Second European Conference on Speech Communication and Technology

Genova, Italy
September 24-26, 1991


Simultaneous Recognition of Concurrent Speech Signals Using Hidden Markov Model Decomposition

A. P. Varga, Roger K. Moore

Speech Research Unit, Defence Research Agency, Electronics Division, R. S. R. E. , Malvern, Great Britain

This paper addresses the problem of automatic recognition of two simultaneous speech signals, that is the recognition of speech in the presence of speech from interfering talkers. The approach used is that of hidden Markov model based signal decomposition in which two or more concurrent signals are recognised simultaneously. In order to accommodate concurrent dynamically varying processes the decomposition technique uses conventional hidden Markov modelling of speech, but a generalisation of the conventional recognition algorithm. The performance of a decomposition based recogniser is compared with that of a conventional recogniser; the results demonstrate the ability of the new technique to decompose and recognise interfering signals as structurally complex as speech. Keywords: Speech recognition, hidden Markov model based decomposition, recognition of simultaneous speech signals.

Full Paper

Bibliographic reference.  Varga, A. P. / Moore, Roger K. (1991): "Simultaneous recognition of concurrent speech signals using hidden Markov model decomposition", In EUROSPEECH-1991, 1175-1178.