Exploring kernel discriminant analysis for speaker verification with limited test data

This paper analyses the probabilistic linear discriminant analysis (PLDA) speaker verification approach with limited development data. This paper investigates the use of the median as the central tendency of a speaker's i-vector representation, and the effectiveness of weighted discriminative techniques on the performance of state-of-the-art length-normalised Gaussian PLDA (GPLDA) speaker verification systems. The analysis within shows that the median (using a median fisher discriminator (MFD)) provides a better representation of a speaker when the number of representative i-vectors available during development is reduced, and that further, usage of the pair-wise weighting approach in weighted LDA and weighted MFD provides further improvement in limited development conditions. Best performance is obtained using a weighted MFD approach, which shows over 10% relative improvement in EER over the baseline GPLDA system on mismatched and interview-interview conditions.

Relevance factor of maximum a posteriori adaptation for GMM–NAP–SVM in speaker and language recognition

In this paper, we present a Mixture Linear Prediction based approach for robust Gammatone Cepstral Coefficients extraction (MLPGCCs). The proposed method provides performance improvement of Automatic Speaker Verification (ASV) using i-vector and Gaussian Probabilistic Linear Discriminant Analysis GPLDA modeling under transmission channel noise. The performance of the extracted MLPGCCs was evaluated using the NIST 2008 database where a single channel microphone recorded conversational speech. The system is analyzed in the presence of different channel transmission noises such as Additive White Gaussian (AWGN) and Rayleigh fading at various Signals to Noise Ratio (SNR) levels. The evaluation results show that the MLPGCCs features are a promising way for the ASV task. Indeed, the speaker verification performance using the MLPGCCs proposed features is significantly improved compared to the conventional Gammatone Frequency Cepstral Coefficients (GFCCs) and Mel Frequency Cepstral Coefficients (MFCCs) features. For speech signals corrupted with AWGN noise at SNRs ranging from (-5 dB to 15 dB), we obtain a significant reduction of the Equal Error Rate (EER) ranging from 9.41% to 6.65% and 3.72% to 1.50%, compared with conventional MFCCs and GFCCs features respectively. In addition, when the test speech signals are corrupted with Rayleigh fading channel we achieve an EER reduction ranging from 23.63% to 7.8% and from 10.88% to 6.8% compared with conventional MFCCs and GFCCs, respectively. We also found that the combination of GFCCs and MLPGCCs gives the highest performance of speaker verification system. The best performance combination achieved is around EER from 0.43% to 0.59% and 1.92% to 3.88%.

Previous researches show that the approaches based on the total variability space (TVS) followed by Gaussian probabilistic linear discriminant analysis (GPLDA) work effectively for dealing with convolutional noise (such as channel noise) and can bring some degree of gains in term of accuracy under additive noisy environment as well. However they meet difficulty while many types of noises are unseen and non-stationary in real world. To address this issue, we introduce the robust principal component analysis (RPCA) into the TVS modeled speaker verification system, called RPCA-TVS, which regards the noise spectrum as the low-rank component and the speech spectrum as the sparse component in short-time Fourier transform (SFT) domain. The highlighting of this paper is to improve the robustness of speaker verification under additive noisy environment, especially in non-stationary and unseen noise conditions. For evaluating the performance, we designed and generated an additive noisy corpus, based on the TIMIT and NUST603-2014 database, using the NaFT tools with 12 types of noise samples deriving from NOISEX-92 and FREESOUND. Experimental results demonstrate that the proposed RPCA-TVS can achieve better performance than the competing methods at various signal-to-noise ratio (SNR) levels. Especially, RPCA-TVS reduces the equal error rate (EER) by 5.12 % in average than the multi-condition system under additive noise conditions at SNR = 8 dB.

