Zoom Logo

MIT Heavy Metal 101: AI Metal w/ Dadabots & Colin Marston - Shared screen with speaker view
Mitchell Chaiet
30:36
hello hello!
Kelly Phillips
30:47
hey everyone! Been at least 24 hours haha
Dan Curhan
31:00
hi Kelly! hi colin! hi everyone!
Symmetry Alexander
31:03
hi again kelly!
Kelly Phillips
31:06
hey Sym and Dan!
Kasra Sammak
31:10
Hey I can’t speak cuz my girlfriend is DJing breakcore in the background haha
Kelly Phillips
31:30
and I see Katie, Colin, Frank, Kim, and Pedro from last night too!
Kasra Sammak
31:33
But my name is Kasra, I’m in Berlin, and CJ can do a short intro for me
Pedro Sarmento
31:37
Hi everybody! I’m Pedro, attending from Porto, Portugal! :)
Pedro Granada
31:38
Hi all!
Frank Liu
31:43
yas for concert
Mitchell Chaiet
31:44
atx!
Sebastian Castro
31:53
I'm a robotics engineer at MIT's Computer Science & AI Laboratory, also a metal head. Super cool confluence of topics \m/
Luke Rymarz
32:13
Hi Joe!
james s
32:20
hey gang
sym (they/them)
32:21
shoutout to Pedro with the BTBAM background
Jack Eaton
32:22
Hey Joe!
Emma Gersten
32:25
I'm Emma. northeastern u student finishing CET degree
Dan Curhan
32:38
intro: musician, metalhead, mechanical engineer but did tons of robotics club stuff in college!
Kasra Sammak
32:39
I’m into Data Science and EEG Analysis/Machine Learning BMI development. Met CJ and Zach in Berlin at Data Natives was fun!
Mason H
32:49
Liz Marks and Mason here!
Kasra Sammak
32:52
Also producing electronic beatsss
Mark Jones
33:12
I’m interested in machine learning. Worked at the MathWorks for 11 years, now working at Athenahealth; nothing related to AI at either place.
Mitchell Chaiet
33:17
I build midi controllers and epic guitar solos. I'm also a media manipulation researcher
Mitchell Chaiet
33:23
also hi CJ!
Kasra Sammak
34:20
Yes!
Kelly Phillips
34:21
Only Colin’s stuff!
Pedro Sarmento
34:23
Yes!
Mitchell Chaiet
34:24
yep
Meaghan T'ao
34:24
ye
Elliott Brown
34:26
Huge fan!!!
Mason H
34:27
not yet!! post a link please!
Emma Gersten
34:29
I came because CJ invited me. I'm staying to hear more about AI
Mark Jones
34:37
I also play around with music production a little with Ableton, but more into dj’ing EDM with Serato. Started to take the ML class on coursera but never finished due to getting distracted by a mobile app I wrote.
Mark Jones
34:50
I’ve seen CJ perform 1 or 2 times
Mitchell Chaiet
34:57
https://www.youtube.com/channel/UCLjU_5t9kP3DjNPmvVJx2eg
Mitchell Chaiet
35:03
^dadabots
Jack Eaton
35:05
Definitely a fan of Colin's stuff as another bassist
Emma Gersten
35:40
listened to the dadabots stuff before
sym (they/them)
35:44
heck yeah!
james s
35:46
love the Relentless Doppleganger stream
Kasra Sammak
35:48
oops also meant to post to main
Dan Curhan
35:48
was that how you did xazraug's "not programmed" drums?
Emma Gersten
35:50
it's kewl
Kelly Phillips
36:16
Eric Import - Night Verses
Kelly Phillips
36:22
*Improta
Ana Maria Velasquez Giraldo
36:54
Yes, you can log into zoom in two devices
sym (they/them)
37:08
forgot to answer the question earlier but i went to Dadabots’ guest lecture last time so that’s how i know of them!
