JURNAL :PREDIKSI JEDA DALAM UCAPAN KALIMAT BAHASA INDONESIA DENGAN HIDDEN MARKOV MODEL

ABSTRAK - This study describes the design of the pauses predictor application of speech sentences in Bahasa Indonesia with Hidden Markov Model (HMM). This application serves to determine the pauses event that occur in Bahasa Indonesia sentences. There are two main processes in this application which is train to train the corpus, and prediction to predict pause. On the train, the input text is produced from sound files, and the output is training corpus for HMM engine. In the prediction process, inputs are words of Bahasa, and outputs are pause prediction that occurred earlier in the input sentence. The results of this study is the sentence that has been predicted in each pause events. Testing is done using precision and recall of training corpus and tagging pause prediction results. The results of precision and recall is calculated back to the f-score. Based on the testing that has been done, showed that the designed applications can already predict a pause in the Indonesian sentences with precision of 0.364, recall of 0.132, and F-score of 0.194

Keywords: pause prediction, Hidden Markov Model, precision & recall, f-score.

PENDAHULUAN - Sintesa ucapan adalah teknologi yang dapat membangkitkan ucapan buatan yang mirip ucapan manusia.

Namun pada penerapannya sering kali pengucapan kalimat yang dihasilkan sistem tidaklah sewajar pengucapan manusia, baik dalam intonasi, jeda antar kata, nada, penekanan, dan irama. Kemampuan prosodi sistem sering menghasilkan pengucapan yang monoton. Meskipun keterbatasan ini tidak berpengaruh krusial untuk beberapa aplikasi, hal ini lebih menimbulkan masalah dalam pengucapan teks-teks yang panjang seperti audio book atau robot.