Croatian HMM-based Speech Synthesis

Abstract

The paper describes the development of a trainable speech synthesis system, based on hidden Markov models. An approach to speech signal generation using a source-filter model is presented. Inputs into the synthesis system are speech utterances and their phone level transcriptions. A method using context-dependent acoustic models and Croatian phonetic rules for speech synthesis is proposed. Croatian HMM-based speech synthesis experiments are presented and generated speech results are discussed.