This article presents Conformity as a new method for text-independent speaker recognition. Its principle is to compare statistical distributions of samples. We estimate these distributions by vector quantization using a single, universal codebook. One computes the frequency of apparition of each codeword for test and reference speech, and the frequencies are compared.
We perform several experiments with a database consisting of many speakers and a high number of tests, and conduct experiments in an opened test methodology. We compare the results to those obtained by other well-known methods.
Our new method yields good results when used alone. It may also be combined advantageously with the classical vector quantization method to which it is complementary. The joint use of these two methods permits an enhanced recognition rate.
Cite as: Thévenaz, P., Hügli, H. (1994) Conformity, a new method for text-independent speaker recognition. Proc. ESCA Workshop on Automatic Speaker Recognition, Identification and Verification, 63-66
@inproceedings{thevenaz94_asriv,
author={Philippe Thévenaz and Heinz Hügli},
title={{Conformity, a new method for text-independent speaker recognition}},
year=1994,
booktitle={Proc. ESCA Workshop on Automatic Speaker Recognition, Identification and Verification},
pages={63--66}
}