custom_tokenizations

Tokenization Explanations

RelTok

2D tokenization that replaces bar and position tokens with a single token called same_or_not token. By using the cumulativeness of music, we can think notes (including monophonic ones) and rests as they are in a stack. All we need to pay attention to is whether they share the same time or not, their exact position can easily be calculated by cumulative duration.

ABCD

Symbols means:

• a go up 0.5 interval.
• b go down 0.5 interval.
• c note player.
• d note separator.

NOTE: Right now implementation uses TokSequence.bytes to store the string abcd representation. To modify this feature,, one must subclass TokSequence and add another representation similar to ids, bytes, toks etc.

Usage

Uses miditok as backend, therefore you may use them just like official miditok tokenizations.

You may refer to recipes.py for tokenizing.

Test

Run the following:

python -m unittest test/test_tokenizations.py 

This will create output files under output/test

Build

Run the following:

python -m build .