Support vector machines using GMM supervectors for speaker verification

By William M. Campbell, Douglas E. Sturim, Douglas A. Reynolds. 2006

1,107 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Signal Processing

Citations in the corpus

Exploring the Encoding Layer and Loss Function in End-to-End Speaker and Language Recognition System (2018, background): “They are problems in that we are trying to retrieve information about an entire utterance rather than specific word content [1].”
Generalized I-vector Representation with Phonetic Tokenizations and Tandem Features for both Text Independent and Text Dependent Speaker Verification (2015, methodology): “Third, within this i-vector space, variability compensation methods, such as Within-Class Covariance Normalization (WCCN) [7], Linear Discriminative Analysis (LDA) and Nuisance Attribute Projection (NAP) [1] are performed to reduce the variability for subsequent scoring methods (e.”
Spoofing and countermeasures for speaker verification: A survey (2015, background): “The combination of GMM supervectors and support vector machine (SVM) classifiers was developed to enable the discriminative training of generative models (Campbell et al., 2006).”
Deep Speaker Feature Learning for Text-Independent Speaker Verification (2017, methodology): “, the SVM model for the GMM-UBM approach [4] and the PLDA model for the i-vector approach [5].”
Non-parallel voice conversion using i-vector PLDA: towards unifying speaker verification and transformation (2017, positive): “These two GMMs correspond either to the (source, target) pair without any conversion or (converted, target) pair.”
Finding Difficult Speakers in Automatic Speaker Recognition (2011, methodology): “One such approach is an SVM classifier that uses GMM supervectors as features [14].”
EmotionSense: a mobile phones based adaptive platform for experimental social psychology research (2010, methodology): “Alternative SVM-based schemes, including the popular GMMsupervector [7] and MLLR [28] kernel classifiers, were not considered as they are generally suitable for binary classification tasks.”
An overview of text-independent speaker recognition: From features to supervectors (2010, methodology): “Given the demonstrated excellent performance of the JFA compensation and Gaussian supervector SVMs [38], it seems appropriate to ask how they compare with each other, and whether they could be combined?”
A review of biometric technology along with trends and prospects (2014, background): “Table 4: Summary of behavioral modalities: Modality Feature Set Recognition Techniques Datasets Voice Spectrum, Glottal pulse features, pitch, energy, duration, rhythm, temporal features, phones, idiolects, semantics, accent, pronunciation, LPCC [204], MFCC [205], VQ [206-208], HMM [209], GMM [210, ”
What Does the Speaker Embedding Encode? (2017, positive): “To address this limitation, researchers explored methods to correlate different Gaussian components through global transformations, leading to the super-vector representation [3, 4], which achieved competitive results when combined with support vector machines.”
Speaker verification and spoken language identification using a generalized i-vector framework with phonetic tokenizations and tandem features (2014, methodology): “Third, within this i-vector space, variability compensation methods, such as Within-Class Covariance Normalization (WCCN) [4], Linear Discriminative Analysis (LDA) and Nuisance Attribute Projection (NAP) [5], are performed to reduce the variability for the subsequent modeling methods (e.”
Spoken Language Recognition: From Fundamentals to Practice (2013, methodology): “;MM T by stacking the mean vectors of all the adapted mixture components [18].”
A novel scheme for speaker recognition using a phonetically-aware deep neural network (2014, methodology): “In the GMM-support vector machine (SVM) [13] framework, the speakerspecific GMM is obtained just as in the GMM-UBM framework, but the means of the GMM are concatenated to generate a vector called a supervector, which is then input to the SVM.”
A Novel Learnable Dictionary Encoding Layer for End-to-End Language Identification (2018, methodology): “ID 3s 10s 30s 3s 10s 30s 1 GMM i-vector SDC GMM Supervector 20.46 8.29 3.02 17.71 7.00 2.27 2 CNN-TAP CNN FeatureMaps TAP 9.98 3.24 1.73 11.28 5.76 3.96 3 CNN-LDE(C=16) CNN FeatureMaps LDE 9.61 3.71 1.74 8.89 2.73 1.13 4 CNN-LDE(C=32) CNN FeatureMaps LDE 8.70 2.94 1.41 8.12 2.45 0.98 5 CNN-LDE(C=64)”
SVM Based Speaker Verification using a GMM Supervector Kernel and NAP Variability Compensation (2006, methodology): “Second, for a nonlinear kernel [10], SVM NAP uses a nonlinear expanded version of the GMM supervector.”
Audio-based multimedia event detection using deep recurrent neural networks (2016, methodology): “The mean vectors and the diagonals of the covariance matrices of a GMM are concatenated to yield a “supervector” [1], which is used as the clip-level representation, such as in [2].”