Automated Generation of Chinese Lyrics Based on Melody Emotions

Abstract

Natural Language Processing enables computers to understand human natural languages and assists humans to perform many tasks. Recently, language research was gradually moving towards the field of natural language generation, and lyrics generation is one of these research directions. In the past, many studies have explored how to generate lyrics using seeds or prompt words. But, they usually ignored the relationship between music and lyrics and even melody emotions when generating lyrics. In this paper, we propose the Chinese lyrics generation system based on notes and melody emotions. In the first phase, a support vector regression model is trained to detect melody emotions. In the second phase, a sequence-to-sequence model is trained to generate lyrics using notes, melody emotions, and lyrics. Finally, we conduct the experiments and evaluate the detected melody emotions and the generated lyrics. Results show that the generated lyrics have fluent sentences, and the sentences are with connected semantics.