Data-based Synthesis

A speech synthesis technique that generates speech by selecting and concatenating segments from a large database of prerecorded human speech, rather than using rules to generate acoustic waveforms from scratch. The database is indexed with phoneme boundaries, pitch, and prosodic information, and the synthesis engine searches for the best-matching recorded segments to assemble each new utterance. Data-based synthesis produces more natural and intelligible speech than older formant synthesis methods, and because it uses actual recordings of a specific speaker, it can preserve that individual's unique vocal characteristics — making it the foundation for voice banking and personalized synthetic voice systems.