Conditional LSTM-GAN for Melody Generation from Lyrics

--Melody generation from lyrics has been a challenging research issue in the field of artificial intelligence and music, which enables to learn and discover latent relationship between interesting lyrics and accompanying melody. Unfortunately, the limited availability of paired lyrics-melody dataset with alignment information has hindered the research progress. T o address this problem, we create a large dataset consisting of 12,197 MIDI songs each with paired lyrics and melody alignment through leveraging different music sources where alignment relationship between syllables and music attributes is extracted. Most importantly, we propose a novel deep generative model, conditional Long Short-T erm Memory - Generative Adversarial Network (LSTM-GAN) for melody generation from lyrics, which contains a deep LSTM generator and a deep LSTM discriminator both conditioned on lyrics. In particular, lyrics-conditioned melody and alignment relationship between syllables of given lyrics and notes of predicted melody are generated simultaneously. Experimental results have proved the effectiveness of our proposed lyrics-to-melody generative model, where plausible and tuneful sequences can be inferred from lyrics. I NTRODUCTION Music generation is also referred to as music composition with the process of creating or writing an original piece of music, which is one of human creative activities [1]. Without understanding music rules and concepts well, creating pleasing sounds is impossible. To learn these kinds of rules and concepts such as mathematical relationships between notes, timing, and melody, the earliest study of various music computational techniques related to Artificial Intelligence (AI) has emerged for music composition in the middle of 1950s [2]. Markov models as a representative method of machine learning have been applied to algorithmic composition [3]. However, due to the limited availability of paired lyrics-melody dataset with alignment information, research progress of lyrics-conditioned music generation has been obstructed.