Ai Dreams Forum

Software => General Software Talk => Topic started by: chung on April 24, 2019, 10:44:50 pm

Title: not neural network text and music generation
Post by: chung on April 24, 2019, 10:44:50 pm
i made a non neural network ( RNN , LSTM ...) music and text generation using SORT sequences algorythm ( gentext_chung & brainpiano3 of midipiano_chung )  ::)

=>  http://chungswebsite.blogspot.com/2019/03/gentextchung-text-generation-based-on.html (http://chungswebsite.blogspot.com/2019/03/gentextchung-text-generation-based-on.html)
=>  http://chungswebsite.blogspot.com/2019/03/brainpiano3-sort-not-neural-network.html (http://chungswebsite.blogspot.com/2019/03/brainpiano3-sort-not-neural-network.html)
Title: Re: not neural network text and music generation
Post by: goaty on April 25, 2019, 01:38:14 am
Nice 2 see a fellow developer.  I usually work without traditional neural meshes as well.

Synonymity in English means that two different words have the same meaning.
In music, this would translate to a phrase of notes which has a common surrounding of notes around it.

So the computer decides it "means the same thing" even tho they are different note "words".

Every time you say somethings the same, you get more "music" to come out,  but eventually it breaks down if you do it too much.
Title: Re: not neural network text and music generation
Post by: Korrelan on April 25, 2019, 08:11:58 am
Welcome Chung... Interesting project.

 :)
Title: Re: not neural network text and music generation
Post by: LOCKSUIT on April 25, 2019, 10:29:08 am
Heh goaty you really got that idea from me didn't you >:)

This is what I do too chung, no net whitebox stuff, I will check it out.

Can you show us samples of the music?
Title: Re: not neural network text and music generation
Post by: chung on April 26, 2019, 06:04:44 am
i made a video  https://www.youtube.com/watch?v=LsyjQ1X6yfQ (https://www.youtube.com/watch?v=LsyjQ1X6yfQ)

but it is monophonic (chords are recreated in the end)  ::)

i am studying another version
  =>  import midifile (done)
  =>  save it as text    (done)
  = > with gentext generate new text , and play it as midiOuttmsg (done)
Title: Re: not neural network text and music generation
Post by: chung on April 26, 2019, 02:00:06 pm
(26/04/2019) i have added to brainpiano3 of midipiano_chung  save as miditext txt files from imported midifiles (save with  [A]noautochord selected)

then when loaded these txt files gentext_chung generates mid music sent to the universal default midi mapper port instead of speach .   ::)


new video => https://www.youtube.com/watch?v=egfXspTBeJI (https://www.youtube.com/watch?v=egfXspTBeJI)
Title: Re: not neural network text and music generation
Post by: LOCKSUIT on April 26, 2019, 03:45:30 pm
so it uses blankies/blank notes, and once the output is done, it spices up/converts the notes into cool sounds like piano or synth? Can you show us dark synths like drum or vibes?
Title: Re: not neural network text and music generation
Post by: chung on April 26, 2019, 08:26:49 pm
for now i just send noteon / noteoff midimessages , so it plays the piano on the default midi mapper out port (i prefer the piano sound) .
if you are interested i can add send output to a midi input selection port so you can play another instrument .

the last version includes time data in the text strings generated so it is polyphonic with any tempo and velocity.
Title: Re: not neural network text and music generation
Post by: LOCKSUIT on April 26, 2019, 08:43:46 pm
I'm not sure yet I'm interested enough in the outputs compared to the music I would like to hear, but, maybe it is interesting if I heard it with other sound mask, can you show us an output of a techno like music using this sound mask below (or something similar)?:

https://www.youtube.com/watch?v=YCT2VUXjJTg&feature=youtu.be

So, instead of hearing "piano, pianopianopiano, piano,,, piano"
we hear
"boom, boomboom,,, boom boom boomboomboom, boomboom"
or
"vvbvv,vbbvvvv,vvv,,,,vvv,vv,vvvvv,,vvvvv,vvbbvvvv,"

