• Record hours of human speech
• Cut into small units
• Stitch together to form new sentences

Types

• Diphone concatenation
  • record transitions between phonemes
• Unit Selection
  • record full sentences
  • segment sentences
  • index phonemes, diphones, syllables, words

Strengths

• Natural Sounding
• Dominated commercial tts for years

Weaknesses

• Many recordings
• glitchy transitions
• limited expressivity
• can’t modify emotion/style
• coverage gaps

Methods

• Can use heuristic-based-search-algorithm
  • target cost
  • join cost

Example

• Early Siri