Unit-selection synthesis achieves naturalness at the cost of huge databases and inflexibility

Evidence

One benefit of statistical parametric synthesis is that the source–filter-type synthesis of speech signals provides possibilities for changing the characteristics of voice, such as speaking styles and prosodic features, and possibly even provides multilingual support (Zen et al., 2009). With the unit-selection method, such modifications are a bit complicated, as voice conversion techniques have to be utilized (Stylianou et al., 1998), which may degrade the quality of the speech. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2263 (#expands, via linked claim)

Categorization

• XREF: connects to: Statistical parametric synthesis quality surpasses knowledge-based but trails unit-selection synthesis.; Source-filter parametric synthesis enables voice modification that unit-selection cannot. — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2263; Computer power growth enabled natural-sounding voices trained on large speech databases (+1 more)

Broader Connections

• Source-filter parametric synthesis enables voice modification that unit-selection cannot. #expands (0.85) — Ville Pulkki, Matti Karjalainen, Communication Acoustics- An…, 2263
• Shorter synthesis units lower concatenation cost but syllables are the optimal balance #expands (0.85)
• Statistical parametric synthesis quality surpasses knowledge-based but trails unit-selection synthesis. #expands (0.80)
• Computer power growth enabled natural-sounding voices trained on large speech databases #corroborates (0.75)