Mitchell Chaiet
37:37
joe can you send me your email? would love to connect
Gerald Chau
37:55
hail satan
Gerald Chau
38:23
yo from scotland
Kasra Sammak
38:35
coool
Ana
39:00
London, UK
Mark Jones
40:06
Just curious, are the folks from all over students at MIT or just joining b/c they heard about the event? I heard about it from CJ, not a student at MIT.
Katie Means
40:16
I work at MIT!
Sebastian Castro
40:32
I also work at MIT -- not a student.
Meaghan T'ao
40:37
heard about event
Dan Curhan
40:55
I'm in Boston - somerville, but unaffiliated with mit, hah
Frank Liu
40:55
I lurked in from twooter
Emma Gersten
40:57
I'm not an mit student or employee (wish I was though)
Kelly Phillips
41:07
lol same Emma
Kelly Phillips
41:14
I can say I was an MIT professor for 90 minutes
sym (they/them)
41:19
same Emma
Jeff Warmouth
41:34
friends with CJ - he posted on FB
Ana Maria Velasquez Giraldo
41:35
I am not from MIT but I admire their work, lol
Mason H
41:36
I met joe's beautiful curls and have been following him ever since. not an mit student.
Jack Eaton
41:52
RPI not MIT lol
Emma Gersten
42:18
RPI?
Jack Eaton
42:54
Rensselaer Polytechnic Institute, another engineering school
Kelly Phillips
43:06
Jack, is that in New York? Can’t remember
Katie Means
43:12
That's where my bf went for undergrad. We visited a few years ago. I really like Troy
Jack Eaton
43:29
Yup, Troy. I've only been in MA for a few years
Joe Diaz
43:35
Music Hack Day! I remember!
Sebastian Castro
46:27
that's my uncle!
CJ Carr
46:35
XD
Joe Diaz
46:41
Hahaha
Joe Diaz
50:27
Feel free to drop any questions or comments in the chat! If you have a question about any terms, etc., let us hear them!
Joe Diaz
52:03
Woo Davis Square, Somerville
Dan Curhan
52:18
I thought I heard davis square, I'm within a mile from davis!
Emma Gersten
52:20
is this being recorded so I can rewatch later?
Kelly Phillips
52:21
Red Line? Am I getting this right as a non-Bostonian?
Dan Curhan
52:33
heh, davis is indeed on the red line
Emma Gersten
52:37
Davis is red line yes
Katie Means
52:39
Red line! Yes!
Joe Diaz
52:39
Emma, yes! I'll be posting on metal.mit.edu later on
Kelly Phillips
52:50
my favorite Red Line station is Alewife fo sho
Joe Diaz
52:55
Kelly, you got it. You're safe
Dan Curhan
52:56
pffffft
Katie Means
52:57
Saaaame
Dan Curhan
53:06
the porter escalator tho
Joe Diaz
53:26
Cardio when it's broken
sym (they/them)
53:29
i’m with dan, the porter escalator is the coolest
Mason H
53:36
is the programing randomly selecting bits of music or is there programing to give a theme so to speak?
sym (they/them)
53:43
alewife station is disorienting
Dan Curhan
53:48
you gotta "train" it, right?
Dan Curhan
53:58
pun not intended tbh
Frank Liu
54:43
Obama reads navy seal copypasta (0.9 beta)
Sebastian Castro
54:44
last year they had a President Obama fake officially introducing the MIT Intro to Deep Learning course: https://www.youtube.com/watch?v=l82PxsKHxYc
Kelly Phillips
55:02
CJ I am so unbelievably impressed by your ability to know exactly what tab to go to without much hesitation lol
Joe Diaz
56:34
+1 Kelly
Joe Diaz
56:56
Yes Sebastian! I also recommend the Center for Advanced Virtuality at MIT
Joe Diaz
57:16
https://virtuality.mit.edu/
Mason H
57:19
thank you!!
Jack Eaton
57:35
so you're basically filtering out specific characteristics of amplitudes?
