AccoMontage2

AccoMontage2

Introduction

AccoMontage2 is a system capable of doing full-length song harmonization and accompaniment arrangement based on a lead melody. Based on AccoMontage, AccoMontage2 invents a harmonization module to generated chord progression, and provides a GUI to control the generating styles of chords and textures. Our paper AccoMontage2: A Complete Harmonization and Accompaniment Arrangement System is accepted by ISMIR 2022. This repository stores the code corresponding to the paper.

Binary Data Files Download

You might have to download some data files and store them in chorderator/static before running. You can find them here.

Dataset Download

If you want to explore our dataset or use it for your own project, you can find it here (For those who don’t have access to Google, you can use the mirror at Baidu Netdisk). Note that you don’t need to download them if you only want to run AccoMontage2.

The dataset is re-organized from the original Niko Dataset. It contains 5k+ chord progressions with style labels. We provide our dataset in two formats:

1. MIDI: each chord progression piece stored as a single MIDI file.
2. Quantized Note Matrix: a python dictionary with format like the following. nmatis an 2-d matrix, each row represent a quantized note: [start, end, pitch, velocity]. Each note is quantized at the eighth note level. eg., start=2 means the note begins at the third eighth note. root is also an 2-d matrix. It labels the roots of the chords using an eighth note sample rate. Each row of the root represents a bar. Each element is an integer ranged from 0 (C note) to 11 (B note).

{'piece name': 
 	{'nmat': [[0, 3, 60, 60], ...],    # 2-d matrix: note matrix
     'root': [[0,0,0,0,0,0,0,0], ...], # 2-d matrix: root label
     'style': 'some style',            # pop_standard, pop_complex, dark, r&b, unknown
     'mode': 'some mode',              # M, m
     'tonic': 'C'                      # C, Db, ..., B
    }, 
 ...
}

# load the dataset using pickle
import pickle
with open('dataset_path_and_name.pkl', 'rb') as file:
    dataset = pickle.load(file)

Interfaces and Demo

Run with terminal

A demo is provided. If you want to run demo, please check all the requirements in requirments.txt are satisfied, and run

python demo.py

This demo takes in the melody midi file hpps65 and output the result in folder hpps65_output_results.

# demo.py

import chorderator as cdt

if __name__ == '__main__':
    
    demo_name = 'hpps65'
    input_melody_path = 'MIDI demos/inputs/' + demo_name + '/melody.mid'

    cdt.set_melody(input_melody_path)
    cdt.set_meta(tonic=cdt.Key.G)
    cdt.set_segmentation('A8B8A8B8')
    cdt.set_texture_prefilter((0, 2))
    cdt.set_note_shift(0)
    cdt.set_output_style(cdt.Style.POP_STANDARD)
    cdt.generate_save(demo_name + '_output_results')

Interfaces are provided as follows:

More detailed controllability is provided. Please refer to the Core Class in chorderator/core.py. A core object can be instantiated by calling cdt.get_chorderator().

In addition, if you have problems trying AccoMontage2 with a custom melody, please refer to issue #2. There is also an interactive tutorial available, credit to @asigalov61.

Run with GUI

Please check all the requirements in back-end/requirments.txt are satisfied, and run

python back-end/app.py

You can interact with the GUI at http://127.0.0.1:5000.

Reference and Acknowledgement

Thanks to Prof. Gus Xia for his guidance. Thanks to Jingwei Zhao and the AccoMontage system that provides solid foundations for this research. His repository can be found at AccoMontage. Thanks to all members at New York University Shanghai Music-X-Lab for their generous support.

Cite Our Work

@inproceedings{yi_2022_accomontage2,
  author       = {Li Yi and Haochen Hu and Jingwei Zhao and Gus Xia},
  title        = {{AccoMontage2: A Complete Harmonization and Accompaniment Arrangement System}},
  booktitle    = {{Proceedings of the 23rd International Society for Music Information Retrieval Conference}},
  year         = 2022,
  pages        = {248-255},
  publisher    = {ISMIR},
  address      = {Bengaluru, India},
  month        = dec,
  venue        = {Bengaluru, India},
  doi          = {10.5281/zenodo.7316642},
  url          = {https://doi.org/10.5281/zenodo.7316642}
}