5.8 Summary and Chosen Methods

Table 4.1 shows a comparison of the representations introduced in the preceding sections with a score (++, +, $\textsf{\large o}$, -, -) indicating the degree to which they fulfill the requirements for representations from section 4.1.

The methods chosen to be implemented are filter coefficients, the spectral representation, precise and fuzzy formants. The filter coefficients are kept because they provide the most precise and a very compact representation, they are closest to the source and can be easily converted to the canonical representation. The spectral representation is the most general, and will be the canonical representation. The precise formants are necessary for high quality singing voice synthesis. The fuzzy formants, along with the spectral representation, will make it possible to interpolate by formant shift between spectral envelope in spectral representation and precise formant representation, as will be explained in section 5.1.2.

 

Table 4.1: Comparison and score of representation methods in regard of requirements

Representation	Stability	Locality	Flexibility/Ease	Speed of Synthesis	Space	Manual
			of Manipulation	time-/freq-domain		Input
Filter Coefficients	++	-	-/ -	++/ -	+	-
Spectral	++	++	++/ $\textsf{\large o}$	+/ ++	$\textsf{\large o}$	+
Geometrical	-	+	+/ ++	$\textsf{\large o}$/ -	+	++
Formants	-	+	++/ ++	+/ +	++	++
HRMP	-	+	++/ $\textsf{\large o}$	$\textsf{\large o}$/ $\textsf{\large o}$	$\textsf{\large o}$	-