DETECTION OF SPEECH SECTIONS FROM ACOUSTIC SIGNALS (Content Processing and Information Security)(International Workshop On Advanced Image Technology (IWAIT2004))

概要

論文の詳細を見る
We study a method to detect speech sections from acoustic signals using lip image sequences. This method can reduce mistakes that the non-target person's speeches are recognized as those of the target person, since the lip images of the target person can be used. The proposed method employs the hidden Markov models to learn the characteristics of the speech and the non-speech sections in the lip image sequences. Especially, to reduce the ambiguities caused by the variations of the appearance of the lips in the images and of the brightness of the images, we examine the construction of the features, which are extracted from the images and are used to detect speech sections.
社団法人電子情報通信学会の論文
2004-01-05