https://github.com/tripathiarpan20/midiformers

Applying Transformers to MIDI music for various tasks
https://github.com/tripathiarpan20/midiformers

ai-music-generator deep-learning music music-composition

Last synced: about 1 month ago
JSON representation

Applying Transformers to MIDI music for various tasks

• Host: GitHub
• URL: https://github.com/tripathiarpan20/midiformers
• Owner: tripathiarpan20
• Created: 2022-07-02T06:05:47.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2022-09-03T08:49:36.000Z (over 2 years ago)
• Last Synced: 2024-08-05T06:05:47.471Z (5 months ago)
• Topics: ai-music-generator, deep-learning, music, music-composition
• Language: Python
• Homepage:
• Size: 703 KB
• Stars: 46
• Watchers: 3
• Forks: 5
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# MIDIformers

## Table of Contents

- [MIDIformers](#midiformers)

  - [Table of Contents](#table-of-contents)

  - [DeepMixer ](#deepmixer-)

    - [Installation ](#installation-)

    - [Notebook ](#notebook-)

    - [Output samples ](#output-samples-)

  - [Support](#support)

## DeepMixer  

The project involves the usage of the open-source [MusicBERT](https://github.com/microsoft/muzic/tree/main/musicbert) model to perform mask prediction tasks for MIDI task with customizability.

### Installation  

```

git clone https://github.com/tripathiarpan20/midiformers.git

cd midiformers/DeepMixer/models/musicbert

./setup.sh

```

### Notebook  

The live version of Colab notebook utilising the scripts in `DeepMixer/models/musicbert` can be accessed from this link : 







The notebook supports customisability on top of the original MusicBERT codebase, like masking chosen percentage of random notes from either whole MIDI stream/ notes from selected instruments based on user preference. 

Other features include: 

- [x] Option to leaving notes from the beginning `min_bar_mask` masks out of the masking pool to provide more initial context for mask prediction.  

- [x] Prediction modes with trade-off between speed and quality of predictions.

- [x] Sampling strategies like Temperature, Top-k and Nuclues (Top-p) added for mask predictions. 

- [x] Filtering invalid prediction for more consistent results.  

- [x] Song segment selection and multi-program/ins masking. 

### Output samples  

Some of the samples from the above notebook along with the reference pieces can be found in a [Drive folder](https://colab.research.google.com/drive/1pPFn-HhH7nZvfbWQlwEne7mm1uc2adOV?usp=sharing) , the songs are copyrighted by the respective owners.

A few of our favorites are embedded below:  

* Shock (Attack on Titan):

    - Original & Remix: 

https://user-images.githubusercontent.com/42506819/178094976-a5763a31-70bf-4a17-b9e9-1255c0b9a1f0.mp4

    

https://user-images.githubusercontent.com/42506819/178094989-441471d2-78ce-4b6a-8e88-bd01aeab810e.mp4

* Bohemian Rhapsody (Queen):

    - Original & Remix: 

    

 

https://user-images.githubusercontent.com/42506819/178095093-1e1192bc-76e5-4027-b733-d63bed9ac206.mp4

https://user-images.githubusercontent.com/42506819/178095109-b9a38d56-3b69-4169-9ed1-3eb82a47276a.mp4

* Unforgiven 2 (Metallica): 

    - Original & Remix: 

    

https://user-images.githubusercontent.com/42506819/178095316-6d19c803-3407-4bbd-81e1-0dc476b680d6.mp4

https://user-images.githubusercontent.com/42506819/178095326-f6961d98-c574-4ee2-8d27-27efead79409.mp4

## Support

There are many ways to support a project - starring⭐️ the GitHub repo is just one.