Skip to content

karlaspuldaro/AI_Lyric_Composer

Repository files navigation

Heavy Metal Composer AI

Artificial Intelligence Heavy Metal Songwriter built under a Long Short-term Memory (LSTM) Recurrent Neural Network (RNN) architecture. Song lyrics originally obtained from the Kaggle dataset

*Note: Original full dataset no longer available. lyrics-ds.txt is a filtered dataset which includes over 1k songs from a selection of 10 artists (described in dataprocessing.py).


Requirements

  • Conda
  • JupyterLab
  • Elyra
  • Numpy
  • Tensorflow

Installation

  1. Configure conda virtual environment for python:

    This will help managing dependencies and isolate our project

    conda create -n myenv python=3.7

  2. Activate the environment:

    conda activate myenv

  3. Install dependencies:

  • numpy:

    conda install -c anaconda numpy

  • tensorflow:

    conda install -c conda-forge tensorflow

  1. Install JupyterLab and Elyra:

    conda install -c conda-forge jupyterlab

    conda install -c conda-forge elyra

  2. Build JupyterLab

    jupyter lab build

  3. Verify Installation

    jupyter serverextension list && jupyter labextension list


Build and run JupyterLab

jupyter lab build

In the root of your clone of this github project, run jupyter lab


Open JupyterLab and run the model

Once JupyterLab is launched in your web browser, all files from this repository will be loaded and are accessible from the File Browser on the left pane of JupyterLab's main page.

Select composer-notebook.ipynb file and open it. Run the notebook (Run tab --> Run All)


Model training

This repository includes 2 trained models, saved as checkpoints in hdf5 files.

To re-train the model, from a terminal, run

composer.py train <optional_checkpoint_file>

Training mode will create a checkpoint file after every epoch. The latest file can be passed to this command, the model resumes its training from there.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published