Preserving naturalness in synthetic voices while minimizing variation in formant frequencies and bandwidths.