUCH?

[@SirajRaval]

Thanks Kapil! Lol i do sleep. about 6 hours a night. I JUST LOVE MAKING MACHINE LEARNING VIDEOS I WILL KEEP CRANKING THEM OUT :)

[@tastelesstouch]

The memes in your episodes are the best!

[@kapilsharma1704]

I don't know ,what all you have learned.but the way you presents is terrific and can beats anyone out there

[@infantraj9951]

Siraj your intro was well edited. Keep up the good work!

[@SirajRaval]

Thanks Infant!

[@phuccoiinkorea3341]

love this channel so much

[@SirajRaval]

Thanks Phuc!

[@chrisoffersen]

I don't know how you do it Siraj. You make this stuff very accessible. Many thanks, once again.

[@SirajRaval]

lots of sacrifice np thx

[@WiiiLo100]

This is Great!

[@michaelbowie8315]

Loved the LSTM explanation.. From 4:38 to 6:38, you could totally make this a separate video... "LSTM 100% Explained in 2 Minutes"... I have been reading a bunch of papers, articles, etc on LSTM, and this is by far the most intuitive explanation (for me at least). I wish I had seen this earlier, but it was hidden inside of this video.

[@wobanda5986]

I never understood his any video till date😇 (I watched his all video), but some day this will make sense to me that I am sure🙂

[@SirajRaval]

Yes it absoltuely will

[@rafaellazo1725]

love your videos Siraj

[@larryteslaspacexboringlawr739]

thank you for music deep learning video

[@renanlopes9144]

Am I the only one who thinks those explanations are way too fast to understand? I know the video is intended to be short, but how about 15-20 min? Good job nevertheless.

[@timschafer2536]

Renan Lopes agree

[@yellowbeard1]

He has other videos where he dives into the topic at more length. I agree that if you only look at this video it is inadequate for learning a topic like this but think of this as an introduction and overview, his in depth how to do this video is about an hour long.

[@WilliamDye-willdye]

It doesn't matter much, but I've always heard "MIDI" pronounced as "middy".

[@SirajRaval]

you're right thanks Will

[@ConTejasMusic]

that LSTM beat was magical!

[@SirajRaval]

Thanks Tejas!

[@BrianFaure1]

dude you're channel is like the Rosetta Stone of machine learning

[@SirajRaval]

Thanks so much Brian!

[@pakigyatuladhar]

so true!!

[@easter.bunny.6]

Siraj YOU ROCK!! nice work!!

[@SirajRaval]

thanks Kevin

[@easter.bunny.6]

but plz maybe do a tut about polyphonic music? or even how to transform your modling in mind to mathmatical modeling?

[@Towzeur]

i have trouble to install preprocess library Traceback (most recent call last): File "setup.py", line 28, in import preprocess File "lib\preprocess.py", line 264 except Exception, ex: ^ SyntaxError: invalid syntax

[@HelloThere-xs8ss]

Hey, I'm from the future. We need more veggie gardens. amazing how you do it

[@MatthewHallberg]

vishal batchu, wins everytime.

[@jboss7820]

MatthewHallberg lol

[@ayushkarnawat6817]

its because he is probably already a data scientist at some big company and has a lot of experience

[@chandrashekard.7543]

MatthewHallberg plot twist: Vishal was Siraj all this time. fight club style.

[@Erilyth]

Haha :'D, I'm just a student learning how to use DL :)

[@MatthewHallberg]

hahaha that was too good.

[@aminejadid2702]

awesome !

[@SirajRaval]

thanks!

[@llumino1455]

SIRAJ CAN YOU PLEASE PUT THE CODE IN THE DESCRIPTION OR ON GITHUB

[@FastestPodcastClips]

Where do you get data_fn from ??? NameError: name 'data_fn' is not defined

[@samuelvarghese6881]

+Siraj Ravel Nicely explained about lstm man keep it up. Btw you know when you say batch size is 128 in this instance, how do you do that with your own data from like a excel sheet.

[@md.shahidulislam3178]

Hey Man why i love your video so much ;) !! >>> Nice Explanation Every time

[@SirajRaval]

Thanks!

[@dmitrykarpovich7579]

mid E! Love your videos man!

[@NightmareTV666]

Are the results dramatically better than RBM with a GIBBS sampler? And is the CNN parametric approach the best, but potentially intractable in higher dimensions?

