Non-acoustic speaker recognition

[摘要] ENGLISH ABSTRACT: In this study the phoneme labels derived from a phoneme recogniser are used for phoneticspeaker recognition. The time-dependencies among phonemes are modelled by usinghidden Markov models (HMMs) for the speaker models. Experiments are done using firstorderand second-order HMMs and various smoothing techniques are examined to addressthe problem of data scarcity. The use of word labels for lexical speaker recognition is alsoinvestigated. Single word frequencies are counted and the use of various word selectionsas feature sets are investigated. During April 2004, the University of Stellenbosch, in collaborationwith Spescom DataVoice, participated in an international speaker verificationcompetition presented by the National Institute of Standards and Technology (NIST). TheUniversity of Stellenbosch submitted phonetic and lexical (non-acoustic) speaker recognitionsystems and a fused system (the primary system) that fuses the acoustic system ofSpescom DataVoice with the non-acoustic systems of the University of Stellenbosch. Theresults were evaluated by means of a cost model. Based on the cost model, the primarysystem obtained second and third position in the two categories that were submitted.

[发布机构] Stellenbosch University