Emma Gersten
57:45
there are neural net books made for babies nowadays. the future is going to be wild
Joe Diaz
58:18
Holy crap, Emma. https://www.goodreads.com/book/show/39971473-neural-networks-for-babies
Mark Jones
59:19
And sounding like the Beatles too
Colin Brumley
01:00:04
what was your line of not sounding too much like the original?
Gerald Chau
01:00:06
Have you tried Style Transfer to transplant X genre or band to Y genre / band?
Mark Jones
01:01:16
So do you do it all in the time domain rather than frequency domain?
Mark Jones
01:02:18
Guess you could even split it off to two nn and then recombine via additive mixing
Sebastian Castro
01:02:22
You could imagine a frequency domain use of RNNs that essentially synthesizes new "columns" of a spectrogram given the previous N columns
Kelly Phillips
01:03:23
it’s the black metal version of that AI art thing which generates paintings
Meaghan T'ao
01:04:06
melone
Kelly Phillips
01:04:08
Anthony FANTANOOOOOO
Emma Gersten
01:04:20
I love TND. Saw this one
Joe Diaz
01:04:29
I remember watching this
Kelly Phillips
01:04:30
“washy,” what a word
Mark Jones
01:04:38
Or maybe split the time series up by a bunch of bandpass filters into bands and then a nn for each band and recombine…I really need to learn ML, I get so into thinking about this stuff
Mark Jones
01:04:56
:)
Dan Curhan
01:05:07
he looks so... perplexed.. hahaha
Kelly Phillips
01:05:08
lmao love how he is just blankly staring trying to comprehend what he’s hearing
Meaghan T'ao
01:05:14
lool
Emma Gersten
01:05:18
never seen him so contemplative
Gerald Chau
01:05:23
Nailing Nick's vocals
Kelly Phillips
01:05:28
he looks like that album slapped his mother in front of him
Emma Gersten
01:05:35
lmao
Kelly Phillips
01:07:17
“Aliens”
Mark Jones
01:07:58
Two great albums behind him, that’s for sure
Dan Curhan
01:07:58
like the archspire bot
Jack Eaton
01:08:16
@Mark Jones that's kind of what I was thinking with my last question...like learning what amplitudes associate with what type of sound and kind of 'filtering' them out to resemble previously obtained samples
Dan Curhan
01:08:30
OH WAIT NO THAT'S YOU, FUCK
Joe Diaz
01:08:43
Hahahaha Dan
Mason H
01:08:50
if this AI is able to produce music comparable to "human" made music does that remove the "soul" aspect of music? does the ability for AI to invoke emotional response in humans give you any pause?
Mark Jones
01:08:52
My thought was then it might learn how to kind of emulate instruments
Kelly Phillips
01:08:53
well TIL there’s an Archspire bot???
Dan Curhan
01:09:02
https://www.youtube.com/watch?v=MwtVkPKx3RA
Stephen Zhang
01:09:28
is there any regularisation you can use to try and reduce the noise/unstructured component of the ouptut
Mark Jones
01:09:29
Not that a band pass filter will isolate to just a single instrument, but more so than on the unfiltered
Jack Eaton
01:09:36
Exactly...like being able to separate out sections with vocals because they have certain characteristics
Kelly Phillips
01:09:42
holy that is Archspire
Kelly Phillips
01:09:46
that sounds exactly like it hahaha
Jack Eaton
01:09:49
And then going from there into separating out different instruments
Dan Curhan
01:09:52
it's so legit archspire it's nuts
Dan Curhan
01:09:56
I thought it was snippet-based tbh
Frank Liu
01:10:15
????this is actual archspire
Emma Gersten
01:10:17
I just bought the Neural net baby book, screw it why not. my friend just had a baby too.
Joe Diaz
01:10:29
Yes to all of the above
Kelly Phillips
01:10:30
Petition to ask Dean Lamb to have him and Claire learn all of this and post it on their YouTube channel
Joe Diaz
01:10:36
And below
Dan Curhan
01:10:43
amazing
Gerald Chau
01:10:44
Did you ever try training on stems then feeding the trained stems together?