This paper analyses the short utterance probabilistic linear discriminant analysis (PLDA) speaker verification with utterance partitioning and short utterance variance (SUV) modelling approaches. Experimental studies have found that instead of using single long-utterance as enrolment data, if long enrolled utterance is partitioned into multiple short utterances and average of short utterance i-vectors is used as enrolled data, that improves the Gaussian PLDA (GPLDA) speaker verification. This is because short utterance i-vectors have speaker, session and utterance variations, and utterance-partitioning approach compensates the utterance variation. Subsequently, SUV-PLDA is also studied with utterance partitioning approach, and utterance partitioning-based SUV-GPLDA system shows relative improvement of 9% and 16% in EER for NIST 2008 and NIST 2010 truncated 10sec-10sec evaluation condition as utterance partitioning approach compensates the utterance variation and SUV modelling approach compensates the mismatch between full-length development data and short-length evaluation data.

We propose a method to improve speaker verification performance when a test utterance is very short. In some situations with short test utterances, performance of ivector/probabilistic linear discriminant analysis systems degrades. The proposed method transforms short-utterance feature vectors to adequate vectors using a deep neural network, which compensate for short utterances. To reduce the dimensionality of the search space, we extract several principal components from the residual vectors between every long utterance i-vector in a development set and its truncated short utterance i-vector. Then an input i-vector of the network is transformed by linear combination of these directions. In this case, network outputs correspond to weights for linear combination of principal components. We use public speech databases to evaluate the method. The experimental results on short2-10sec condition (det6, male portion) of the NIST 2008 speaker recognition evaluation corpus show that the proposed method reduces the minimum detection cost relative to the baseline system, which uses linear discriminant analysis transformed i-vectors as features.

In the biometrics community, face and speaker recognition are mature fields in which several systems have been proposed over the past twenty years. While existing systems perform well under controlled recording conditions, mismatch caused by the use of different sensors or a lack of cooperation from the subject still significantly affects performance, especially in challenging scenarios such as in forensics. Furthermore, existing methods suffer from scalability issues, which prevents them from taking advantage of increasingly large amounts of training data. This is otherwise a promising approach to improve accuracy in such challenging scenarios. In this thesis we address these problems of mismatch and complexity by developing scalable probabilistic models that we apply to face, speaker and bimodal recognition. Our contributions are four-fold. First, we propose a unified framework for session variability modeling techniques based on Gaussian mixture models (GMM), that encompasses inter-session variability (ISV) modeling, joint factor analysis (JFA) and total variability (TV) modeling. Second, we propose a novel exact and scalable formulation of probabilistic linear discriminant analysis (PLDA), which is a probabilistic and generative framework that models between-class and within-class variations. This formulation solves a major scalability issue, by improving both the time complexity of the training procedure from cubic to linear with respect to the number of samples per class, and the complexity of the scoring procedure. Furthermore, the implementations of all the proposed techniques are integrated into a novel collaborative open source software library called Bob 1 that enforces fair evaluations and encourages reproducible research. Fourth and finally, large-scale experiments are conducted with all of the above machine learning algorithms on several databases such as FRGC for face recognition, NIST SRE12 for speaker recognition and MOBIO for bimodal recognition, showing competitive performances.

Explicit modelling of session variability for speaker verification

To reduce the memory and timing complexity of the Side-Channel Attacks (SCA), dimensionality reduction techniques are usually applied to the measurements. They aim to detect the so-called Points of Interest (PoIs), which are time samples which (jointly) depend on some sensitive information (e.g. secret key sub-parts), and exploit them to extract information. The extraction is done through the use of functions which combine the measurement time samples. Examples of combining functions are the linear combinations provided by the Principal Component Analysis or the Linear Discriminant Analysis. When a masking countermeasure is properly implemented to thwart SCAs, the selection of PoIs is known to be a hard task: almost all existing methods have a combinatorial complexity explosion, since they require an exhaustive search among all possible d-tuples of points. In this paper we propose an efficient method for informative feature extraction in presence of masking countermeasure. This method, called Kernel Discriminant Analysis, consists in completing the Linear Discriminant Analysis with a so-called kernel trick, in order to efficiently perform it over the set of all possible d-tuples of points without growing in complexity with d. We identify and analyse the issues related to the application of such a method. Afterwards, its performances are compared to those of the Projection Pursuit (PP) tool for PoI selection up to a 4th-order context. Experiments show that the Kernel Discriminant Analysis remains effective and efficient for high-order attacks, leading to a valuable alternative to the PP in constrained contexts where the increase of the order d does not imply a growth of the profiling datasets.

