A comparison between voice as a periodic sound source in speech and voice as a musical instrument is made on the basis of physiological and acoustic differences. Oral acoustic output power is about 1000 times greater in maximal effort singing than in conversational speech, and vocal efficiency can be 100 times greater. This makes optimization of laryngeal and pharyngeal configurations an important performance task. Recent research demonstrates that a somewhat lowered larynx, stable respiratory and phonatory configurations based on nearly isometric contractions of opposing muscles, and formation of an upper laryngeal resonator for high-frequency emphasis (the singing formant) are key factors. Fundamental frequency control over a range of several octaves is accomplished by intrinsic laryngeal muscular contraction, but nonlinear tissue characteristics of the vocal folds can account for an amplitude tuning by pulmonary effort. The quality of the tone is dominated by two major register adjustments in the larynx, but a larger number of minor register effects are perceived on the basis of vocal tract loading. This source-system interaction, which depends on vocal tract dimensions, in conjunction with the size of the vocal folds, establishes a scaling principle for voice classification on both physiological and acoustic levels. Training of the instrument and training of the artist is an inseparable process. Evidence is yet scarce that a good singing voice, like a good violin, is structurally determined. If so, what are the key morphological elements? If not, is the quality improved by overall muscular strengthening, by differential adjustment and accurate coordination of specific muscles, or by biological adaptation of mechanical tissue properties?