Towards PLDA-RBM based speaker recognition in mobile environment: Designing stacked/deep PLDA-RBM systems

Andreas Nautsch; Hong Hao; Themos Stafylakis; Christian Rathgeb; Christoph Busch

doi:10.1109/ICASSP.2016.7472640

Towards PLDA-RBM based speaker recognition in mobile environment: Designing stacked/deep PLDA-RBM systems

Andreas Nautsch, Hong Hao, Themos Stafylakis, Christian Rathgeb, Christoph Busch

Department of Applied Mathematics and Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

Abstract

The vast majority of text-independent speaker recognition systems rely on intermediate-sized vectors (i-vectors), which are compared by probabilistic linear discriminant analysis (PLDA). This paper proposes a PLDA-alike approach with restricted Boltzmann machines for i-vector based speaker recognition: two deep architectures are presented and examined, which aim at suppressing channel effects and recovering speaker-discriminative information on back-ends trained on a small dataset. Experiments are carried out on the MOBIO SRE'13 database, which is a challenging and publicly available dataset for mobile speaker recognition with limited amounts of training data. The experiments show that the proposed system outperforms the baseline i-vector/PLDA approach by relative gains of 31% on female and 9% on male speakers in terms of half total error rate.

Original language	English
Title of host publication	Proceedings of 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)
Number of pages	5
Publisher	IEEE
Publication date	2016
DOIs	https://doi.org/10.1109/ICASSP.2016.7472640
Publication status	Published - 2016
Event	2016 IEEE International Conference on Acoustics, Speech, and Signal Processing - Shanghai, China Duration: 20 Mar 2016 → 25 Mar 2016 Conference number: 41

Conference

Conference	2016 IEEE International Conference on Acoustics, Speech, and Signal Processing
Number	41
Country/Territory	China
City	Shanghai
Period	20/03/2016 → 25/03/2016

Series	I E E E International Conference on Acoustics, Speech and Signal Processing. Proceedings
ISSN	1520-6149

Keywords

Deep learning
MOBIO
PLDA-RBM
Speaker recognition

Access to Document

10.1109/ICASSP.2016.7472640

OpenUrl availability

Full text

Cite this

Nautsch, Andreas ; Hao, Hong ; Stafylakis, Themos et al. / Towards PLDA-RBM based speaker recognition in mobile environment: Designing stacked/deep PLDA-RBM systems. Proceedings of 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2016. (I E E E International Conference on Acoustics, Speech and Signal Processing. Proceedings).

@inproceedings{f84ec9b5ea8344199b34702e1cce464c,

title = "Towards PLDA-RBM based speaker recognition in mobile environment: Designing stacked/deep PLDA-RBM systems",

abstract = "The vast majority of text-independent speaker recognition systems rely on intermediate-sized vectors (i-vectors), which are compared by probabilistic linear discriminant analysis (PLDA). This paper proposes a PLDA-alike approach with restricted Boltzmann machines for i-vector based speaker recognition: two deep architectures are presented and examined, which aim at suppressing channel effects and recovering speaker-discriminative information on back-ends trained on a small dataset. Experiments are carried out on the MOBIO SRE'13 database, which is a challenging and publicly available dataset for mobile speaker recognition with limited amounts of training data. The experiments show that the proposed system outperforms the baseline i-vector/PLDA approach by relative gains of 31% on female and 9% on male speakers in terms of half total error rate.",

keywords = "Deep learning, MOBIO, PLDA-RBM, Speaker recognition",

author = "Andreas Nautsch and Hong Hao and Themos Stafylakis and Christian Rathgeb and Christoph Busch",

year = "2016",

doi = "10.1109/ICASSP.2016.7472640",

language = "English",

series = "I E E E International Conference on Acoustics, Speech and Signal Processing. Proceedings",

publisher = "IEEE",

booktitle = "Proceedings of 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)",

address = "United States",

note = "2016 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2016 ; Conference date: 20-03-2016 Through 25-03-2016",

}

Nautsch, A, Hao, H, Stafylakis, T, Rathgeb, C & Busch, C 2016, Towards PLDA-RBM based speaker recognition in mobile environment: Designing stacked/deep PLDA-RBM systems. in Proceedings of 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, I E E E International Conference on Acoustics, Speech and Signal Processing. Proceedings, 2016 IEEE International Conference on Acoustics, Speech, and Signal Processing, Shanghai, China, 20/03/2016. https://doi.org/10.1109/ICASSP.2016.7472640

Towards PLDA-RBM based speaker recognition in mobile environment: Designing stacked/deep PLDA-RBM systems. / Nautsch, Andreas; Hao, Hong; Stafylakis, Themos et al.
Proceedings of 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE, 2016. (I E E E International Conference on Acoustics, Speech and Signal Processing. Proceedings).

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

TY - GEN

T1 - Towards PLDA-RBM based speaker recognition in mobile environment: Designing stacked/deep PLDA-RBM systems

AU - Nautsch, Andreas

AU - Hao, Hong

AU - Stafylakis, Themos

AU - Rathgeb, Christian

AU - Busch, Christoph

N1 - Conference code: 41

PY - 2016

Y1 - 2016

N2 - The vast majority of text-independent speaker recognition systems rely on intermediate-sized vectors (i-vectors), which are compared by probabilistic linear discriminant analysis (PLDA). This paper proposes a PLDA-alike approach with restricted Boltzmann machines for i-vector based speaker recognition: two deep architectures are presented and examined, which aim at suppressing channel effects and recovering speaker-discriminative information on back-ends trained on a small dataset. Experiments are carried out on the MOBIO SRE'13 database, which is a challenging and publicly available dataset for mobile speaker recognition with limited amounts of training data. The experiments show that the proposed system outperforms the baseline i-vector/PLDA approach by relative gains of 31% on female and 9% on male speakers in terms of half total error rate.

AB - The vast majority of text-independent speaker recognition systems rely on intermediate-sized vectors (i-vectors), which are compared by probabilistic linear discriminant analysis (PLDA). This paper proposes a PLDA-alike approach with restricted Boltzmann machines for i-vector based speaker recognition: two deep architectures are presented and examined, which aim at suppressing channel effects and recovering speaker-discriminative information on back-ends trained on a small dataset. Experiments are carried out on the MOBIO SRE'13 database, which is a challenging and publicly available dataset for mobile speaker recognition with limited amounts of training data. The experiments show that the proposed system outperforms the baseline i-vector/PLDA approach by relative gains of 31% on female and 9% on male speakers in terms of half total error rate.

KW - Deep learning

KW - MOBIO

KW - PLDA-RBM

KW - Speaker recognition

U2 - 10.1109/ICASSP.2016.7472640

DO - 10.1109/ICASSP.2016.7472640

M3 - Article in proceedings

T3 - I E E E International Conference on Acoustics, Speech and Signal Processing. Proceedings

BT - Proceedings of 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

PB - IEEE

T2 - 2016 IEEE International Conference on Acoustics, Speech, and Signal Processing

Y2 - 20 March 2016 through 25 March 2016

ER -

Nautsch A, Hao H, Stafylakis T, Rathgeb C, Busch C. Towards PLDA-RBM based speaker recognition in mobile environment: Designing stacked/deep PLDA-RBM systems. In Proceedings of 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). IEEE. 2016. (I E E E International Conference on Acoustics, Speech and Signal Processing. Proceedings). doi: 10.1109/ICASSP.2016.7472640

Towards PLDA-RBM based speaker recognition in mobile environment: Designing stacked/deep PLDA-RBM systems

Abstract

Conference

Keywords

Access to Document

OpenUrl availability

Fingerprint

Cite this