Musicaiz

What's new

- Minor fixes
- Added algorithm for structure/form segmentation of MIDI files by levels

What's new
- Fix bugs in MMM tokenizer
- Added arg in Pianoroll to save the plot
- Add structure annotations parser for BPS dataset

What's Changed
- Refactor in loaders -> Musa. Now notes are not grouped in instruments or bars but they are referenced to them with attributes.
- Build beats and subbeats objects when loading a MIDI file with the Musa class in laoders.
- Support tempo and time signature changes.
- Refactor pianoroll plots to support multi-track plotting.

What's New
- REMI tokenizer (with thempo and time signature changes).
- Chord detection algorithm at beat-level.
- Add graph representation of symbolic music (features module).

What's new

- [Key profiles](https://github.com/carlosholivan/musicaiz/blob/main/musicaiz/algorithms/key_profiles.py): Krumhansl-Kessler, Temperley, Albretch-Shanahan
- Algorithms: Addition of a change scale algorithm
- MMMTokenizerArguments: Initializes arguments for MMM tokenizer
- Datasets: Include composers, periods and forms in Maestro, refactor code
- Converters: Add protobufs
- Tests: Add more tests for datasets and tokenizers

What's new

- `datasets` module to tokenize MAESTRO, Lakh MIDI and JSB chorales datasets
- `models` module with a GPT-based model to train and generate symbolic music