Speech Audio Retrieval Using Voice Query

Abstract

Multimedia data has increasingly become a prevalent resource in Digital Library system; this includes audio, video, and image archives. However, each type of these data may need specific tools to help facilitate effective and efficient retrieval tasks. In this paper, we focus on retrieval of speech audio collection, which includes audio books, speech recordings, interviews, and lectures. Currently, most of the audio retrieval systems are based on keyword/title/author search typed into the system by users. The system then searches for particular keywords and gives a list of entire audio files that potentially are relevant to the query. Nonetheless, browsing audio content for particular section of the audios without knowing the actual content is yet a very difficult task. Moreover, since audio transcription or keyword annotation is very labor intensive and becomes infeasible for large data, we introduce here a preliminary framework that locates subsections of the audio that correspond to the voice query made by a user. We demonstrate a utility of our approach on query retrieval tasks in various types of audio recordings. We also show that this simple framework can potentially help retrieve and locate the voice query within the audio accurately and efficiently.