Detail publikace

Employment of Subspace Gaussian Mixture Models in Speaker Recognition

MOTLÍČEK, P. DEY, S. MADIKERI, S. BURGET, L.

Originální název

Employment of Subspace Gaussian Mixture Models in Speaker Recognition

Anglický název

Employment of Subspace Gaussian Mixture Models in Speaker Recognition

Jazyk

en

Originální abstrakt

This paper presents Subspace Gaussian Mixture Model (SGMM) approach employed as a probabilistic generative model to estimate speaker vector representations to be subsequently used in the speaker verification task. SGMMs have already been shown to significantly outperform traditional HMM/GMMs in Automatic Speech Recognition (ASR) applications. An extension to the basic SGMM framework allows to robustly estimate low-dimensional speaker vectors and exploit them for speaker adaptation. We propose a speaker verification framework based on low-dimensional speaker vectors estimated using SGMMs, trained in ASR manner using manual transcriptions. To test the robustness of the system, we evaluate the proposed approach with respect to the state-of-the-art i-vector extractor on the NIST SRE 2010 evaluation set and on four different length-utterance conditions: 3sec-10sec, 10 sec-30 sec, 30 sec-60 sec and full (untruncated) utterances. Experimental results reveal that while i-vector system performs better on truncated 3sec to 10sec and 10 sec to 30 sec utterances, noticeable improvements are observed with SGMMs especially on full length-utterance durations. Eventually, the proposed SGMM approach exhibits complementary properties and can thus be efficiently fused with i-vector based speaker verification system.

Anglický abstrakt

This paper presents Subspace Gaussian Mixture Model (SGMM) approach employed as a probabilistic generative model to estimate speaker vector representations to be subsequently used in the speaker verification task. SGMMs have already been shown to significantly outperform traditional HMM/GMMs in Automatic Speech Recognition (ASR) applications. An extension to the basic SGMM framework allows to robustly estimate low-dimensional speaker vectors and exploit them for speaker adaptation. We propose a speaker verification framework based on low-dimensional speaker vectors estimated using SGMMs, trained in ASR manner using manual transcriptions. To test the robustness of the system, we evaluate the proposed approach with respect to the state-of-the-art i-vector extractor on the NIST SRE 2010 evaluation set and on four different length-utterance conditions: 3sec-10sec, 10 sec-30 sec, 30 sec-60 sec and full (untruncated) utterances. Experimental results reveal that while i-vector system performs better on truncated 3sec to 10sec and 10 sec to 30 sec utterances, noticeable improvements are observed with SGMMs especially on full length-utterance durations. Eventually, the proposed SGMM approach exhibits complementary properties and can thus be efficiently fused with i-vector based speaker verification system.

Dokumenty

BibTex


@inproceedings{BUT119895,
  author="Petr {Motlíček} and Subhadeep {Dey} and Srikanth {Madikeri} and Lukáš {Burget}",
  title="Employment of Subspace Gaussian Mixture Models in Speaker Recognition",
  annote="This paper presents Subspace Gaussian Mixture Model (SGMM) approach employed as
a probabilistic generative model to estimate speaker vector representations to be
subsequently used in the speaker verification task. SGMMs have already been shown
to significantly outperform traditional HMM/GMMs in Automatic Speech Recognition
(ASR) applications. An extension to the basic SGMM framework allows to robustly
estimate low-dimensional speaker vectors and exploit them for speaker adaptation.
We propose a speaker verification framework based on low-dimensional speaker
vectors estimated using SGMMs, trained in ASR manner using manual transcriptions.
To test the robustness of the system, we evaluate the proposed approach with
respect to the state-of-the-art i-vector extractor on the NIST SRE 2010
evaluation set and on four different length-utterance conditions: 3sec-10sec, 10
sec-30 sec, 30 sec-60 sec and full (untruncated) utterances. Experimental results
reveal that while i-vector system performs better on truncated 3sec to 10sec and
10 sec to 30 sec utterances, noticeable improvements are observed with SGMMs
especially on full length-utterance durations. Eventually, the proposed SGMM
approach exhibits complementary properties and can thus be efficiently fused with
i-vector based speaker verification system.",
  address="IEEE Signal Processing Society",
  booktitle="Proceedings of 2015 IEEE International Conference on Acoustics, Speech and Signal Processing",
  chapter="119895",
  doi="10.1109/ICASSP.2015.7178811",
  edition="NEUVEDEN",
  howpublished="electronic, physical medium",
  institution="IEEE Signal Processing Society",
  year="2015",
  month="april",
  pages="4445--4449",
  publisher="IEEE Signal Processing Society",
  type="conference paper"
}