A modified linear discriminant analysis technique for speaker verification, referred to as normalized discriminant analysis (NDA), is presented. Using this technique it is possible to design an efficient linear classifier with very limited training data and to generate normalized discriminant scores with comparable magnitudes for different classifiers. The NDA technique is applied to a classifier for speaker verification based on speaker specific information obtained when utterances are processed with speaker independent models. In experiments conducted on a network based telephone database, the NDA technique provides an equal-error rate of 6.13% while the classifier using Fisher linear discriminant analysis has an equal-error rate of 18.18%. Furthermore, when the NDA combined with HMM approach in a hybrid speaker verification system, the rate was reduced from 5.30% (HMM with cohort normalization) to 4.32%.

Kernel discriminant analysis (KDA) and the kernel principal component analysis (KPCA), which are the extension of the linear discriminant analysis (LDA) and the principal component analysis (PCA), respectively, from linear domain to nonlinear domain via the kernel trick, are two very popular nonlinear feature extraction methods. In this paper, we present a new feature extraction algorithm by combing KDA and KPCA, and then apply it to the face recognition task. The experimental results on Yale face dataset show that the proposed method can significantly improve the performance both KDA and KPCA.

Speaker and language recognition in noisy and degraded channel conditions continue to be a challenging problem mainly due to the mismatch between clean training and noisy test conditions. In the presence of noise, the most reliable portions of the signal are the high energy regions which can be used for robust feature extraction. In this paper, we propose a front end processing scheme based on autoregressive (AR) models that represent the high energy regions with good accuracy followed by a modulation filtering process. The AR model of the spectrogram is derived using two separable time and frequency AR transforms. The first AR model (temporal AR model) of the sub-band Hilbert envelopes is derived using frequency domain linear prediction (FDLP). This is followed by a spectral AR model applied on the FDLP envelopes. The output 2-D AR model represents a low-pass modulation filtered spectrogram of the speech signal. The band-pass modulation filtered spectrograms can further be derived by dividing two AR models with different model orders (cut-off frequencies). The modulation filtered spectrograms are converted to cepstral coefficients and are used for a speaker recognition task in noisy and reverberant conditions. Various speaker recognition experiments are performed with clean and noisy versions of the NIST-2010 speaker recognition evaluation (SRE) database using the state-of-the-art speaker recognition system. In these experiments, the proposed front-end analysis provides substantial improvements (relative improvements of up to 25%) compared to baseline techniques. Furthermore, we also illustrate the generalizability of the proposed methods using language identification (LID) experiments on highly degraded high-frequency (HF) radio channels and speech recognition experiments on noisy data.

This is the author’s version of a work that was submitted/accepted for pub-lication in the following source:

An automated wood recognition system is designed to classify tropical wood species. The wood features are extracted based on two feature extractors: Basic Grey Level Aura Matrix (BGLAM) technique and statistical properties of pores distribution (SPPD) technique. Due to the nonlinearity of the tropical wood species separation boundaries, a pre classification stage is proposed which consists of Kmeans clustering and kernel discriminant analysis (KDA). Finally, Linear Discriminant Analysis (LDA) classifier and K-Nearest Neighbour (KNN) are implemented for comparison purposes. The study involves comparison of the system with and without pre classification using KNN classifier and LDA classifier. The results show that the inclusion of the pre classification stage has improved the accuracy of both the LDA and KNN classifiers by more than 12%.

Feature extraction VIA kernel weighted discriminant analysis methods by Guang Dai thesis 2007 xi, 90 leaves : ill. ; 30 cm In recent years, as the kernel extension of linear discriminant analysis (LDA), kernel discriminant analysis (KDA) has become one…Read more ›