A machine learning approach to predict the mood of the music or a song and using that predicted mood to create a customized playlist from a list of available songs in local directory.
We have taken 518 songs and extracted 18 features from them using librosa and saved in dataset.csv file. Mean,standard deviation and variance of all those 18 features is calculated as well. By doing that total number of input features in the dataset are 54. To manually label these songs with ease, we have relied on the top spotify playlists which are created according to the moods. Some of them are:
-
Happy Songs
-
Relax Playlists
-
Angry Playlists
-
Sad Playlists
The issue of a song belonging to one or more category is resolved by assigning them to a particular mood category out of four.
| Models | Training accuracy | Testing accuracy |
|---|---|---|
| Logistic Regression | 88.12% | 82.05% |
| Naive Bayes | 78.45% | 76.28% |
| KNN | 78.17% | 78.84% |
| SVM | 87.84% | 76.92% |
Python 3.7 or any other version above than Python 2.7
https://realpython.com/installing-python/
sklearn
pip install -U scikit-learn
pandas
pip install pandas
numpy
pip install numpy
matplotlib
pip install matplotlib
pickle
pip install pickle
librosa
pip install librosa
tkinter
apt-get install python-tk
pygame
pip install pygame
mutagen
pip install mutagen
ttkthemes
pip install ttkthemes
Just install all the prerequisites and run the musicplayer.py file. A GUI will open, Add your song and wait for few seconds to get the mood of the song. After detection of mood, All the songs having similar mood will be added automatically.