Mark Jones
01:11:06
What are stems?
Dan Curhan
01:11:25
hey colin, I volunteer my stems for Gerald's experiment
Mitchell Chaiet
01:11:27
different parts of a song
Dan Curhan
01:11:27
hahaha
Mitchell Chaiet
01:11:33
separated out
Gerald Chau
01:11:34
@MarkJones the individual instruments (not the full mix)
Mark Jones
01:11:39
ah
Mitchell Chaiet
01:11:42
vocals, guitar, bass, drums
Mitchell Chaiet
01:11:56
used for remixing usually
Mark Jones
01:11:59
Definitely better way than bandpass filters probably
Joe Diaz
01:12:13
I'm noting questions for later on! Just in case
Jack Eaton
01:12:15
Engineers love to overcomplicate
Sebastian Castro
01:12:22
^ it is known.
Joe Diaz
01:12:31
+1 Jack
Sebastian Castro
01:15:06
@all, there are ways to extracy "style" out of music using unsupervised learning techniques -- one common example today is Variational Autoencoders (VAE). Basically you train a neural network to go through a low-dimensional "bottleneck" and perfectly reconstruct the output. Then, you start messing with that bottleneck in the middle and you can pull out different style "clusters" inn your data
Sebastian Castro
01:15:14
extract*
Joe Diaz
01:15:48
Thanks Colin!
Gerald Chau
01:16:05
Is there something like this but for audio?
Gerald Chau
01:16:06
https://www.tensorflow.org/tutorials/generative/style_transfer
Dan Curhan
01:16:11
lmao that albu title
Jack Eaton
01:16:12
My mind is being blown in like 3 different ways right now
Gerald Chau
01:16:34
that stream was great
Viktor Simjanoski
01:16:35
what's the typical size of your NNs that you use in terms of number of layers/nodes?
Kelly Phillips
01:16:45
Question for Colin - is that kind of what you did on the split Indricothere album with Geryon? Just saw on the Bandcamp that it has “effected bass” and I didn’t know if that was implying ‘randomized’ or something
Kelly Phillips
01:16:57
#infinitegrindcore
Dan Curhan
01:17:02
it's got the bree
Dan Curhan
01:17:04
it's brutal
Kelly Phillips
01:17:06
BLEGH
Gerald Chau
01:17:34
The cymbal stop starts and vocal punches were top notch
Sebastian Castro
01:18:51
here is an example of generating music at a particular style: https://magenta.tensorflow.org/transformer-autoencoder
Sebastian Castro
01:19:37
and a more "tutorial" version: https://www.kaggle.com/basu369victor/generate-music-with-variational-autoencoder
Joe Diaz
01:19:46
Thanks Sebastian
Jack Eaton
01:20:51
All recommended reading on this topic is highly appreciated :)
Gerald Chau
01:21:02
Thanks Sebastian
Kelly Phillips
01:21:05
Me, to someone in the elevator: “Yeah, if you love XavlegbmaofffassssitimiwoamndutroabcwapwaeiippohfffX, you’ll love Thrakhrathraatevviceatencengorath”
Dan Curhan
01:21:12
GP5 for life!
Jack Eaton
01:21:14
Guitar Pro <3
Emma Gersten
01:21:16
nope
Joe Diaz
01:22:08
The chimes sold me
Sebastian Castro
01:22:08
sounds like the band Unexpect
Kelly Phillips
01:22:25
this is giving me major Black MIDI vibes
Kelly Phillips
01:22:31
for all the Adam Neely fans out there
Jack Eaton
01:22:31
When the drummer writes the guitar parts
Dan Curhan
01:22:36
fables is top 5 albums of all time, I love me some unexpect
Joe Diaz
01:23:11
https://dadabots.com/science.php
Emma Gersten
01:23:17
black metal is my fav!
