Ahots kantatuaren sintesiaren egokitzapena bertsolaritzarako

This paper describes the process of creation of a singing voice synthesis system for bertsolaritza, the art of singing improvised songs in Basque. The database acquisition process and the tools used to achieve this goal are explained. Our system uses Hidden Markov Model (HMM) based statistical speech synthesis as base to build a singing voice. We use a singing voice database instead of a speech database to train the HMM system with music labels in addition to the usual linguistic labels.