Lyrics generation and evaluation modules

In the computational creativity field there have been many researches related to the automatic generation of music and forms of literature. In this work we propose a solution to the automatic generation of songs, first writing the music and the lyrics independently, and later joining them. Regarding music generation, a module developed by Guillermo Villeta Torres that would be explained in a later TFG, has been used. This module generates a song in MIDI format that would be then joined to the lyrics, and is also capable of modifying the music based on the feedback it receives after this merging. As for the generation of lyrics, we explored the use of Markov Chains which helped us obtain a poetic text. During this process, we developed some tests for each generated sentence to carry a syntactic and a sentimental evaluation. For both evaluations, different configurations were tried until the result obtained was the most satisfactory one. When both music and lyrics are ready, they are received by another module that put them together and evaluated the outcome. During this evaluation process, we came up with a series of tests first to check the sentiment analysis of the music, and then to obtain a quality index of the resulting song. These tests are performed iteratively through the music with the lyrics already assigned, and later the quality score is normalized. Depending on the outcome of this evaluation process, either the result is given as an output or a series of parameters are sent to the generator modules as feedback for them to modify the lyrics and music with the intention of improving the result.