Search

The most common method used to determine the identity of an individual bird is the capture-mark-recapture technique. The method has several major disadvantages, e.g. some species are difficult to capture/recapture and the capturing process itself may cause significant stress in animals leading even to injuries of more vulnerable species. Some studies introduce systems based on methods used for human identification. An automatic system for recognition of bird individuals (ASRBI) described in this article is based on a Gaussian mixture model (GMM) and a universal background model (GMM-UBM) method extended by an advanced voice activity detection (VAD) algorithm. It is focused on recognizing the bird individuals on an open set, i.e. any number of unknown birds may appear anytime during the identification process as is common in nature. The introduced ASRBI processes the recordings just as if they were recorded by an ornithologist: with durations from seconds to minutes, containing noise and unwanted s...

Probabilistic Linear Discriminant Analysis (PLDA), used particularly in image and speech processing for face and speaker recognition, respectively, is a generative model requesting lots of data to be trained. In the paper several enhancements concerning the implementation of the estimation algorithm of PLDA are proposed providing substantial computational savings. At first, an inverse of a huge matrix is replaced by an inversion of two significantly smaller matrices. Subsequently, it is shown how to avoid the need to process the whole data set in each iteration of the estimation algorithm. Supplementary results are presented on NIST SRE 2008.

An estimation of parameters of a multivariate Gaussian Mixture Model is usually based on a criterion e.g. Maximum Likelihood that is focused mostly on training data. Therefore, testing data, which were not seen during the training procedure, may cause problems. Moreover, numerical instabilities can occur e.g. for low-occupied Gaussians especially when working with full-covariance matrices in high-dimensional spaces. Another question concerns the number of Gaussians to be trained for a specific data set. The approach proposed in this paper can handle all these issues. It is based on an assumption that the training and testing data were generated from the same source distribution. The key part of the approach is to use a criterion based on the source distribution rather than using the training data itself. It is shown how to modify an estimation procedure in order to fit the source distribution better despite the fact that it is unknown, and subsequently new estimation algorithm for diagonal- as well as full-covariance matrices is derived and tested.

Nowadays, Factor analysis based techniques become part of state-of-the-art Speaker Recognition (SR) systems. These are the Joint Factor Analysis, its modified version called the concept of i-vectors, and the Probabilistic Linear Discriminant Analysis (PLDA). PLDA, as a generative statistical model, is usually used as the back end of a SR system, e.g. once i-vectors have been extracted, a PLDA model is used in the i-vector space to provide a verification score of two given i-vectors. In order to train the system huge amount of development data are utilized. In this paper the behaviour of the PLDA model is investigated. It is shown how does the amount of development data influence the system’s performance. PLDA has several parameters to be tuned, i.e. dimensions of latent variables/subspaces, which represent the speaker and the channel variabilities. These will be examined too.

Probabilistic Linear Discriminant Analysis PLDA and the concept of i-vectors are state-of-the-art methods used in the speaker recognition. They are based on Factor Analysis, in which a data covariance matrix is decomposed in order to find a low dimensional representation of given feature vectors. More precisely, the Factor Analysis based methods seek for directions/subspaces in which the projected overall/between/within variance is highest. In order to train models related to individual methods, development speech corpora comprising various acoustic conditions are utilized. The higher are the variations in some of these acoustic conditions, the more will the model tend to reflect them. Strong data variations in some of the development corpora may suppress conditions present in other corpora. This can lead to poor recognition when acoustic variations in test conditions significantly differ. In this paper techniques alleviating such effects are investigated. The idea is to use several background and i-vector models related to different parts of development data so that several i-vectors are extracted, processed and handed over to the PLDA modelling. PLDA model is then used to utilize all the extracted information and provide the verification result.

Single-Gaussian and Gaussian-Mixture Models are utilized in various pattern recognition tasks. The model parameters are estimated usually via Maximum Likelihood Estimation MLE with respect to available training data. However, if only small amount of training data is available, the resulting model will not generalize well. Loosely speaking, classification performance given an unseen test set may be poor. In this paper, we propose a novel estimation technique of the model variances. Once the variances were estimated using MLE, they are multiplied by a scaling factor, which reflects the amount of uncertainty present in the limited sample set. The optimal value of the scaling factor is based on the Kullback-Leibler criterion and on the assumption that the training and test sets are sampled from the same source distribution. In addition, in the case of GMM, the proper number of components can be determined.

The aim of this work is to propose a refinement of the shift-MLLR (shift Maximum Likelihood Linear Regression) adaptation of an acoustics model in the case of limited amount of adaptation data, which can lead to ill-conditioned transformations matrices. We try to suppress the influence of badly estimated transformation parameters utilizing the bottleneck Artificial Neural Network (ANN). The ill-conditioned shift-MLLR transformation is propagated through a bottleneck ANN (suitably trained beforehand), and the output of the net is used as the new refined transformation. To train the ANN the well and the badly conditioned shift-MLLR transformations are used as outputs and inputs of ANN, respectively.

In the paper recent methods used in the task of speaker recognition are presented. At first, the extraction of so called i-vectors from GMM based supervectors is discussed. These i-vectors are of low dimension and lie in a subspace denoted as Total Variability Space (TVS). The focus of the paper is put on Probabilistic Linear Discriminant Analysis (PLDA), which is used as a generative model in the TVS. The influence of development data is analyzed utilizing distinct speech corpora. It is shown that it is preferable to cluster available speech corpora to classes, train one PLDA model for each class and fuse the results at the end. Experiments are presented on NIST Speaker Recognition Evaluation (SRE) 2008 and NIST SRE 2010.

