A text-to-speech system for Spanish with a frequency domain based prosodic modification algorithm

From the input text, the linguistic-prosodic module obtains the phonetic transcription and prosodic marks that reflect both the syntactic structure and some rhythmical constraints. The synthesis module is a variation of the MBE (multiband excitation) vocoder with an LPC (linear predictive coding) filter that is very flexible for prosodic modifications. From a parametrized acoustic database, the algorithm decodes the speech units and modifies their prosody in a single process. The frequency baseness of the synthesis algorithm allows a fine pitch modification without spectral envelope distortion. The prosody modeling is done using the acoustic module by a close copy stylization method.<>