High-level speaker verification with support vector machines

Improved Phonetic Speaker Recognition Using Lattice Decoding

“…In their 2003 NIPS paper, Campbell et al demonstrated a phonetic speaker recognition system based on support vector machines [3]. One of the main innovations of the paper was the following "kernelized" version of the log-likelihood ratio:…”

“…In this paper, we compare 1-best phone decodings vs. lattice phone decodings for the purposes of performing phonetic speaker recognition. The phone decodings are used to compute relative frequencies of phone bigrams, which are then used as inputs for two standard phonetic speaker recognition systems: a conventional system based on loglikelihood ratios (LLRs) [1,2], and an SVM-based system similar to that of Campbell et al [3]. The results indicate that lattice decodings provide a much richer sampling of phonetic patterns than 1-best decodings.…”

“…For the experiments in this paper, we used relative frequencies computed from splits 6 through 10 as a background model-that is, to estimate the p(dijbkg) terms in equation (3)-for all speaker models belonging to splits 1 through 5, and vice versa. Similarly, we used relative frequencies derived from the training conversation sides of speaker B to estimate p(dijspkB) for all i 2 1; : : : ; M. Note that the form of the LLR used in equation (3) is equivalent to that used in [1], [2], and [3].…”

“…In this paper, we present results on the Switchboard-2 corpus, where we compare 1-best phone decodings versus lattice phone decodings for the purposes of performing phonetic speaker recognition. The phone decodings are used to compute relative frequencies of phone bigrams, which are then used as inputs for two standard phonetic speaker recognition systems: a system based on log-likelihood ratios (LLRs) [1,2], and a system based on support vector machines (SVMs) [3]. In each experiment, the lattice phone decodings achieve relative reductions in equal-error rate (EER) of between 31% and 66% below the EERs of the 1-best phone decodings.…”

“…This work was subsequently extended in various papers, such as the work of the "SuperSID" team at the JHU 2002 Summer Workshop [5,6,7]. In 2003, Campbell et al used support vector machines (SVMs) to train phonetic speaker models [3].…”