Modelling voiceless speech segments by means of an additive procedure based on the computation of formant sinusoids.