Lyrics Generator

Overview:

This is a Deep learning model, which uses LSTM and is implemented through TensorFlow. Its main function is to generate Lyrics(text) upto "n" number of words for a given input line.

Files:

1. train_model.ipynb:

   This Jupyter notebook creates and trains the LSTM model. It has 2-Layer LSTM model followed by 2 Dense layers having total of 3.1M trainable params. The model achieves an accuracy of about 55%

2. lyrics_generator.ipynb:

   This jupyter notebook loads the model and predicts "n" (user input) of words for a given input line

3. lyrics.txt:

   This is a dataset of about 1000 English songs from various artists such as Adele, Britney Spears, Justin Beiber, Elvis presley, Rihanna and others. This file can be replaced to train the model on other datasets such as English Essays, Stories, chat conversations etc.

Model:

The first layer of the model is the Embedding layer that vectorizes the input words and holds them into clusters of similar-meaning words. Next,The model uses 2 layers of LSTM. The first layer is Bidirectional Model having 256 nodes with a dropout of 0.2 and the other LSTM layer has 128 nodes. Next we have 2 dense layers having nodes equal to (total_number_of_words/2) and (total_number_of_words) respectively. Below is the summary of the model:

  Layer (type)                 Output Shape              Param #   
  =================================================================
   embedding (Embedding)        (None, 23, 100)           763400    
  _________________________________________________________________
  bidirectional (Bidirectional (None, 23, 512)           731136    
  _________________________________________________________________
  dropout (Dropout)            (None, 23, 512)           0         
  _________________________________________________________________
  lstm_1 (LSTM)                (None, 128)               328192    
  _________________________________________________________________
  dense (Dense)                (None, 3817)              492393    
  _________________________________________________________________
  dense_1 (Dense)              (None, 7634)              29146612  
  =================================================================

Result:

The lyrics_generator.ipynb file requires a input line from the user, continuing on which, the model will predict its own set of words.

Example 1

"I miss you"

I miss you when the lights go out 
from the rest who is standing there 
i will sustain 
closing down in your eyes 
i see through the dark of your body 
strikes a extra road outta lights
know something the whole block on her name 
i can't go back to the same old lane
How could you come be true 
i was wrong
oh maybe i was burned but i called it a lesson learned
i see your head don't do about you 
darlin' i'll find someone to find you up my head

Example 2

"I will see you"

I will see you out of control 
don't you leave me thinking of me feeling this way i'm leaving myself 
i wish i could lay down beside you 
you see the way you look at me now 
i'll be standing right next to you yeah yeah yeah yeah 
closed there ain't nothing here for me 
i'm full of mud on the wall sitting there 
it is just you and i can't help myself from shining like a rocket that takes off the lights 
lay me down pour it up pour it up pour it up all night on my soul