High-quality synthesized speech requires tens of hours of manually-segmented recorded speech.

Evidence

A database consisting typically of tens of hours of recorded speech has first to be created for unit-selection synthesis. During the creation of the database, each recorded utterance is segmented into some or all of the following: individual phones, diphones, half-phones, syllables, morphemes, words, phrases, and sentences. The segmentation can be done using automatic systems, although often the segments have to be manually fine-tuned. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2244