Formant synthesis of speech based on Fant's acoustic model has proved to give a good approximation to natural speech spectra. Unfortunately analogue and digital filters do not behave similarly, especially at frequencies near the half sampling frequency. This paper deals with the deviation between Fant's model and a digital realization and how to minimize this problem. The emphasis is on female vowels.
Keywords: Speech synthesis, Female voice, Digital filters, Spectral tilt.