| Title |
Lietuvių kalbos atpažinimas, panaudojant Julius programinę įrangą / |
| Translation of Title |
Speech Recognition of Lithuanian Using Julius Software. |
| Authors |
Braubartas, Ernestas |
| Full Text |
|
| Pages |
48 |
| Keywords [eng] |
speech recognition ; Hidden Markov Model ; acoustoc model ; categorization of words ; Julius software |
| Abstract [eng] |
The theme of Master project of signal technology is actual, because not enough usual information introduction ways. Therefore information search and processing, complicated devices and programs control would be more handily if computers and devices understood human speech. Similar systems are designing for many years in the world. However Lithuanian speech recognition systems are still developing in nowadays. The thesis treats of isolated Lithuanian words recognition dividing them into category and using Hidden Markov Models. The idea of research is to explore categorization of Lithuanian words influence on the accuracy of recognition. The recognition of single words and word groups is under research too. Acoustic model is constructed by using HTK toolkit which is based on Hidden Markov Models. Categorization of words is described with Backus-Naur form. Experiments are made with Julius software speech recognition system Julian witch performs words category based recognition. Best results are got trying to recognize single words set into categories. The accuracy rate of recognition reaches 91 %. While trying to recognize uncategorized word sequences – the accuracy rate of recognition reaches only 51 %. The accuracy rate of Microsoft Office Word 2003 control menu recognition reaches 82 %. |
| Type |
Master thesis |
| Language |
Lithuanian |
| Publication date |
2008 |