Kelly Phillips
01:24:10
YES
Mitchell Chaiet
01:24:12
yessss
Kelly Phillips
01:24:17
one of the greatest videos on the internet
Dan Curhan
01:24:29
there are your stems
Kelly Phillips
01:25:00
https://www.youtube.com/watch?v=dbbque0Y4FU
sym (they/them)
01:25:25
i’m not familiar with the bad flute meme, can someone link it please?
Sebastian Castro
01:25:43
sym, this I think is the original: https://www.youtube.com/watch?v=BG6EtT-mReM
Sebastian Castro
01:25:47
and then people memed.
Kelly Phillips
01:25:58
[Obama not bad face]
Gerald Chau
01:26:02
Is this Colab? https://magenta.tensorflow.org/demos/colab/
Emma Gersten
01:26:05
this colabs?https://colab.research.google.com/notebooks/intro.ipynb#recent=true
Mark Jones
01:26:39
Ha I found https://research.google.com/colaboratory/faq.html
Mark Jones
01:27:22
Sounds like clown metal with blow horns, like Harpo Marx
Jack Eaton
01:27:23
yup that's death grips
Dan Curhan
01:27:33
how do you generate the "instrument" synth sounds like flute? using traditional synth models?
Jack Eaton
01:28:21
https://magenta.tensorflow.org/ddsp ?
CJ Carr
01:28:26
yup
CJ Carr
01:28:35
ddsp takes care of it all
Kelly Phillips
01:28:46
man can you imagine if GG Allin had this back in the day!?!? good lord
Joe Diaz
01:29:03
Bad reading memes and ddsp, yes
Dan Curhan
01:29:14
grown empathetic furniture lmao
Meaghan T'ao
01:29:40
w a t e r m e l o n s a d n e s s
Jack Eaton
01:29:52
N0)))FX
Joe Diaz
01:30:49
https://openai.com/blog/better-language-models/
Sebastian Castro
01:31:18
notably, their newer version (GPT-3) is not an open model, but is rather a paid service hosted by OpenAI. Sad :(
Kelly Phillips
01:31:44
DESKTOP *pig squeal*
Sebastian Castro
01:31:57
condemned to life of perineum...
Kelly Phillips
01:32:05
oh man that last one was a Travis Ryan lyric for sure
Kelly Phillips
01:32:09
Carbonization of life on earth
Meaghan T'ao
01:32:33
lool
Frank Liu
01:32:40
lmao eek
Jack Eaton
01:32:52
Didn't one of the first experiments of this turn into a Nazi in like hours
Meaghan T'ao
01:33:03
YES
Joe Diaz
01:33:03
Microsoft chatbot right?
Mason H
01:33:16
is this what the Malestrome peeps in Cyberpunk 2077 listen to?
Meaghan T'ao
01:33:19
im so happy
Joe Diaz
01:33:27
Mikumetal
Meaghan T'ao
01:34:01
joe that's just wagakki band
Joe Diaz
01:34:12
Haha
Kelly Phillips
01:34:31
so it’s just industrial music, that’s sick
Meaghan T'ao
01:34:43
this is amazing
Ana
01:34:46
Yeah hahaha great lyrics
Jack Eaton
01:34:56
These lyrics are incredible
Dan Curhan
01:35:00
fantastic hahahaha
Mason H
01:35:11
amazing
Jack Eaton
01:35:12
My favorite is 'sores sewn on your skin'
Kelly Phillips
01:35:38
this guy is amazing
Emma Gersten
01:36:00
I want to see an Obama deepfake do an AI death metal performance
Kelly Phillips
01:36:13
Emma, this is the future liberals want
Dan Curhan
01:36:18
yes emma!
Emma Gersten
01:36:25
lmao!
