Improved formant frequency estimation from high-pitched vowels by downgrading the contribution of the glottal source with weighted linear prediction.