Discussion-Beginners

[ashwani-rathee]

Tiny Doubt:- Does this look how it should look??

[boblsturm]

Hello. You need to remove all the spaces before converting to staff notation, or converting to MIDI.

[ashwani-rathee]

Ok I changed that,I found it a little confusing to convert abc to mid

Notes for people who are exploring this in ubuntu: The result through the program was: X:0 M:4/4 K:Cdor c 3 d c 2 B G | G F F 2 G B c d | c 3 d c B G A | B G G F G 2 f e | d 2 c d c B G A | B G F G B F G F | G B c d c d e f | g b f d e 2 e f | d B B 2 B 2 d c | B G G 2 B G F B | d B B 2 c d e f | g 2 f d g f d c | d g g 2 f d B c | d B B 2 B G B c | d f d c B 2 d B | c B B G F B G B | d 2 c d d 2 f d | g e c d e 2 f g | f d d B c 2 d B | d c c B G B B c | d 2 c d d 2 f d | c d c B c 2 d f | g b b 2 g a b d' | c' d' b g f d d f |

The problem is instantly converting it to mid file which is easier to handle acc to me. This site helped quite a lot: http://clivew.com/abc.php All other ways like abc2midi,easyabc need a lot to do first . Put the result and it generates the mid right there and allows you to download it. mid files can be directly loaded to audacity using the import for mid,but It can't be listened to(as much as I know,I may be wrong there) So we kind of need to convert it to wav file(I don't know how to convert it to mp3 directly :) Install timidity for conversion to wav files...

timidity -Ow filename.mid

It converts mid file to wav file which can easily be converted to mp3 file using ubuntu's sound converter.. which can be loaded to audacity and can be explored :) Hey bob,there seems some kind of error while I convert mid to wav file causing those spikes in the samples in the beginning,how can I avoid them?? Edit: folkrnn.org does it all(sheet,midi download),I searched for this a hour and answer was right here :)

[ashwani-rathee]

I want to try my hand at training for double jigs,I believe the ones with M: 6/8 are double jigs in the training data(v2_data file)??

[boblsturm]

For those using Linux/Unix, do the following for converting the output of folk-rnn. First install the abctools (https://ifdo.ca/~seymour/runabc/abcguide/abc2midi_guide.html). You need abc2midi. Also install timidity.

1. Create a file with a .abc extension, e.g., temp.abc, and open in your favorite text editor.

2. The first line of the file should be X:0.

3. Paste in the output from folk-rnn, and make sure the M: and the K: fields are on lines of their own.

4. In the transcription, remove all spaces. I do this with a sed command in vi, i.e., :%s/ //g The transcription above appears like this: X:0 M:4/4 K:Cdor c3dc2BG|GFF2GBcd|c3dcBGA|BGGFG2fe|d2cdcBGA|BGFGBFGF|GBcdcdef|gbfde2ef|dBB2B2dc|BGG2BGFB|dBB2cdef|g2fdgfdc|dgg2fdBc|dBB2BGBc|dfdcB2dB|cBBGFBGB|d2cdd2fd|gecde2fg|fddBc2dB|dccBGBBc|d2cdd2fd|cdcBc2df|gbb2gabd'|c'd'bgfddf|

5. Close the file.

6. convert the abc file to midi, e.g., abc2midi temp.abc. There may be counting errors, but ignore.

7. To hear the output, use timidity, e.g., timidity temp0.mid.

[boblsturm]

Double jigs are in 6/8, but not all transcriptions in 6/8 are double jigs. The 365 double jigs in O'Neill's 1001 are here: http://www.norbeck.nu/abc/book/

[ashwani-rathee]

Hey after a day of experimenting with folk-rnn..I wanted to ask a few things,as a beginner I only know how to use and train models from existing resources and see how it sounds..I haven't learnt how to develop AI system yet..

• How many abc notated songs did the team use for the training data in file data_v2??
• It seems folk-rnn(v2) model produce only 4/4 probably or maybe I can't tell the difference,is it so??
• Were 9/8,6/8,4/4,3/4 notations randomly sprinkled in training data of data_v2 or any particular way(there seems few 6/8 then few 4/4 then 6/8 and so on)?? Ofcourse as this resource is under MIT license,it would be fun to use it and things I would love to implement in it is direct conversion to midi,producing multiple mid files at same time,saving it all in a particular directory(partly because magenta does that,makes it much more accessible.. :) though I know it is easier said than done...

[boblsturm]

Hello.

• There are several versions of folk-rnn that we have created (see here: https://github.com/IraKorshunova/folk-rnn/tree/master/data). folk-rnn (v2) was trained on all of data_v2, which has 23,636 transcriptions.
• folk-rnn (v2) does not produce only M:4/4. You can see 30,000 transcriptions generated by v2 here: https://highnoongmt.wordpress.com/2018/01/05/volumes-1-20-of-folk-rnn-v1-transcriptions/ You can also play online with v2 here: https://folkrnn.org/ (It's the model thesession.org (w/ :||:)
• folk-rnn (v2) is trained using SGD with minibatches of size 64. Minibatches are created without regard to the meter, but do consider the lengths of the transcriptions. Each minibatch should have transcriptions that have about the same number of tokens. We did this to make the training more efficient.