:)))

I have another idea
say our output is:
Note_Note_Note_Note_Note_Note_Note_Note
how about we make every other a 2nd type of instrument sound like this, 2 instruments!:
boom_whoosh_boom_whoosh_boom_whoosh
Therefore we get:
Boom......whoosh..boom.........boom boom BOOM.....WHOOSH boom WHOOSH.........boom whoosh
Title: Re: not neural network text and music generation
Post by: chung on April 26, 2019, 09:53:27 pm
will keep it like it for now  ::)

just added a combobox to gentext for selecting the instrument (grand piano, electric piano, ... vibes ,...,  gunshot ) 


=>  https://www.youtube.com/watch?v=cDDf7f7ToI0 (https://www.youtube.com/watch?v=cDDf7f7ToI0)
=>  https://www.youtube.com/watch?v=qC56waG52fE (https://www.youtube.com/watch?v=qC56waG52fE)
Title: Re: not neural network text and music generation
Post by: ivan.moony on April 27, 2019, 11:58:12 am
Hello Chung :)

Did you think of offering this to someone in game industry? Or maybe as a pluggable library for embedding in C, javascript, flash, or something?

I remember some attempt in good old days where music mood followed the game action intensity. It was some samurai game or something, but back then there were no more developed neural networks, so I guess it had to be some music-transition advanced if-then switching mechanism. I think it was about year 1995., but I don't remember the game name. Anyway, the concept was used as a curiosity to promote the game. The game ended up described even in one Croatian monthly computer magazine...
Title: Re: not neural network text and music generation
Post by: chung on April 27, 2019, 04:55:43 pm
well, for now i'll have a little rest ...  ::)
Title: Re: not neural network text and music generation
Post by: chung on May 02, 2019, 03:53:33 pm
(02/05/2019) many bugs corrections + new look  ::)
Title: Re: not neural network text and music generation
Post by: goaty on May 02, 2019, 03:59:45 pm
The way I get my little critter to walk splitbrained is by forming sequences of the other leg,  very similar to this.  :)

.
Title: Re: not neural network text and music generation
Post by: chung on May 03, 2019, 09:02:34 am
as my chatbot says : "thats too complicated for me ..."

(04/05/2019) genmidi_chung version with midi data size x3 (adds transpose +2, -3 )  ::)
Title: Re: not neural network text and music generation
Post by: chung on May 06, 2019, 04:05:32 pm
(06/05/2019) genmidi_chung : works with notes variations (pitch independent) (transposes removed)  ::)
Title: Re: not neural network text and music generation
Post by: goaty on May 06, 2019, 06:34:17 pm
If you want the robot to talk,  you can look 20 words in front, for a million leaves to check for a reward.  8)

//EDIT//
actually that's not true, that's only if it was morse code, and its per bit.   20 bits then.

and my walky bot, I need knn to match my smaller database, because its 3 space vectors and I need a good approximate result without too much time needed for training.
Title: Re: not neural network text and music generation
Post by: 8pla.net on May 08, 2019, 01:49:47 am
Thank you, Chung Nguyen.

I have been following your work with A.I. for a few years now.
Title: Re: not neural network text and music generation
Post by: chung on May 15, 2019, 10:39:02 am
(07/05/2019) midiTotext small program to export all folder midi/.mid midifiles to midi/.txt files

use then _copyAlltxt.bat program in text/merge folder to merge all txt files in folder to all.txt

(see README.txt )  ::)

(15/05/2019) algo1-5 added
Title: Re: not neural network text and music generation
Post by: chung on May 16, 2019, 12:40:43 pm
(16/05/2019) genmidi_chung_dll library dll version of genmidi_chung added

this is not AI but is not bad too => https://sourceforge.net/projects/minimidichung/ (https://sourceforge.net/projects/minimidichung/)