EUROSPEECH 2003 - INTERSPEECH 2003
8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003

        

On the Amount of Speech Data Necessary for Successful Speaker Identification

Ales Padrta, Vlasta Radova

University of West Bohemia in Pilsen, Czech Republic

The paper deals with the dependence between the speaker identification performance and the amount of test data. Three speaker identification procedures based on hidden Markov models (HMMs) of phonemes are presented here. One, which is quite commonly used in the speaker recognition systems based on HMMs, uses the likelihood of the whole utterance for speaker identification. The other two that are proposed in this paper are based on the majority voting rule. The experiments were performed for two different situations: either both training and test data were obtained from the same channel, or they were obtained from different channels. All experiments show that the proposed speaker identification procedure based on the majority voting rule for sequences of phonemes allows us to reduce the amount of test data necessary for successful speaker identification.

Full Paper

Bibliographic reference.  Padrta, Ales / Radova, Vlasta (2003): "On the amount of speech data necessary for successful speaker identification", In EUROSPEECH-2003, 3021-3024.