Kelly Phillips
01:36:29
/s obviously but yeah, I support that
Kasra Sammak
01:36:38
ouu
Kasra Sammak
01:37:10
bita glitch
Kasra Sammak
01:37:11
nice
Kelly Phillips
01:37:29
so this was machine learning pretty much applied on the fly?
Kasra Sammak
01:37:57
CJ I want you to talk about the audio integrity of these neural synths on electronically synthesized kicks, the problem we were having, etc
Mason H
01:37:59
I must break him
Joe Diaz
01:39:01
Hahaha
Dan Curhan
01:41:16
is training with one song as input too little input?
Dan Curhan
01:44:42
you'd have to convert into midi-esque domain from audio, and then back
Gerald Chau
01:44:51
Thanks for the answers B-)
Mark Jones
01:44:53
Would there be a way to exploit the more limited frequency range of single instruments to somehow limit the amount of memory needed. I’m thinking of how like an MP3 might be smaller maybe if the track has only one instrument vs a full band (I haven’t tested this, just theorizing) because the data is more limited in the amount of variance and maybe more easily compressed? Maybe this makes no sense in the time domain given you have a set sample rate and number of bits/sample, but maybe in the frequency domain?
Meaghan T'ao
01:46:20
ay
Pedro Sarmento
01:46:33
@Mark Jones, I think that’s exactly what dadabots did with Adam Neely on one of their latest videos. By reducing the sampling rate, you can train on “longer chunks” of memory
Mark Jones
01:47:20
That’s another option but have to watch out for lost higher frequencies (nyquist rate)
Joe Diaz
01:47:46
Have a question? Use the raise hand button, and we'll get you in
Pedro Sarmento
01:47:53
Because it’s only bass, we can get rid of the of the higher freq content safely"
Ana
01:48:22
So true ;)
sym (they/them)
01:48:32
whenever a new genre gets “invented” there’s always someone (or in this case an AI) who does it for the first time, as in they came up with something that hadn’t been heard before
Liz Surette
01:50:03
Sorry I'm late. Just wanted to say I'm a huge Gorguts fan and have seen you guys live twice, picture with Luke Lemay, etc. This is an awesome project, thanks for all you do \m/
Sebastian Castro
01:50:42
In the context of music, destroying humans would mean destroying your training data... but the AI can generate its own data so I guess we're not needed
Joe Diaz
01:51:17
Fantasia brooms and water
Dan Curhan
01:52:33
have you spent all your time dabbling in sonically dense ai generation, or have you done any generation of music based on a single, pure instrument like a violin or acoustic guitar?
Joe Diaz
01:52:42
Thanks Kelly!
Joe Diaz
01:53:02
or so we think...
Meaghan T'ao
01:53:13
must...kill...colin...marston
Kelly Phillips
01:53:19
looooool
Meaghan T'ao
01:53:19
must...make...more...music
Dan Curhan
01:54:02
do the bots judge it themselves?
sym (they/them)
01:54:59
thank you Dadabots and Colin!!
Mason H
01:55:02
thank you so much! this has been incredible
Kelly Phillips
01:55:12
yeah this has been wild, thank you to all three of you
CJ Carr
01:55:12
thx thx ill be hanging out and still answering questions
Mark Jones
01:55:20
I have to drop at 8 but this was awesome! Thanks for the invite again CJ and thanks to all presenters and those in the chat!
Jack Griffiths
01:55:27
great talk, thank you!
Ana Maria Velasquez Giraldo
01:55:40
thank you!
Elliott Brown
01:55:41
Thank you CJ, Zach, and Colin!!!
sym (they/them)
01:55:44
and thanks Joe of course!!
Emma Gersten
01:55:47
@cj where can I find that academic paper you published?
CJ Carr
01:56:07
http://dadabots.com/science.php
Viktor Simjanoski
01:56:09
what's the typical size of your NNs that you use in terms of number of layers/nodes?
