Normalizacija teksta u strojnoj tvorbi hrvatskoga govora

master's thesis

• Title (english): Text normalization in Croatian speech synthesis
• Author: Domagoj Mijić
• Mentor: Šandor Dembitz (mentor)
• Committee member: Šandor Dembitz (predsjednik povjerenstva)
• Committee member: Gordan Gledec (član povjerenstva)
• Committee member: Marko Banek (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2015-07-09, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Electrical Engineering

Abstract

Normalizacija teksta u strojnoj tvorbi govora vrlo je kompleksan problem, jer pri pojavi strane riječi dolazi do prelaska iz jednog jezika u drugi, pa prema tome i narušavanje jezičnog identiteta. Pri prijelazu iz jednog jezika u drugi mijenjaju se pravila pisanja i čitanja, cijela fonetika i dolazi do pojave nekih novih grafema. Za rješavanje tog problema razvijen je program u programskom jeziku Java, korištenjem razvojnog okruženja Eclipse. Preko svojih metoda on učitava bazu imena koje zna u strukturu podataka koju može brzo pročitati i s korištenjem algoritma za Jaro-Winkler udaljenost izračunava udaljenost između testnih podataka I podataka u bazi.
Nakon što je program našao najbolji rezultat u svojoj bazi, ako to rješenje ima istu točnost u više jezika, program provjerava kontekst te vraća procjenu najtočnijeg rješenja prema pravilima čitanja za jezike.
Akronimi i kratice imaju svoja pravila i pri njihovom pronalasku, program zamijeni ih zadanim nizom znakova koje će sintetizator govora pročitati.

Abstract (english)

Normalization of text for mechanical speech synthesis is a complex problem, because on the occurrence of foreign words, there is a transition from one language to another, and therefore linguistic identity is violated. In the transition from one language to another, the rules of writing and reading are changed, the whole phonetics is changed as well and some new graphemes appear. To solve this problem, the program is created in Java using the Eclipse development environment. Through its methods, it loads database of names that it knows into data structure that can be quickly read and uses algorithm for calculating Jaro-Winkler distance to calculate distance between test string and strings from database..
Once a program found best result in its database, if the solution has the same accuracy in multiple languages, the program checks the context and returns the most accurate assessment of the solution according to the rules of reading for languages.
Acronyms and abbreviations have their own rules and on their discovery, program replaces them with the default string that will read the speech synthesizer.

• Keywords: normalizacija; akronimi; imena; kratice
• Keywords (english): normalization; acronyms; names; abbreviations
• Language: croatian
• URN:NBN: urn:nbn:hr:168:888162
• Study programme: Title: Information and Communication Technology; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka informacijske i komunikacijske tehnologije (magistar/magistra inženjer/inženjerka informacijske i komunikacijske tehnologije)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2018-02-06 10:41:58