Slovak text-to-speech synthesis in ARTIC system

This paper presents a brand-new Slovak text-to-speech system. It was developed within the framework of ARTIC system (primarily designed to synthesize Czech speech) with respect to the knowledge of Slovak language. Thus, statistical approach (using hidden Markov models) was employed to build an acoustic unit inventory of Slovak language in a fully automatic way. Both phonetic transcription and prosodic rules were proposed to convert an input text to its phonetic form and to estimate its suprasegmental features. As a result, a fully working text-to-speech system that converts an arbitrary Slovak text to the corresponding output speech was designed. The informal listening tests show the system is capable of producing speech of a high quality (with high level of intelligibility and good naturalness).