Speaker-dependent recognition of continuous speech is generally held to be an easier task than speaker-independent continuous speech recognition. In this paper we describe experiments which attempt to quantify this assertion for a particular feature-based recogniser. The primary basis for the study is an investigation of how well a feature-based recogniser developed to work near-perfectly on one particular speaker works on continuous speech examples from a large number of other speakers.
Bibliographic reference. O'Kane, M. / Kenne, P. / Landy, D. / Atkins, S. (1991): "Generalising from single-speaker recognition in a feature-based recogniser", In EUROSPEECH-1991, 409-412.