太赫兹科学与电子信息学报Issue(6)：958-963,6.DOI:10.11805/TKYDA201306.0958
基于点过程模型连续语音关键词检测
Spotting keywords in continuous speech based on Point Process Models
王勇 1张连海1
作者信息
- 1. 信息工程大学信息 信息系统工程学院，河南 郑州 450002
- 折叠
摘要
Abstract
A keyword spotting method is proposed based on Point Process Model(PPM) in continuous speech. Frame-level phone posterior probability is computed by using TempoRAl Patterns(TRAP) and Multiple Layer Perception(MLP). The speech can be considered as independent events(phones), and PPM can be set up by using Poisson process. The likelihood ratio is calculated to estimate whether the keyword is uttered. The experimental results show that the average recall and precision rate of keywords are above 69.5%and 82.0%with 8 kHz sampling frequency for speech, respectively.关键词
检测/音素后验概率/泊松过程/点过程Key words
spotting/phone posterior probability/Poisson process/Point Process分类
信息技术与安全科学引用本文复制引用
王勇,张连海..基于点过程模型连续语音关键词检测[J].太赫兹科学与电子信息学报,2013,(6):958-963,6.
