https://github.com/d-dawg78/mva_dl
Master MVA - Deep Learning project
https://github.com/d-dawg78/mva_dl
bert cnn fasttext genre-identification glove gru lstm lyrics word-embeddings
Last synced: 5 months ago
JSON representation
Master MVA - Deep Learning project
- Host: GitHub
- URL: https://github.com/d-dawg78/mva_dl
- Owner: d-dawg78
- Created: 2020-12-06T23:13:56.000Z (almost 5 years ago)
- Default Branch: main
- Last Pushed: 2023-06-22T12:35:09.000Z (over 2 years ago)
- Last Synced: 2025-06-15T07:02:30.059Z (5 months ago)
- Topics: bert, cnn, fasttext, genre-identification, glove, gru, lstm, lyrics, word-embeddings
- Language: Jupyter Notebook
- Homepage:
- Size: 32.8 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
#### Final Project Submission for the Deep Learning course at the École Normale Supérieure (Master MVA)!
#### Group: Dorian Desblancs, Kodjo Mawuena Amekoe
#### Abstract:
In this project, we explore the field of musical genre recognition using lyrics. We first introduce a 10000 track dataset that spans 12 genres. The lyrics present in it are then classified using a multitude of popular NLP algorithms such as CNNs, Bidirectional LSTMs, and Bidirectional GRUs superimposed on pre-trained GloVe or fastText word embedding layers. We also report results obtained using a pre-trained BERT model. Our best model obtained an average test set accuracy of 70%. This result is close to those obtained by more elaborate audio-based classifiers on much smaller datasets. It also surpasses past benchmarks on lyrical genre recognition tasks.