LatAksLate: Javanese Script Translator based on Indonesian Speech Recognition using Sphinx-4 and Google API

Abstract

Javanese script is one of the cultural heritage of Indonesia, especially the Javanese. Currently, the understanding of the younger generation of Javanese culture, especially the Javanese script has decreased. The main cause is, Javanese script only taught in school but never used in everyday life. This research proposes the application of speech recognition combined with the process of translating Indonesian into Javanese script to facilitate the process of learning Javanese script. The application is built based on Android using the Sphinx-4 framework for speech recognition process in Indonesian language and Google API as a database for the translation process. In order to obtain good accuracy, several steps are introduced phonetic recognition of speech input sequences involving model dictionaries, phonetic sequence matching transcripts of language models acting as datasets, testing the sequence of acoustic senones of speech properties to match phonemic properties, decoding acoustic models evaluated by word error rate (WER), syllables category categorization into syllables Javanese text, most recently by classifying syllables of Javanese text to be converted into Javanese script with true type font (TTF). Based on the test results obtained recognition accuracy and translation of 82.46%.