VoiceGraph is a user interface for retrieval of audio recordings based on automatic processing of speech contained in the recordings. Our ultimate goal is to create a responsive interface that integrates speaker identification, speech recognition, language identification and machine translation to support effective interactive speech-based search strategies. VoiceGraph is being developed by the College of Information Studies, the Human Computer Interaction Laboratory, and the National Public Broadcasting Archives at the University of Maryland. The research is supported by a Shared University Research equipment grant from IBM.

Papers and Presentations

User Interface Design for Speech-Based Retrieval (HTML)

A paper that appeared in the June/July 2000 issue of the Bulletin of the American Society for Information Science.

User Interaction with Speech-Based Retrieval Systems (Powerpoint)

The slides from a presentation given at the Annual Conference of the American Society for Information Science in Washington, DC in November, 1999.

A Graphical Interface for Speech-Based Retrieval (Postscript)

A short paper that was presented at the Third ACM Digital Library Conference, Philadelphia, in June, 1998 that described our work at that time. A Rich Text Format version is also available.

Speech-Based Information Retrieval for Digital Libraries (HTML)

A paper describing the initial conceptual design of the the VoiceGraph interface that was presented as a poster at the AAAI Symposium on Cross-Language Text and Speech Retrieval in March, 1997. For black and white printing is it better to use the postscript version.

Specification (Postscript)

The original interface specification. This is now rather dated, but retained here for historical reasons.

Audio Collections

National Association of Educational Broadcasters

Reel to reel Recordings made between 1950 and 1970, from the holdings of the National Public Broadcasting Archives. About 10% of this 2,000 hour collection has been digitized for use with VoiceGraph.

Contemporary Broadcast News

Two standard test collections totaling approximately 600 hours that are distributed by the National Institute of Standards and Technology for the Spoken Document Retrieval track of the Text Retrieval Conference.

The VoiceGraph Interface

Static Screen Shot

Best viewed at 1280x1024 resolution, this screen shot illustrates the four key components of the interface, starting in the upper left and moving around to the right: query formulation, recording selection, transcript examination, and audio replay. We are presently enriching the recording selection interface to include metadata and extracted text as well as the timelines that are shown here.

The VoiceGraph Team

Faculty

• Doug Oard

Students

• Tara Hess
• Jinmook Kim

Staff

• Tom Connors
• Chuck Howell

Past team members

• Jody Condit
• Aaron Hallberg
• Chunyun Hao
• Julie Harding
• Inna Kachko
• Tapas Kanungo
• Zhongliang Liu
• Laura Slaughter
• Tony Tse
• Skip Warnick
• Galen Wilkerson
• Cynthia Zujko-Miller

Related Resources

Speech-Based Information Retrieval

Links to other speech-based IR resources on the Web.