Indexing join costs for faster unit selection synthesis

A corpus-oriented concatenative speech synthesis has become a leading method for generating high quality output in the last two decades. This method creates speech by re-sequencing pre-recorded speech units selected from a very large speech databases. In a CHATR style selection the final synthesized sequence of units is obtained by searching all of the candidate sequences for the minimal combination of target and join costs. The searching process is a very time consuming task, so several acceleration methods has been created to make the synthesis available for real time applications. This paper proposes one such a method that was implemented in the S2 system - the Slovak low-level synthesizer.