Speaker verification using sequence discriminant support vector machines

Wan, Vincent; Renals, Steve

doi:10.1109/tsa.2004.841042

Cited by 167 publications

(93 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…During testing, the segment score is obtained by averaging the scores of the SVM output for each frame. There are also others applications of SVM in ASV that used kernels sequences [8] [9].…”

Section: Introductionmentioning

confidence: 99%

Support Vector Gmms for Speaker Verification

Dehak

Chollet

2006

2006 IEEE Odyssey - The Speaker and Language Recognition Workshop

View full text Add to dashboard Cite

This article presents a new approach using the discrimination power of Support Vectors Machines (SVM) in combination with Gaussian Mixture Models (GMM) for Automatic Speaker Verification (ASV). In this combination SVMs are applied in the GMM model space. Each point of this space represents a GMM speaker model. The kernel which is used for the SVM allows the computation of a similarity between GMM models. It was calculated using the Kullback-Leibler (KL) divergence. The results of this new approach show a clear improvement compared to a simple GMM system on the NIST2005 Speaker Recognition Evaluation primary task.

show abstract

Section: Introductionmentioning

confidence: 99%

Support Vector Gmms for Speaker Verification

Dehak

Chollet

2006

2006 IEEE Odyssey - The Speaker and Language Recognition Workshop

View full text Add to dashboard Cite

show abstract

“…The magic transpires now: if the graph will contain an abundance of pairs of th1`s and th2`s from the same musical composition, a diagonal line will compose. The conception abaft the formation of that line is straightforward: the rate at which the crests (the minuscule crosses from the improved spectrogram) in the database musical composition appear will be identically tantamount rate in which the tops show up in the recorded specimen, so on the off chance that you match these circumstances, the directions on the dissipate diagram will develop perpetually (to one side top of the diagram) as the time goes on both tomahawks [9][10] [13].…”

Section: International Journal Of Computer Applications (0975 -8887) mentioning

confidence: 99%

Audio Scenarios Detection Technique

Kadam¹,

Kagalkar²

2015

IJCA

View full text Add to dashboard Cite

The proposed research objective is to add to a framework for programmed recognition of sound. In this framework the real errand is to distinguish any information sound stream investigate it & anticipate the likelihood of diverse sounds show up in it. To create and industrially conveyed an adaptable sound web crawler a flexible sound search engine. The calculation is clamor and contortion safe, computationally productive, and hugely adaptable, equipped for rapidly recognizing a short portion of sound stream caught through a phone microphone in the presence of frontal area voices and other predominant commotion, and through voice codec pressure, out of a database of over accessible tracks. The algorithm utilizes a combinatorial hashed time-recurrence group of stars examination of the sound, yielding ordinary properties, for example, transparency, in which numerous tracks combined may each be distinguished.

show abstract

“…It is therefore important that the kernel function includes some form of dynamic range normalisation. One option is Spherical Normalisation [7] where each feature vector is mapped onto the surface of a unit sphere. An alternative approach is to perform normalisation at the kernel level.…”

Section: Dynamic Range Normalisationmentioning

confidence: 99%

Multiple kernel learning for speaker verification

Longworth

Gales

2008

2008 IEEE International Conference on Acoustics, Speech and Signal Processing

View full text Add to dashboard Cite

Many speaker verification (SV) systems combine multiple classifiers using score-fusion to improve system performance. For SVM classifiers, an alternative strategy is to combine at the kernel level. This involves finding a suitable kernel weighting, known as Multiple Kernel Learning (MKL). Recently, an efficient maximum-margin scheme for MKL has been proposed. This work examines several refinements to this scheme for SV. The standard scheme has a known tendency towards sparse weightings, which may not be optimal for SV. A regularisation term is proposed, allowing the appropriate level of sparsity to be selected. Cross-speaker tying of kernel weights is also applied to improve robustness. Various combinations of dynamic kernels were evaluated, including derivative and parametric kernels based upon different model structures. The performance achieved on the NIST 2002 SRE when combining five kernels was 7.78% EER.

show abstract

Speaker verification using sequence discriminant support vector machines

Cited by 167 publications

References 15 publications

Support Vector Gmms for Speaker Verification

Support Vector Gmms for Speaker Verification

Audio Scenarios Detection Technique

Multiple kernel learning for speaker verification

Contact Info

Product

Resources

About