Concatenation-Based MIDI-to-Singing Voice Synthesis
Michael W. Macon, Leslie Jensen-Link, E. Bryan George, James C. Oliverio, Mark A. Clements · 1997
In this paper, we propose a system for synthesizing the human singing voice and the musical subtleties that accompany it. The system, Lyricos, employs a concatenation-based text-to-speech method to synthesize arbitrary lyrics in a given language. Using information contained in a regular MIDI file, the system chooses units, represented as sinusoidal waveform model parameters, from an inventory of data collected from a professional singer, and concatenates these to form arbitrary lyrical phrases. Standard MIDI messages control parameters for the addition of vibrato, spectral tilt, and dynamic musical expression, resulting in a very natural-sounding singing voice.