This approach advances the singing voice closer towards a structured audio model in which musical parameters such as pitch, duration, and phonemes are represented orthogonally to the synthesis technique and can thus be modified prior to re-synthesis. 1.