TEXT-TO-SPEECH SYNTHESIS: A PROTOTYPE SYSTEM FOR CROATIAN LANGUAGE
AbstractThis paper presents the development of a Croatian text-to-speech system capable of synthesizing speech from arbitrary text. Input text in normalized form is first transcribed into a phonetic string (grapheme-to-phoneme conversion) and then processed by a TD-PSOLA based synthesizer. A procedure for automatic selection of diphones from a spoken corpus is proposed. A Croatian language diphone database was built for the system. Subjective quality evaluations of the resulting speech were performed, as well as tests for intelligibility.
Engineering review uses the Creative Commons Attribution-NonCommercial-NoDerivatives (CC-BY-NC-ND) 4.0 International License, which governs the use, publishing and distribution of articles by authors, publishers and the wider general public.
The authors are allowed to post a digital file of the published article, or the link to the published article (Enginering Review web page) may be made publicly available on websites or repositories, such as the Author’s personal website, preprint servers, university networks or primary employer’s institutional websites, third party institutional or subject-based repositories, and conference websites that feature presentations by the Author(s) based on the published article, under the condition that the article is posted in its unaltered Engineering Review form, exclusively for non-commercial purposes.
The journal Engineering Review’s publishing procedure is performed in accordance with the publishing ethics statements, defined within the Publishing Ethics Resource Kit. The Ethics statement is available in the document Ethics Policies.