Modern synthetic voices learn from single-speaker speech databases rather than hand-crafted phonetic rules.

Evidence

Data-based synthesis methods have largely been adopted in academia and in industry since the 1990s. The increase in power and resources of computer technology has enabled the building of natural-sounding synthetic voices based on the utilization of large, single-speaker databases of natural speech (Zen et al., 2009). In contrast to knowledge-based synthesis, each phonetic unit need not be crafted for each applicable context anymore, but the information in the database is used to train the system on how to produce correct-sounding speech. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2230

Categorization

• DEFINE: Defines data-based synthesis as distinct from knowledge-based synthesis, establishing the core terminological split of the chapter.