Automatic labeling and digesting for lecture speech utilizing repeated speech by shift CDP
Automatic labeling and digesting for lecture speech utilizing repeated speech by shift CDP
复制标题
利用轮班 CDP 的重复语音对讲座语音进行自动标记和消化
DOI：
10.21437/eurospeech.2001-426
复制
发表时间：
2001
期刊：
影响因子：
--
通讯作者：
Kazuyo Tanaka
中科院分区：
文献类型：
--
作者：
Y. Itoh;Kazuyo Tanaka
本文提出了一种演讲语音的自动标注和文摘方法。该方法利用相同的部分，例如被认为是重要的并且在语音中重复的相同单词或相同短语。为了提取相同的部分，我们提出了一种新的高效算法，称为移位连续DP，因为它是连续DP的扩展，并实现了两个语音数据集的任意部分之间的快速匹配帧同步。本文对移位CDP算法进行了扩展，使其能够从单个长语音数据中提取相同的语音段。本文介绍了如何将该算法应用于标注和摘要的演讲稿。我们进行了一些初步的实验，以显示该方法可以提取相同的部分，提取的部分序列可以被看作是一个摘要的语音。
This paper proposes an automatic labeling and digesting method for lecture speech. The method utilizes same sections, such as same words or same phrases that are thought to be important and are repeated in the speech. To extract the same sections, we have proposed a new efficient algorithm, called Shift Continuous DP, because it is an extension of Continuous DP and realizes fast matching between arbitrary sections in two speech data sets frame-synchronously. Shift CDP is extended to extract same sections in single long speech data in this paper. This paper describes ways to apply the algorithm to labeling and digesting for a lecture speech. We conduct some preliminary experiments to show the method can extract same sections and a sequence of extracted sections can be regarded as a digest of the speech.