Acoustics of Human Voice Production

Human voice production is studied as an acoustic process inside the vocal tract using the wave equation for air, and compared with large-scale databases of simultaneously acquired microphone and electroglottograph signals. The following theories are confirmed: For voiced sounds, the process starts at glottal closures. Immediately before a glottal closure, there is a continuous airflow in the vocal tract. Each glottal closure abruptly blocks the air flowing into the vocal tract, excites a zero-velocity d’Alembert wavefront, which then resonates in the vocal tract to form an acoustic wave. Its waveform is determined by the geometrical configuration of vocal tract, thus represents instantaneous timbre. Kinetic energy of airflow is converted into acoustic energy. For plosives and affricates, the process starts at various places inside the vocal tract. The mechanism is similar to that of voiced sounds. Linear superposition of those acoustic waves plus noise-based disturbances constitutes speech signal. Using Kramers-Kronig relations, Fourier analysis and Laguerre functions, digitized speech signals can be converted into a parametric representation in terms of timbre vectors; and the speech can be regenerated from those timber vectors, possibly with a variety of desirable modifications. Examples of voice transformation are shown.