mathildebuenerd / music-emotion-classifier

Music Emotion Classifier

Description

Classifies musics and sounds into 4 emotions: happy, calm, angry or sad, depending on the music features, using Tensorflow.js.

The classification is based on a dataset of already classified sounds by danz1ka19.

Usage

1. Extract the features of your sound files

• Download the Music-Emotion-Recognition repo.
• Create a "Dataset" folder at the root of the folder, and put the mp3 files you want to classify in it.
• Launch the Feature-Extraction.py script with the command line (might take few minutes if you have a lot of files).

py Feature-Extraction.py

• It generates an Emotion_features.json file at the root of the folder.

2. Add the sound features

• Download this repo.
• Run npm install.
• Place the generated Emotion_features.json file in the toClassify folder.

3. Tune the model

• Change the default options of the model in scripts/classifier.ts, at the declaration of the classify() function, or directly with the form (next step).

4. Run

• Run with using npm start at the root of the project.
• You can access the results in the console at http://localhost:1234, with Google Chrome.
• You can also tune the model here, by entering your own parameters and pressing the "Classify" button. There is no checking for the values entered here so be careful.
• The results will appear in the console.

Credits

Development by Mathilde Buenerd.

Based on this dataset and feature extractor by danz1ka19.