Emma Gersten
01:56:13
danke
Joe Diaz
01:56:24
https://padlet.com/joediaz/mitmetal101
Jeff Warmouth
01:56:43
thanks, this was awesome \m/
CJ Carr
01:56:50
our SampleRNN has 66 million parameters, fits on a 16GB GPU, has two tiers of 5 LSTM layers and a few fully connected layers
CJ Carr
01:56:54
with skip connections
Emma Gersten
01:56:58
padlet reminds me of Tumblr for some reason. first time seeing it
Mark Jones
01:57:04
Oh, did anyone collect all of the links shared in the chat? Will they be posted anywhere? I can get the link from CJ later
Pedro Sarmento
01:57:27
Gotta go guys, thank you so much, was great!
Joe Diaz
01:57:35
Thanks Pedro!
Joe Diaz
01:57:52
https://krallice.bandcamp.com/album/diotimahttps://beholdthearctopus.bandcamp.com/
CJ Carr
01:58:04
<3
Joe Diaz
01:58:08
https://dysrhythmia.bandcamp.com/musichttps://encenathrakh.bandcamp.com/http://gorguts.com/
Joe Diaz
01:58:20
Infinite Bass https://www.youtube.com/watch?v=VBXNr-msjfQ&t=0s
CJ Carr
01:58:20
Feel free to DM me on twitter http://twitter.com/dadabots and I can answer more questions too
CJ Carr
01:59:44
Infinite Bass Solo, new link hey I
Viktor Simjanoski
01:59:50
where does the name krallice come from? It sounds like a word I know from a foreign language, wondering if that was the source or it's just a random coincidence :)
CJ Carr
01:59:52
https://www.youtube.com/watch?v=RRwJzw4VvZA
Katie Means
01:59:56
I've gotta go. Thanks for everything!
Joe Diaz
01:59:59
Oops thanks!
Joe Diaz
02:00:02
Thanks Katie!
Emma Gersten
02:00:08
Does anyone know of a good Discord server for AI/ML topics?
CJ Carr
02:01:44
@Emma join our discord! we just started. hold on let me make an invite link
Joe Diaz
02:01:56
Thanks CJ!
CJ Carr
02:01:56
Also there's a few more I can post
Gerald Chau
02:02:07
I'd be interested in a Discord server
Emma Gersten
02:02:07
danke!
CJ Carr
02:03:20
https://discord.gg/wahgfKdq secret invite to the CJ+Zack discord super special for you guys
Emma Gersten
02:03:29
<3
CJ Carr
02:03:48
Jukebox Community https://discord.gg/U57Cvjd8
CJ Carr
02:04:07
ArtBreeder community https://discord.gg/TQYgpUFm
sym (they/them)
02:04:10
can anyone here join the discord?
Kelly Phillips
02:05:06
I’m in, sum, try copying the link that goes into your browser into a different tab
Kelly Phillips
02:05:10
*sym
Jack Eaton
02:05:39
I need to bounce, thank you everybody!
Mark Hanslip
02:05:40
have you guys worked with WaveGAN at any point?
Joe Diaz
02:05:46
Later Jack!
sym (they/them)
02:05:49
i meant i wasn’t sure if that invite was specifically for a couple people who asked questions or for the whole class
CJ Carr
02:05:52
Audio DeepFakes is another discord community
Joe Diaz
02:06:22
Process vs. Product indeed
Zack Zukowski
02:06:53
https://en.wikipedia.org/wiki/Bible_of_Kralice
CJ Carr
02:06:57
Mark: WaveGAN not yet but MelGAN yes. Interesting discussion is at what level of audio hierarchy GANs works best at
Joe Diaz
02:06:58
Nice
Viktor Simjanoski
02:07:08
cool, thanks
Mark Jones
02:07:17
Process vs Product: Bushwick Bill said if it don’t make dollars then it just don’t make cents. :)
Dan Curhan
02:07:25
also gotta bounce, but thanks so much CJ, zack, colin + Joe! This was super informative, inspirational, and a good laugh. Cheers!
