The paper presents a fully functional prosody generation module developed for Macedonian text-to-speech (TTS) synthesis. The module is based on research of prosody generation modules in high-end TTS synthesis systems, previous prosody experiences in Macedonian TTS, as well as original research of prosody carried through by the authors. The paper starts with an overview of the basic tasks, problems and solutions in prosody generation modules. Then it continues to give a detailed account of the workings of the developed module. The module first segments the input speech into intonation phrases, and determines their intonation type. Next it generates durations for each of the units that will be used to synthesize the speech output. Then it determines the positions of the lexical stresses and modifies these units’ durations. After determining the intonation phrase’s pitch accent location, it generates an adequate pitch contour and calculates the pitch targets needed for unit modification. The synthesis module uses this data to generate prosody in the output speech. Generated prosody patterns in the output speech are of satisfactory quality for arbitrary text input. The presented results are of significant value for Macedonian TTS and can be used for other underrepresented languages.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=15783
(724KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
