Factorized Sub-Space Estimation for Fast and Memory Effective I-vector Extraction

Cumani, Sandro; Laface, Pietro

doi:10.1109/taslp.2013.2290505

Cited by 14 publications

(21 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The slowness of training has often forced many researchers to limit their experimental validation, for example by limiting the number of training iterations, or by relaying on the results from a single run with random initialization. In addition, simplifications and approximations of the model have been proposed to reduce the computational load [6,7,8].…”

Section: Introductionmentioning

confidence: 99%

Unleashing the Unused Potential of i-Vectors Enabled by GPU Acceleration

et al. 2019

View full text Add to dashboard Cite

Speaker embeddings are continuous-value vector representations that allow easy comparison between voices of speakers with simple geometric operations. Among others, i-vector and x-vector have emerged as the mainstream methods for speaker embedding. In this paper, we illustrate the use of modern computation platform to harness the benefit of GPU acceleration for i-vector extraction. In particular, we achieve an acceleration of 3000 times in frame posterior computation compared to real time and 25 times in training the i-vector extractor compared to the CPU baseline from Kaldi toolkit. This significant speed-up allows the exploration of ideas that were hitherto impossible. In particular, we show that it is beneficial to update the universal background model (UBM) and re-compute frame alignments while training the i-vector extractor. Additionally, we are able to study different variations of i-vector extractors more rigorously than before. In this process, we reveal some undocumented details of Kaldi's i-vector extractor and show that it outperforms the standard formulation by a margin of 1 to 2% when tested with VoxCeleb speaker verification protocol. All of our findings are asserted by ensemble averaging the results from multiple runs with random start.

show abstract

Section: Introductionmentioning

confidence: 99%

Unleashing the Unused Potential of i-Vectors Enabled by GPU Acceleration

et al. 2019

View full text Add to dashboard Cite

show abstract

“…SPAM models are Gaussian mixture models with a subspace constraint, where each covariance matrix is represented as a weighted sum of globally shared full-rank matrices. In 2014, Sandro Cumani proposed an i-vector extractor factorization [16], for faster i-vector extraction and smaller memory footprint, where each row of the i-vector extractor matrix is represented as a linear combination of the atoms of a common dictionary with the assumption that it is not necessary to store all rows this matrix to perform ivector extraction.…”

Section: Introductionmentioning

confidence: 99%

“…In our approach to factorization, we were inspired by [16], but instead of factorizing each row, we perform factorization on the level submatrices of the i-vector extractor that represent individual GMM-UBM components. Also, our motivation is different, as we aim to greatly decrease the memory footprint and therefore substantially speedup the discriminative training.…”

Section: Introductionmentioning

confidence: 99%

Factorization of Discriminatively Trained i-Vector Extractor for Speaker Recognition

Novotny¹,

Plchot

Glembek

et al. 2019

Interspeech 2019

View full text Add to dashboard Cite

In this work, we continue in our research on i-vector extractor for speaker verification (SV) and we optimize its architecture for fast and effective discriminative training. We were motivated by computational and memory requirements caused by the large number of parameters of the original generative ivector model. Our aim is to preserve the power of the original generative model, and at the same time focus the model towards extraction of speaker-related information. We show that it is possible to represent a standard generative i-vector extractor by a model with significantly less parameters and obtain similar performance on SV tasks. We can further refine this compact model by discriminative training and obtain i-vectors that lead to better performance on various SV benchmarks representing different acoustic domains.

show abstract

“…Previous studies on rapid i-vector extraction have primarily optimized computations in the standard front-end factor analysis (FEFA) approach [1,2] by adopting new computational algorithms, often approximative in nature [6,8,9]. In this study, however, we focus on an alternative and straightforward compression of classic maximum a posteriori (MAP) [3] adapted GMM supervectors with a goal of obtaining fast execution times without compromising on ASV accuracy.…”

Section: Introductionmentioning

confidence: 99%

Supervector Compression Strategies to Speed up I-Vector System Development

Vestman¹,

Kinnunen²

2018

The Speaker and Language Recognition Workshop (Odyssey 2018)

View full text Add to dashboard Cite

The front-end factor analysis (FEFA), an extension of principal component analysis (PPCA) tailored to be used with Gaussian mixture models (GMMs), is currently the prevalent approach to extract compact utterance-level features (i-vectors) for automatic speaker verification (ASV) systems. Little research has been conducted comparing FEFA to the conventional PPCA applied to maximum a posteriori (MAP) adapted GMM supervectors. We study several alternative methods, including PPCA, factor analysis (FA), and two supervised approaches, supervised PPCA (SPPCA) and the recently proposed probabilistic partial least squares (PPLS), to compress MAP-adapted GMM supervectors. The resulting i-vectors are used in ASV tasks with a probabilistic linear discriminant analysis (PLDA) back-end. We experiment on two different datasets, on the telephone condition of NIST SRE 2010 and on the recent VoxCeleb corpus collected from YouTube videos containing celebrity interviews recorded in various acoustical and technical conditions. The results suggest that, in terms of ASV accuracy, the supervector compression approaches are on a par with FEFA. The supervised approaches did not result in improved performance. In comparison to FEFA, we obtained more than hundred-fold (100x) speedups in the total variability model (TVM) training using the PPCA and FA supervector compression approaches.

show abstract

Factorized Sub-Space Estimation for Fast and Memory Effective I-vector Extraction

Cited by 14 publications

References 26 publications

Unleashing the Unused Potential of i-Vectors Enabled by GPU Acceleration

Unleashing the Unused Potential of i-Vectors Enabled by GPU Acceleration

Factorization of Discriminatively Trained i-Vector Extractor for Speaker Recognition

Supervector Compression Strategies to Speed up I-Vector System Development

Contact Info

Product

Resources

About