Semantic Computing in Scalable Text-To-Speech System

摘要

Because of diversity of hardware environments, building scalable text-to-speech system is an important issue of Corpus-based text-to-speech system. This paper proposes and analyses three semantic computing problems of building scalable text to speech system: similarity calculation, granular computing and automated instances-pruning process framework. According to these, an acoustic clustering algorithm-NuClustering-VPA and a data ranking algorithm-StaRp-VPA are constructed to pruning synthesis instances. In experiments, the naturalness scored by MOS remains almost unchanged when less than 50% instances are pruned off using these two algorithms and the MOS does not severely degrade when reduction rate is above 50% using StaRp-VPA algorithm.