"Music sequence generation of Python image features" on the idea transformation of music generation

In the seminar a few days ago, a brother pointed out that the original method is really difficult to train, so I improved the idea of music generation.

First of all, I used lstm to generate a sequence of musical expressions, then I can use some of the data set to train lstm. To avoid a lot of duplication between the generated music and the existing music, we can consider changing the lstm to make it more "forgetful", which should solve some of the problems. There will definitely be more problems to be revealed, and I will continue to explain my ideas.

First, find a suitable data set. Open it

http://www.mln.io/resources/datasets/

Then I'll explain the usage of these datasets:

The first classical music Datasets, which stores a lot of midi classical music. Combined with the ABC and MIDI format conversion tool, you can extract a serialized format from the ABC format.

The conversion tool address is as follows: Https://github.com/leesavide/abcmidi

The second is I intend to collect, called Nottingham Music Database, but unfortunately the data download has expired, I found the data set as follows:

Http://www.chezfred.org.uk/freds/music/tunes/index.htm

is saved in a special format.

The third musedata is also a music data set, which is described below:

The musedata database is a project of the Center for Computer assisted the Humanities (Ccarh). The database was created by Walter Hewlett. Data entry have been primarily done by Frances Bennion, Edmund Correia, Walter Hewlett, and Steve Rasmussen.

offers a variety of formats for storage, more classical music.

The fourth one seems to be hanging.

The fifth is mainly to do music recommendation, it seems unable to know the original melody sequence.

=====================

The preliminary decision is Nottingham Music Database, there are two formats to choose, the first is the ABC format, as follows:

X:1t:a and D% Nottingham Music databases:efm:4/4k:am:6/8p:af| " A "ECC c2f|" A "ECC c2f|" A "ECC c2f|" Bm "BcB" E7 "b2f|" A "ECC c2f|" A "ECC c2c/2d/2|" D "Efe" E7 "dcb| [1 "A" Ace a2:| [2 "A" Ace ag=g| | K:dp:b "D" F2F fdd| " D "AFA f2e/2f/2|" G "G2G ecd|" Em "EFD" A7 "cba|" D "F^ef dcd|" D "AFA f=ef|" G "GFG" A7 "ABc | |" D "D3 d2e:|2" D "D3 d2| |

ASCII code:

. Mstitle= "\f3a and D\FP"; rtitle = "\F2EF\FP"; key = a;bps = 16;autobeam;chords;bars = 18;timesig = 6 8.f^ ' A ' |e^ "a" c^ c^ C ^>  f^ |  e^ "A"  c^ c^c^> f^  |e^ "a" c^ c^ c^ b A  |  B "Bm"  c^ b b> "E7" f^ |e^ "A" c^ c^ c^>  f^ |  e^ "A"  c^ c^  c^> c^< d^< |e^ "D" f^ e^  d^ "E7" c^ b  |a "a" ' \1st. '  c^ e^ a^>: |a "a" ' \2nd. '  c^ e^ a^ g^ g=^ |! \endstave.\key = d. f^> ' B ' "D" f^      d^>  d^  | A "D"    F  a    f^> e^ < f^< | G^> "G" g^  e^  c^ d^  | e^ "Em"  f^ d^   c^ "A7"  b   a  |f^ "D" e+^ f^ d^ c^ d^  | A " D "F  a   f^ e^  f^ | g^" G "  f^ g^ a" A7 "b c^ |d^>. ' \1st ' "D" d^> e^: |d^>. ' \2nd. ' "D" d^> |! \endstave. ME

This analysis has yet to be studied.

"Music sequence generation of Python image features" on the idea transformation of music generation