Infinite Remixer is a Python application that creates remixes, patching together multiple songs.
To install the package, move to the root of the repo and type in the console:
$ pip install .
If you plan to develop the package further, install the package in editable mode also installing the packages necessary to run unittests:
$ pip install -e .[test]
To run unittests, issue the following command from the root of the repo:
$ pytest
The package is divided into 4 sub-packages:
- segmentation
- data
- search
- remix
segmentation is responsible to segment groups of songs into beats and store the beats as separate wav files.
data can be used to batch extract audio features from a directory containing audio files, aggregate the features, and storing the data ready to be consumed by search.
search contains facilities to fit a scikit-learn NearestNeighbour object, and a wrapper class around this object to apply nearest neighbour search.
remix is responsible for generating new remixes.
To generate a remix, there are two main high-level steps to carry out. First, preprocess a group of songs you want to remix. Second, generate a remix, which will leverage the preprocessed data.
To preprocess data follow these steps:
1- Segment a group of tracks into beats 2- Extract features and prepare data from the beats 3- Fit a Nearest Neighbour object
To run the steps above, use the entry points below. You can find more info on the entry points, in the respective modules.
$ segment path/to/dir/with/files path/to/save/dir/for/beats
$ create_dataset path/to/dir/with/audio/files save/dir
$ fit_nearest_neighbours dataset/path save/path
Once you have gone through the preprocessing steps, you're ready to create remixes with the command below:
$ generate_remix 0.1 50 save/path/example.wav
The first positional argument is the jump rate. It's a value between 0 and 1, which indicates how frequently the system should jump from one song to another. The greater the value the higher the chance to jump to a new track.
The second positional argument indicates the number of beats in the remix.
In order to load the necessary artifacts for running the remix, you'll have
to change the paths to your artifacts directly in the top part of the
remix/generateremix.py script.
Infinite Remixer uses the packages below to extract audio features and for nearest neighbour search:
- librosa
- scikit-learn
Learn more about Infinite Remixer in this project presentation video on The Sound of AI YouTube channel.