Downloading music is done by many different people for many different reasons, whether it be for backups, data hoarding, or to listen to music while offline. However, those who do download music, come across the issue of organizing all the files into categories that they can reference when needed. The most known way to organize music is by genres. The issue with genres however, is that they are set by humans and liable to be incorrect, or there could be no genre assigned to the audio file based on how it was retrieved. Currently, if you want to find what genre a song belongs to, you have to manually search for it online and assign it to the song through a tool. While this may work for a short list of songs, if you have a larger list (say about 1000 songs), this process becomes much more time consuming. MusiSort aims to fix this issue by automatically sorting songs into categories based not on genre, but by musical similarity. It accomplishes this task by utilizing new and known musical analysis methods to generate data and feed that information to K-Means Clustering to generate categories. Through this, MusiSort can take a large list of audio files, and separate them by melodies, tempo, mood, and more.
MusiSort is a tool being developed to collect music and put them into similar groups or clusters based on their waveform. The program uses artificial intelligence to check similarities and differences between the different songs. The main goal of the project is to create a tool which removes the need to manually sort music into different genres as this can be quite a difficult, and tedious, task.
As MusiSort is a pip tool, it can be easily installed with the help of Python and Pip through the following command:
pip install musisort
MusiSort files are stored under the user directory as musisort. For instance, on windows, C:\\Users\\ComputerName\\AppData\\Local\\user\\musisort.
To upgrade MusiSort to its latest version, run the command pip install --upgrade musisort.
MusiSort Command Line Interface
There are a total of 7 different commands currently available to use in musisort. The following is a list of each command:
- musisort help
- musisort info
- musisort dir
- musisort config
- musisort analyze
- musisort classify <list_name> <cluster_count>
- musisort debug
help : displays the commands in musisort along with a description of what each one does.
info : when used with a command as an argument, displays a more in-depth description of the command compared to the help menu.
dir : lists all directories on the system associated with MusiSort.
config: when used without arguments, shows current config values in the terminal. If used with the key and value arguments, changes that key's value in the config.
analyze: retrieves song data from the folder all\songs and uses the classes under the analysis_methods folder in MusiSort's code to analyze each song.
classify: takes song's stored in the list/folder specified in the arguments and clusters them based on the amount given in the command. To classify, each song in the list must first be analyzed with the analyze command.
debug : uses a song located in the debug folder of MusiSort's directory to run tests and check MusiSort's performance\accuracy.
Using MusiSort to Classify Songs
- Open terminal and run the command
musisort dir. Go to the directory located under theAll Songs List FoldercalledSongs Folder. An example of this path on Linux would be :/home/user/.local/share/musisort/songlists/all/songs. - Add all songs you want to sort into this folder.
- Run
musisort analyze. The first time this is run, it will go through every song and gather data about it. Everytime you add a new song toall/songs, this command will need to be run again to analyze the newly added songs. - After the analysis is completed, run
musisort classify <song_list> <catergory_count>. For this example, we will usemusisort classify all 5. Since our songs are located in the directorysonglists/all/songs, the list name isallas that is the name of the folder the songs are located in. The 5 means we will have categories ranging from 0 to 4 returned to us for each song. If the value -1 is used for thecategory_countargument, the program will attempt to automatically assign a number of categories. Once the program has finished classification, it will display the categories to the console in a list as so:
If using the windows command prompt, unicode will not display in the final classification output. Rather than viewing results in the command prompt, you can view the results in the categories folder for the list. The category folder for the all list is located in the MusiSort directory ./musisort/songlists/all/categories. The list of labels can be viewed by opening the text files named labels_n.txt, with n referring to the amount of labels classified, within the folder using a text editor that supports unicode.
Data Used for Testing
https://drive.google.com/drive/folders/1kCke4O5IVPndeUuDvmzUe2n7M6pokksR?usp=share_link
The folder containing data is split into two sections. The folder Accuracy Song contains the audio file used to test the program using the musisort debug command. To use the debug command using this file, drag the song located in the folder into the songlists\debug\songs folder in the MusiSort directory. Then you can run musisort debug to get data related to the accuracy of the classification and analysis methods.
Inside the Efficiency Data folder, 4 different zip files contain 1, 5, 10, and 25 songs. The speed of the program was determined by placing various sys.time checks within the code of MusiSort and running the program with these set amounts of audio files. This can be done by cloning the MusiSort repository, and then running pip install . or pip install --upgrade . in the cloned directory when adding time checks in the locally stored code.
[✅] Develop the algorithm to sort songs into categories.
[✅] Optimize the algorithms used to sort songs for faster completion.
[✅] Create a more user friendly terminal interface for easier usage.
[✅] Display list of labels in terminal after classification has finished.
[✅] Develop framework for testing accuracy of data analysis methods and classification. - Current Release 1.0
[🏗️] Add option to append classification labels onto audio file's metadata genre tag.
[🏗️] Explore implementation of other clustering methods.
[❌] Develop a GUI for more interactivity with the program.
(🏗️ : in progress , ❌ : not started yet , ✅ : completed)
For users who want to further develop MusiSort's code, a different approach to installation is needed to build the program. As with most open source projects, the first step is to fork, and then clone MusiSort's repository on Github to a local system. If you want to contribute to MusiSort directly, instead of creating a new version of it, you can create a branch in the original repository, clone the branch, and then create a pull request for finished releases. The link to the Github repository is located at the top of this thesis. Now that the program's code has been successfully copied to your device, you can modify it using any code editor of your choice.
Since MusiSort is a pip package, you cannot run the main.py file directly, but you will instead need to run a few commands to build your modified version. If you have MusiSort already installed on your system, from using the command pip install musisort, you will need to remove it. You can uninstall the currently installed version through the command pip uninstall musisort. After this is completed, you can begin the process to build the updated code you created. First, open your terminal and navigate to the cloned directory within your file system. This can be done through the command cd <directory>. After you have entered the directory in your terminal, type pip install . --upgrade. This command installs the pip package located in the current directory. Since the argument --upgrade is used in the command, it will automatically remove the currently installed version of the program and install the latest version you have created. After the program has finished installing, test its installation by running the command musisort. If no errors occur, and a help menu appears in the terminal window, you have successfully installed your custom version of MusiSort.
Spotify's 15 Personalized Mood Filters
Sony’s 12 Tone Analysis
https://www.sony.com/electronics/support/articles/00009093
Jiang, Yanru & Jin, Xin. (2022). Using k-Means Clustering to Classify Protest Songs Based on Conceptual and Descriptive Audio Features. 10.1007/978-3-031-05434-1_19.
Kim, Kyuwon, et al. “Clustering Music by Genres Using Supervised and Unsupervised Algorithms.” Stanford, pp. 1–5.
https://cs229.stanford.edu/proj2015/129_report.pdf
- Librosa : https://librosa.org/doc/latest/index.html
- Sklearn : https://scikit-learn.org/stable/
- Scipy : https://scipy.org/
- Numpy : https://numpy.org/
- Appdirs : https://pypi.org/project/appdirs/
- Progress : https://pypi.org/project/progress/