Eigenvoice modeling with sparse training data

By Patrick Kenny, Gilles Boulianne, Pierre Dumouchel. IEEE Transactions on Speech and Audio Processing, 2005

550 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computational Mechanics, Signal Processing

View Paper: https://www.semanticscholar.org/paper/db8b2be021e81a36c038a31b9d4217706e4eae2b

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 Computational Mechanics

In Signal Processing

Citations in the corpus

Joint Factor Analysis of Speaker and Session Variability: Theory and Algorithms (2006, methodology): “The proof of Theorem 3 is formally identical to the proof of Proposition 2 in [2] To evaluate the integral we appeal to the formula for the Fourier-Laplace transform of the Gaussian kernel [10] We now turn to the problem of estimating the hyperparameter set Λ from a training set comprising several…”
Front-End Factor Analysis for Speaker Verification (2011, background): “The process of training the total variability matrix is exactly the same as learning the eigenvoice matrix (see [11]), except for one important difference: in eigenvoice training, all the recordings of a given speaker are considered to belong to the same person; in the case of the total variability ”
Joint Factor Analysis Versus Eigenchannels in Speaker Recognition (2007, methodology): “…recognition is discriminative rather than generative, and it is concerned with finding feature representations which are immune to channel variability rather than modeling this type of variability, the key algorithm in “nuisance attribute projection” is also formulated as an eigenvalue problem.”
Generalized I-vector Representation with Phonetic Tokenizations and Tandem Features for both Text Independent and Text Dependent Speaker Verification (2015, methodology): “Considering a C - component GMM and D dimensional acoustic features, the total variability matrix T is a CD × K matrix which is estimated the same way as learning the eigenvoice matrix in [10] except that here we consider that every utterance is produced by a new speaker [3].”
Bayesian Speaker Verification with Heavy-Tailed Priors (2010, methodology): “The feature vector associated with a given recording is just the MAP estimate of w, calculated as in [16].”
Support vector machines using GMM supervectors for speaker verification (2006, methodology): “This GMM supervector can be used to characterize the speaker and channel using eigenvoices and eigenchannels methods, respectively [3].”
Learning Speaker-Specific Characteristics With a Deep Neural Architecture (2011, background): “• FA-GSL, a combination of factor analysis (FA) [113] and GSL.”
A Study of Interspeaker Variability in Speaker Verification (2008, methodology): “and u = 0, the assumption is the same as in eigenvoice MAP [8].”
A Review of Speaker Diarization: Recent Advances with Deep Learning (2021, methodology): “process of extracting an i-vector w for the given recording is formulated as a MAP estimation problem [123, 30] using the Baum–Welch statistics extracted using the UBM, mean supervector mI , and total variability matrix T trained from the EM algorithm as parameters.”
How Far Are We from Robust Voice Conversion: A Survey (2020, methodology): “In i-vector [18], the speaker and the utterance-dependent GMM supervector M are modeled as”
Speaker verification and spoken language identification using a generalized i-vector framework with phonetic tokenizations and tandem features (2014, methodology): “Considering a C -component GMM and D dimensional acoustic features, the total variability matrix T is a CD × K matrix which is estimated the same way as learning the eigenvoice matrix in [19] except that here we consider that every utterance is produced by a new speaker [2].”
Deep bottleneck network based i-vector representation for language identification (2015, methodology): “The loading matrix T ∈ RKd×q is a lowrank rectangular matrix, which is trained similarly to the eigenvoice method [10].”
Language Recognition via i-vectors and Dimensionality Reduction (2011, methodology): “Speech activity detection is then applied and the speech is normalized following a standard normal distribution.”