CHATR: 音声特徴に 基づく単位選択

our synthesis philosophy:

• other synthesisers sound `bad' because they have a small number of units, and try to change the prosody by `force'
• we try to select units that are close to the prosodic targets, and so preserve the `naturalness' of the speech.
• our algorithm finds speech segments from the database that fit both the `target' and each other.

---

(C) Copyright ATR ITL 1997

---