Evaluation of speech synthesis

Tema ovog rada bila je izraditi sučelje za dva algoritma sinteze govora (TD-PSOLA i konkatenacija) u obliku web stranice. Web stranica nudi ocjenjivanje sintetiziranog govora u kategorijama kvalitete, nepravilnosti, razumljivosti i prirodnosti ocjenom od jedan do pet. Nadalje korisniku je omogućeno označavanje loše sintetiziranih riječi, također mogu glasati za algoritam za koji smatraju da je pružio bolju sintezu. Prosječne ocjene i raspodjela ocjena za svaku kategoriju se spremaju u bazu podataka, a prikazuju se u obliku zvjezdica, dok se loše riječi prikazuju u obliku liste. Svaka sinteza sprema se u bazu podataka zasebno i ima svoj unikatni URL tako da se može dijeliti, odnosno više korisnika može ocijeniti istu sintezu. Ocjenjivanje i lista loših riječi pomažu nam u unaprjeđivanju algoritma.The topic of this paper was to create an interface for two speech synthesis algorithms (TDPSOLA and concatenation) in the form of a web page. The website offers an evaluation of synthesized speech in the categories of quality, irregularity, intelligibility and naturalness with a score of one to five. Furthermore, the user is allowed to mark poorly synthesized words, additionally, they can also vote for the algorithm they consider to have provided better synthesis. Average grades and the distribution of grades for each category are stored in a database and displayed as stars, while bad words are displayed as a list. Each synthesis is stored in a database separately and has its own unique URL so that it can be shared, with the goal of allowing multiple users to rate the same synthesis. Grading and a list of bad words help us improve the algorithm