Intelligent Lyric Generator of Different Genres

Project Description

Welcome to visit the homepage of our intelligent lyric generator.

Main Functionality

- The main functionality of this project is to automatically generate the lyric based on the keywords extracted from users' input sentence and the genre users choose.

DataSet

- We chose a raw lyric dataset from Kaggle which contains nearly 380,000 songs. We designed an algorithm to clean and split the data, part of which were used to train the model.

Keyword Extraction

- The keyword extraction algorithm [RAKE](https://github.com/ruby/rake) was used for our keyword extraction module.

Seq2Se2 Model

- In this project, we used a improved Seq2Seq model which is similar to the model of the research paper [Chinese Poetry Generation with Planning based Neural Network](https://arxiv.org/pdf/1610.09889.pdf).
  We trained the improved Seq2Seq model by LSTM and GRU and found GRU outperformed LSTM for this certain task.

Project Structure

src

- data_clean.py wraps the data_cleaning algorithm.
- keyword_extraction.py wraps the RAKE keyword extraction algorithm.
- Model_Training file contains the src for the process of training LSTM and GRU (pop and rock).
- Lyric_Generator file contains the src for automatically generating the lyric for rock and pop style using trained GRU model.

data

- data file contains the dataset we used for training the model including the primary dataset and the cleaned dataset.

report

- project report introduces the project from the view of background, related work, technical implementation, result demo, user study, future work, reference.