Blockchain Scoops

VAE: A neural network that generates lyrics!!

0

A song features in a film. Whom does it take to make a song? A singer, a lyricist, musicians, composer, editor, and the list varies according to the needs of the song. While making a song the first and the basic requirement is the lyrics. The lyrics play all the game. Either it lifts the song or makes it a flop. What if there comes to a technology that gives you lyrics?? Well, there is one coming soon. Researchers have developed an Artificial Neural Network (ANN) that can actually generate lyrics for a song.

What does the ANN do?

The researchers found out that the neural generative model can help lyricists with the lyrics. The model can actually suggest lyrics. The suggestions will be based on the artist’s previous work. The training set of the network will be the previous songs. Studying them the network will be providing its suggestions.

The lyrics need not be copied as they are suggested. The suggestions might just inspire to write the actual lyrics. Finding words is a tough task. And the network is trying to make this task easier. The model can produce unusual arrangements of words that may or may not be used by the lyricist.

The results were much satisfactory as it produced lines that actually match the feel and emotion of the actual person. This was an achievement in itself because every lyricist has a unique style. The main aim was to teach the network this style.

The same system produced poems. The VAE produced the lines and arranged them in an artistically meaningful way.

 Actual working

The actual system that the team developed worked on neural networks. The researchers named this neural network as the variational autoencoder (VAE). The neural network was a multi-dimensional vector of real numbers and a CNN classifier. The training set to the network was the song clips of the artist. It rearranged the original lines of the text for suggestions.

Inspiration

The team wanted to find if a machine can actually learn the lyrical style of a person. There have been many experiments in the past but none that included creative stuff. The experiment was specifically challenging because it demanded the neural network to learn the way a person thinks and feels because that is the basics of good music. And this is how mostly one lyricist differs from the other.

The main challenge was to make a machine that could produce lyrics similar to the person. The neural network might be a system imitating the human brain but was it really of that capability was the real question.

The team plans to later make models that can learn external vocabulary too. This will help to generate lyrics of more variety in the style of the artist.

Conclusion

The aim is not to replace lyricists but to make them available a source of help. The system can later be used in larger scopes that we might have not imagined as yet. The step will definitely make a difference. To use it or not will depend on the artist but now we actually have a machine that can make songs!!

 

 

 

Drones in agriculture: redefining the way India farms

Previous article

A New Gmail: introduction to the new features

Next article

Comments

Leave a Reply

Login/Sign up