지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
논문 기본 정보
- 자료유형
- 학술대회자료
- 저자정보
- 발행연도
- 2014.6
- 수록면
- 2,015 - 2,018 (4page)
이용수
초록· 키워드
오류제보하기
We propose a new discriminative model for robust keyword spotting under an always-on listening framework. In comparison to the hidden Markov model (HMM), which has been generally used for keyword spotting, the proposed model is based on the semi-Markov model (SMM) framework due to its extended capability to incorporate long-range statistical dependencies allowing simultaneous segmentation and labeling of sequential data. For keyword spotting from a continuously received input sound stream, the SMM framework which decides whether the keyword is spoken for each new input frame is modified to allow a transition from a non-keyword state to one of initial keyword states during the Viterbi-like SMM inference. In order to estimate the parameters of the proposed SMM, we develop a discriminative training algorithm based on structured SVM and cutting-plane algorithm. Experimental results show that the proposed discriminative SMM framework outperforms the HMM and performs fast and robust keyword spotting enough to be used under an always-on listening framework.
목차
Abstract
I. Introduction
II. Semi-Markov Model For Keyword Spotting
Ⅲ. Discriminative Training
Ⅳ. Experiments
참고문헌
참고문헌 (0)
참고문헌 신청이 논문의 저자 정보
최근 본 자료
댓글(0)
0
UCI(KEPA) : I410-ECN-0101-2015-560-001688462