[@Frankthegravelrider]

Siraj, your T-Shirt is pretty good - your videos are alright too ;)

[@SirajRaval]

Fr Ya! Good to see you. Thanks!

[@modelworkzseo]

Lol, it's pronounced "middy". Great video man, keep em coming!

[@abhiksingla7274]

Siraj, you are doing a great work! Can you do a video on Image Captioning using Deep Learning?

[@SirajRaval]

Thanks Abhik! I have one called build an AI writer, more soon

[@bjyr]

it's lit!

[@SirajRaval]

thanks Bjorn!

[@MichaelRicksAherne]

"Middy" not "Mid-eye" Otherwise flawless. :)

[@WillKriski]

also meh-theeny :)

[@SirajRaval]

lol thanks!

[@aidangomez4726]

yo man your channel is blowing up

[@SirajRaval]

yes needs to grow faster

[@ConTejasMusic]

Anyone notice the background music at the start is Moonlight Sonata 1st movement, but a hip hop version?

[@rgrimoldi]

We should have gone for Hendrix instead! I definitely have to build that.

[@RonoTron01]

If I want to feed like a series of midi files into this neural network, how to do that? I have maybe a few hundred midi files of the style that I like, is this possible to use as a source?

[@sukumarhonkote9936]

can you provide the link for the preprocess library code? I am getting an import error

[@timehrensberger6479]

I can't figure out how to generate based on other midi files than the standard metheny piece

[@roguenoir]

I'm curious if it would be legal to use music you generate this way as your own.. especially if you train the deep learning net on a commercial artist's music then use that to generate new music which you use for, say, a movie or game you make.

[@HikikomoriDev]

How long did it take for the computer to finalize the final music piece ?

[@lesmizzle]

What was the training data used? I missed it.

[@prateekthakur4060]

I don't see the code Siraj explained in the GitHub link. Can somebody help here:;)

[@eumieumi8766]

hmm the output kinda sucks when i train it on own midis ..sure would be nice if it would train multiple times

[@NoName-iz8td]

can you explain please the input data, cuz for me it is difficult to transform from midi file to vector numbers :p

[@vigneshsuresh3671]

Great series mate! Can you do a video on data points extraction from text with ML .. Regex sucks :(

[@owena7434]

Great vid! VERY new to this stuff. so just to clarify, hope this isn't horribly wrong, the reason we need to put in an LSTM is to keep the program learning from the jazz, and not accidentally learning its own randomness?

[@SirajRaval]

Thanks Owen, LSTM helps it remember ALL the jazz it trained on, not just the most recent jazz (long short term memory)

[@owena7434]

Thanks Siraj! Cool stuff! I think it would be cool to explore how a machine might navigate a more complicated classical work with key changes with LSTM so the machine-generated piece has overall structure. Or even reverse engineer different composer's harmonic structures. But I'm still on the basics lol!

[@Mosfet510]

I didn't do it! It was my algorithmic counterpart, I swear!! lol Great video.

[@SirajRaval]

lol thanks Scott!

[@bomb3r422]

i started programming in python 5 months ago . i have a good idea of oop and algorithms . should i start with machine learning now or is it too early?

[@y.z.6517]

Don't touch ML at the moment, or you will hate it. Find a project that solves your problem, and work on it. Something involves parsing texts and doing maths you can understand. Then make codes reuseable and easy to maintain. At that moment, you will have experience in researching, debugging, and maintaining long and complicated codes along with someone else's libraries.

[@klaudialustig3259]

I used to think that Markov chains are complicated stuff. I learned that it's basically just a state machine.

[@SirajRaval]

exactly

[@aseemKurhekar]

In one of his videos siraj talked about how neural networks can make music from images. If some one could point me to its source it would be great 😊

[@shivanasr8127]

i want extract pichbend from midi file can you help me?

[@debarokz]

Nice video.. I am planning to buy a rig for stuffs related to deep learning ..what's your system config that you use for deep learning??? will MacBook pro suffice??

[@SirajRaval]

Macbook pro works also I prefer the cloud to a dedicated rig (AWS)

[@debarokz]

Siraj Raval thanks bud

[@tylerangert9121]

do you have scripts for all of your videos or are you just explaining off the top of your head? I think a big thing you can do to help make our learning experiences better is providing us with the transcripts of the videos. for me at least, having text to reference and look back to is much easier than just going back and listening to audio again (this is why i usually watch TV shows with closed captioning).

