Lyrics to Music Generator: Statistical Approach

Abstract

Natural Language Processing is in growing demand with recent developments. This Generator model is one such example of a music generation system conditioned on lyrics. The model proposed has been tested on songs having lyrics written only in English, but the idea can be generalized to various languages. This paper’s objective is to mainly explain how one can create a music generator using statistical machine learning methods. This paper also explains how effectively outputs can be formulated, which are the music signals as they are million sized over a short period frame. The parameters mentioned in the paper only serve an explanatory purpose. This paper discusses the effective statistical formulation of output thereby decreasing the vast amount of estimation of output parameters, and how to reconstruct the audio signals from predicted parameters by using ‘phase-shift algorithm’.