I-vector Transformation Using Conditional Generative Adversarial Networks for Short Utterance Speaker Verification

Jiacen Zhang, Nakamasa Inoue, Koichi Shinoda


I-vector based text-independent speaker verification (SV) systems often have poor performance with short utterances, as the biased phonetic distribution in a short utterance makes the extracted i-vector unreliable. This paper proposes an i-vector compensation method using a generative adversarial network (GAN), where its generator network is trained to generate a compensated i-vector from a short-utterance i-vector and its discriminator network is trained to determine whether an i-vector is generated by the generator or the one extracted from a long utterance. Additionally, we assign two other learning tasks to the GAN to stabilize its training and to make the generated i-vector more speaker-specific. Speaker verification experiments on the NIST SRE 2008 “10sec-10sec” condition show that after applying our method, the equal error rate reduced by 11.3% from the conventional i-vector and PLDA system.


 DOI: 10.21437/Interspeech.2018-1680

Cite as: Zhang, J., Inoue, N., Shinoda, K. (2018) I-vector Transformation Using Conditional Generative Adversarial Networks for Short Utterance Speaker Verification. Proc. Interspeech 2018, 3613-3617, DOI: 10.21437/Interspeech.2018-1680.


@inproceedings{Zhang2018,
  author={Jiacen Zhang and Nakamasa Inoue and Koichi Shinoda},
  title={I-vector Transformation Using Conditional Generative Adversarial Networks for Short Utterance Speaker Verification},
  year=2018,
  booktitle={Proc. Interspeech 2018},
  pages={3613--3617},
  doi={10.21437/Interspeech.2018-1680},
  url={http://dx.doi.org/10.21437/Interspeech.2018-1680}
}