Creating singing vocal expressions by means of interactive evolutionary computation

Abstract

Today, research on singing by computers has attracted attention of researchers. VOCALOID is an application to realise that aim. By inputting lyrics and melody, users can make songs sung by the computer. In order to make the singing voice sound more human, users must control frequency curve very carefully. Compared with inputting lyrics or melody, this controlling presents heavy overhead for users. In this research, we propose a system for easily optimising frequency curves with Interactive Evolutionary Computation (IEC). We compared various frequency models by evaluating the convergence performance of GA to fit one of the frequency curves of real human singing, and found a suitable one for achieving our goal. And we made a questionnaire to compare previous interfaces and our IEC interface. From the results of this survey, we found our IEC interface can optimise frequency curve more easily than previous interfaces.