VOICE2TUBA: transforming singing voice into a musical instrument

摘要

In this paper, a scheme to synthesize and convert singing voice into tuba sound is presented. First, our method estimates the fundamental frequency (F (0)) and the aperiodicity of a monophonic audio signal, in order to obtain the pitch and volume variations of human voice. Then, the parameters extracted are used to generate a musical excerpt emulating a certain musical instrument (tuba) in such a way that the melody resembles the original sung song. To this end, two different generation approaches are devised. One of them is based on additive signal synthesis from harmonic amplitudes. The other one converts the F (0) curve into a MIDI stream, in order to allow the play back with a virtual tuba.