원문정보
Recent Trend on Speech Recognition with Deep Neural Network
초록
영어
In this study, we discuss the structure and the principle of the speech recognition system which is currently developing. There are two types of speech recognition models: a generation model and an identification model. In the past, a method based on a generation model was used. Specifically, a speech recognition method based on a GMM-HMM model was developed and evolved into a DNN-HMM method in which an deep learrning technique was introduced. Subsequently, the deep neural network technology was applied to the identification model. We explain te CTC method based on the end-to-end model and the RNN method including the attention mechanism as an end-to-end model.
목차
I. 서론
II. 음성인식 시스템의 구성
2.1 음성인식 시스템 구조
2.2 식별모델과 생성모델
III. 심층학습을 이용한 음성인식
3.1 GMM-HMM 하이브리드 방식
3.2 End-to-End 방식
3.3 End-to-End CTC 모델
3.4 RNN에 주의기구 도입방식
IV. 결론
참고문헌