next up previous
Next: Results Up: Size and scaling Previous: Lexicons and Letter-to-sound rules

Unit databases

The second largest data structure in a synthesizer is the unit database representing the speech units that are to be concatenated.

In Festival although a number of synthesis techniques are supported, at present only one basic type, with a number of options, has been ported to Flite. The residual excited pitch synchronous LPC method [4] is used as a method for modifying pitch and duration independently. LPC coefficients plus encoded residual also has the advantage of being smaller than the full pulse coded modulated signal (PCM).

The basic representation of the units is a short term pitch synchronous signals consisting of a set of coefficients and possible a residual. In the the default case, these are LPC coefficients and encoded residuals.



Alan W Black 2001-08-26