Data-based speech synthesis splits into sample-reassembly and parameter-model approaches.

Evidence

Data-based methods can be divided into two subclasses: unit-selection synthesis and statistical parametric synthesis. Unit selection synthesis does not perform actual ‘synthesis’ of sound, but the output signal is composed of audio samples taken from the database (Beutnagel et al., 1999). Statistical parametric synthesis, on the other hand, synthesizes the speech signal using a model, the parameters of which are controlled using a system trained with real speech samples. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2231

Categorization

• DEFINE: Clarifies the two distinct technical approaches that both fall under data-based synthesis.