ESCA Workshop on Automatic Speaker Recognition, Identification, and Verification

Martigny, Switzerland
April 7-9, 1994

Public Databases for Speaker Recognition and Verification

John Godfrey (1), David Graff (1), Alvin Martin (2)

(1) Linguistic Data Consortium (LDC), University of Pennsylvania, Philadelphia, PA, USA
(2) National Institute of Standards and Technology (NIST), Gaithersburg, MD, USA

In this paper we review several major speech corpora which are designed to support research in speaker recognition and related areas: the KING corpus; KING-SAM, a derivative of the KING corpus; the YOHO corpus; the SWITCHBOARD corpus; and SPIDRE, a derivative subset of SWITCHBOARD. Each one has design characteristics which make it more appropriate for certain types of research or technology development than others. Our purpose here is to acquaint researchers with these properties so that they can make the best choice for their purposes. We will attempt to highlight the amount and nature of the speech data in each corpus, its intended use for training or test where applicable, and the strengths and limitations of each dataset for research and development in such areas as speaker identification, speaker verification, and speaker monitoring.

Full Paper

Bibliographic reference.  Godfrey, John / Graff, David / Martin, Alvin (1994): "Public databases for speaker recognition and verification", In ASRIV-1994, 39-42.