자연스러운 TTS 구현을 위한 ㄴ 첨가 처리 방안 ― 표준 발음과 현실 발음의 조화 모델 중심으로

초록

In this study, we first confirm that n-insertion pronunciation is an optional phonological phenomenon; thereafter, we assess the realization pattern of n-insertion pronunciations by examining the results of past pronunciation studies. In addition, we examine how n-insertion words are pronounced in text-to-speech (TTS) utterances provided by Google Translator and Naver Papago, revealing that the n-insertion pronunciations of TTS utterances are unrealistically realized and that the pronunciation is not properly reflected. Based on this, we propose an n-insertion method for natural TTS implementation. The main content is to construct a real pronunciation dictionary, reflect it in the TTS algorithm, and adjust the rate of applying real pronunciation according to user preference. In other word, to implement a natural TTS pronunciation, the pronunciation of the words placed in the n-insertion environment is basically calculated according to the standard pronunciation. However, since the degree of preference of the standard pronunciation or the degree of preference of the real pronunciation is different in relation to the n-insertion pronunciation according to the user, we suggest a method that allows the user to use the TTS by adjusting the n-insertion pronunciation preference.

키워드