Improvements of Hungarian Hidden Markov Model-based text-to-speech synthesis

Abstract

Statistical parametric, especially Hidden Markov Model-based, text-to-speech (TTS) synthesis has received much attention recently. The quality of HMM-based speech synthesis approaches that of the state-of-the-art unit selection systems and possesses numerous favorable features, e.g. small runtime footprint, speaker interpolation, speaker adaptation. This paper presents the improvements of a Hungarian HMM-based speech synthesis system, including speaker dependent and adaptive training, speech synthesis with pulse-noise and mixed excitation. Listening tests and their evaluation are also described.