Personal versus normal spectral characteristics in the phonetogram

Peter Pabon

During the past 3 years a special computer phonetogram database was built up that today holds recordings of 40 normal voices and 40 singers. What is so special is that each contributing phonetogram by itself contains an all including amount of detailed information on an individual’s voice quality, as every recording stores the narrow band spectra for each possible dynamic position/cell within the phonetogram.
There are several fields of interest; the first aim was to average phonetograms and so to construct a normalized image on the general way that the voice spectrum changes with Fo and SPL. Moreover, as the phonetogram mapping allows a designation of specific phonetogram areas to specific registers, this also allowed for a separate analysis of the dynamic spectral variations typical to each register.
This paper will focus on the specific spectral properties that are normal to a voice type or register. It will go into the question if there is a representative spectral-voice quality pattern for male and female voices? Will the averaging lead to an average voice and what does it look like and sound like in its bare form? The opposite is also considered; what spectral qualities make up the individual differences? Analysis methods are presented that aim to define a specific modifier/template that represents the quality that is typical to one unique voice.
Apart from discussing the numerical results, direct voice synthesis from the stored spectral qualities in the phonetogram will be used to illuminate what acoustical qualities relate to the specific spectral patterns and their modifiers.

Peter Pabon , Royal Conservatory, The Hague / University Utrecht / Voice Quality Systems pabon@koncon.nl