Gerald Chau
02:07:31
https://en.wikipedia.org/wiki/Bible_of_Kralice
Kelly Phillips
02:07:32
peace Dan!
CJ Carr
02:07:36
thx dude
Mitchell Chaiet
02:07:38
how long of a guitar solo do you need to make a virtual mitch
Gerald Chau
02:07:42
I'm off to bed, night all. Great talk.
CJ Carr
02:07:45
30-60 minutes
Joe Diaz
02:07:48
Thanks for joining!
Sertan Senturk
02:08:00
Thanks a lot! It was a great session! Rock on!
Joe Diaz
02:08:04
:D
Emma Gersten
02:08:46
when will ya'll get on the Lex Fridman podcast?
CJ Carr
02:08:55
@emma LOL ask him for us
Sebastian Castro
02:09:01
just say Elon Musk's name 17 times and he'll notice
Joe Diaz
02:09:07
I'm glad you asked
Jess Yang
02:09:09
Got to take off, thanks!
Meaghan T'ao
02:09:11
yep
Joe Diaz
02:09:40
https://www.youtube.com/watch?v=-VsmF9m_Nt8
Meaghan T'ao
02:09:42
simlish
Joe Diaz
02:09:44
Later!
Sebastian Castro
02:10:23
it's like the band Sigur Ros, who made up the "hopelandic" language just because it sounded good to sing
Mark Hanslip
02:10:40
I didn't manage to get MelGAN working when I tried it ~a year ago. FWIW I <3 WaveGAN for my use case (saxophone improv)
CJ Carr
02:10:55
Mark: single instruments definitely train better, even in the time domain: simpler, because less diversity of timbre
Meaghan T'ao
02:11:10
iirc he wanted to make it sound like English to Italians
Mark Jones
02:11:11
I’ve noticed that if I turn down a song really low, my mind will sometimes create an entirely new song. I just wish I knew how to play an instrument to make it a reality when it happens.
CJ Carr
02:11:11
oh cool can you share your WaveGAN stuff
Jacob Schwartz
02:11:18
If only Borges was here.
Mark Hanslip
02:11:44
Yes. I found WaveGAN 'played' stuff that I would never think to play even though fed on me practicing
Meaghan T'ao
02:12:05
absolute banger
Sebastian Castro
02:12:13
it's like that Spanish pop song "Asereje"
Sebastian Castro
02:12:20
but a whole song of it vs. a chorus
Kasra Sammak
02:12:20
https://www.youtube.com/watch?v=Vt4Dfa4fOEY&ab_channel=BrianandKarl
Kasra Sammak
02:12:23
this is what i was talking bout
Viktor Simjanoski
02:12:28
thanks for the great presentation, good-bye
Mark Hanslip
02:12:30
https://soundcloud.com/mark-hanslip/gander
Emma Gersten
02:12:36
this is a bop!
Kasra Sammak
02:12:37
its a better presentation its more of a conversation
Sebastian Castro
02:15:14
or half-Creed, half-Cannibal Corpse
Joe Diaz
02:15:24
Quebecois?
Joe Diaz
02:15:53
I'd buy the album Seb
sym (they/them)
02:15:55
thanks CJ and Zack and Colin and Joe, hope everyone has a good night!
Kelly Phillips
02:16:02
putting my Spotify listening history to shame
Sebastian Castro
02:16:09
thank you -- this was a lot of fun! TIme to listen to some bot music \m/
Mark Hanslip
02:16:34
waveGAN stuff: https://soundcloud.com/mark-hanslip/gan-with-no-name
CJ Carr
02:17:08
https://www.youtube.com/watch?v=RRwJzw4VvZA
Kelly Phillips
02:17:28
this was absolutely amazing, thanks for being so passionate about this topic! This was amazing guys, thanks
CJ Carr
02:17:46
After party ==> https://www.youtube.com/watch?v=RRwJzw4VvZA
Mark Hanslip
02:18:02
thanks! super interesting. is later here in UK, goodnight x