[@SirajRaval]

scripts. youtube has automatic captioning

[@rhezabudiono2385]

Hey Siraj, as an undergrad should I prepare differently to be a machine learning researcher vs ML engineer? I was thinking that to be to be a researcher I should take more stats, math, physics whereas to be an engineer I should take more programming-heavy CS classes

[@AkshayAradhya]

Rheza Budiono yup correct

[@yousofebneddin7430]

Nice shirt :) can you also provide a video on speech recognition?

[@SirajRaval]

thanks! i have one 5 videos back

[@yousofebneddin7430]

Do you mean "How to Do Sentiment Analysis" ? What I meant is that the input of the Algorithm is the sound and the output is the pronounced words.

[@calvin5371]

Honestly, I had to watch the video twice ... cause the first time I was just focussing on the t-shirt

[@shaikrasool1316]

Is it better choice to learn machine learning ai,,, in present day

[@KETANRAMTEKE]

Go back to vdo. It's more amazing than comment section :D

[@MultiSciGeek]

This is really confusing. Subscribed

[@lxxanderfans267]

this is so interesting lol

[@alexhormann8931]

What are you doing with your hands?

[@travisgoodwin6802]

Ok, let's hear it already, the music that is. Something inspired, not sounding derivative.

[@sir_aken9706]

I have Pycharm and I'm not sure I can replicate what you did in it.

[@dominiksabat]

Briliantly explained, you should work at the university.

[@diegoantoniorosariopalomin4977]

sinesthesia as an artificial nn (natural sinestesia is also on neural networks ) , in other works real life theme music

[@SirajRaval]

good insight! i have sinesthesia

[@diegoantoniorosariopalomin4977]

Siraj Raval you do ?! .... oh that explains why you break your videos with musical numbers that explain the topic

[@raj-cp5nu]

2:20 I press play, so so true haha

[@WilsonMar1]

[2:32] Coding begins

[@EduardoRenz]

Cool Can you make a ML that learns to make music from a real music audio? Analysing a whole album of a band?

[@RitobanRoyChowdhury]

You would need it midi format. Analyzing mp3s is a much harder problem (because they store audio data rather than notes)

[@danieltv123]

there are some people already doing it from real audio, search on youtube

[@sumantkanala]

Hmm. this challenge is a tad bit harder to understand than previous ones!

[@SirajRaval]

indeed but give it a try! You just might win

[@easter.bunny.6]

next time maybe polyphonic music?

[@RicardoMV95]

Pat Metheny is a guitarist not a pianist. I’m not sure if you chose to make piano pieces based on his music or if you didn’t realize that he is a guitarist. Regardless, cool video!

[@mallikarjun480]

What's the best way to start Artificial Intelligence.

[@SirajRaval]

my videos

[@OrthoGunner]

you should go work at OpenAI!

[@SirajRaval]

they are my homies

[@ili626]

Pat Metheny (ma - thee - nee)

[@sukumarh3646]

I liked your video but why didn't you cite the original code in the videos or at least in the description? The original code is available here github.com/jisungk/deepjazz

[@SirajRaval]

just added it thanks

[@aarondinesh3657]

When I try to run the code this error pops up: Traceback (most recent call last): File "generator.py", line 21, in from music21 import * ImportError: No module named music21

[@sleep4103]

Who else had a Frank Jev Cee association :'D

[@nevokrien95]

Lstm usualy have atention gates

[@jiexun9679]

Hi Siraj, I attempted the challenge but results are not very good for now. Here's my code anyway: github.com/jiexunsee/rudimentary-ai-composer Thank you for the video!

[@SirajRaval]

great job Jie for going for it

[@berrym8901]

The music generated by RNN is good... Well, anyway, it's better than Kenny G! Poor Kenny G~ :p

[@SirajRaval]

pwnt

[@anthonyparks505]

There's a lot of good info in this video, but you move so incredibly quickly that it is impossible to follow. The pacing is frenetic...

[@kracejiciAlchymista]

Noooo, Kenny G`s goood man.

[@simthology5770]

CODE SHMODE WHEN ARE YOU NERDS GONNA MAKE A GUI FOR CRYING OUT LOUD?

[@neelpendharkar4793]

You need help. Thanks tho

[@Acc22-mr1qk]

I didnt understand anything

[@gawni1612]

Lost me at mid eye

[@ukaszOgan01]

Can you speak more slow please? :)

[@cryhav0k2112]

Love all your videos but you should talk slower, it gets a little overwhelming for people who aren't so advanced