The worst problem the adaptation is dealing with is the lack of adaptation data. This work focuses on the feature Maximum Likelihood Linear Regression (fMLLR) adaptation where the number of free parameters to be estimated significantly decreases in comparison with other adaptation methods. However, the number of free parameters of fMLLR transform is still too high to be estimated properly when dealing with extremely small data sets. We described and compared various methods used to avoid this problem, namely the initialization of the fMLLR transform and a linear combination of basis matrices varying in the choice of the basis estimation (eigen decomposition, factor analysis, independent component analysis and maximum likelihood estimation). Initialization methods compensate the absence of the test speaker’s data utilizing other suitable data. Methods using linear combination of basis matrices reduce the number of estimated fMLLR parameters to a smaller number of weights to be estimated. Experiments are aimed to compare results of proposed basis and initialization methods.

Článek popisuje řešení problému malého mknožství dat pro fMLLR adaptaci. Navrhuje inicializaci statistik z dat od nejbližších řečníků s využitím informace o fonetické příslušnosti. In this work we deal with the problem of small amount of data when estimating a feature transformation for the speaker adaptation of an acoustic model. Our goal is to compensate for the lack of adaptation data by a proper initialization of transformation matrices. Methods used in such situations are described, they are based on collecting additional accumulated statistics from nearest speakers. The proposed initialization approach is based on accumulated statistics too, but it incorporates also phonetic information when selecting the “nearest” statistics. Initialization methods compensating for the absence of actual speaker’s data are tested on telephone recordings with different amounts of adaptation data. In worst situation with extremely small amount of adaptation data relative improvement of 5% is obtained.

In this paper recent methods used in the task of Speaker Recognition (SR) are reviewed and their complementarity is analysed. At first, methods based on Supervectors (SVs) related to Gaussian Mixture Models (GMMs) and Support Vector Machines (SVMs) used as a discriminative model are described along with the Nuisance Attribute Projection (NAP). NAP was proposed to suppress undesirable influences of high channel variabilities between several sessions of a speaker. Next, recent methods focusing on the extraction of so called i-vectors (low dimensional representations of GMM based SVs) are discussed. The space in which i-vectors lie is denoted the Total Variability Space (TVS) since it contains both between-speaker and session/channel variabilities. Once i-vectors have been extracted a Probabilistic Linear Discriminant Analysis (PLDA) model is trained in the TVS. In the training phase of PLDA the TVS is decomposed to a channel and a speaker subspace, hence each i-vector is supposed to be composed from a speaker identity component and a channel component. The complementarity of PLDA and SVM based modelling techniques is examined utilizing the linear logistic regression as a fusion tool used to combine the verification scores of individual systems leading to significant reductions in error rates of the SR system. The results are presented on the NIST SRE 2008 and NIST SRE 2010 corpora.

Gaussian Mixture Models (GMMs) are widely used among scientists e.g. in statistics toolkits and data mining procedures. In order to estimate parameters of a GMM the Maximum Likelihood (ML) training is often utilized, more precisely the Expectation-Maximization (EM) algorithm. Nowadays, a lot of tasks works with huge datasets, what makes the estimation process time consuming (mainly for complex mixture models containing hundreds of components). The paper presents an efficient and robust implementation of the estimation of GMM statistics used in the EM algorithm on GPU using NVIDIA's Compute Unified Device Architecture (CUDA). Also an augmentation of the standard CPU version is proposed utilizing SSE instructions. Time consumptions of presented methods are tested on a large dataset of real speech data from the NIST Speaker Recognition Evaluation (SRE) 2008. Estimation on GPU proves to be more than 400 times faster than the standard CPU version and 130 times faster than the SSE version, thus a huge speed up was achieved without any approximations made in the estimation formulas. Proposed implementation was also compared to other implementations developed by other departments over the world and proved to be the fastest (at least 5 times faster than the best implementation published recently).

One of the most utilized adaptation techniques is the feature Maximum Likelihood Linear Regression (fMLLR). In comparison with other adaptation methods the number of free parameters to be estimated significantly decreases. Thus, the method is well suited for situations with small amount of adaptation data. However, fMLLR still fails in situations with extremely small data sets. Such situations can be solved through proper initialization of fMLLR estimation adding some a-priori information. In this paper a novel approach is proposed solving the problem of fMLLR initialization involving statistics from speakers acoustically close to the speaker to be adapted. Proposed initialization suitably substitutes missing adaptation data with similar data from a training database, fMLLR estimation becomes well-conditioned, and the accuracy of the recognition system increases even in situations with extremely small data sets.

This paper deals with robust estimations of data statistics used for the adaptation. The statistics are accumulated before the adaptation process from available adaptation data. In general, only small amount of adaptation data is assumed. These data are often corrupted by noise, channel, they do not contain only clean speech. Also, when training Hidden Markov Models (HMM) several assumptions are made that could not have been fulfilled in the praxis, etc. Therefore, we described several techniques that aim to make the adaptation as robust as possible in order to increase the accuracy of the adapted system. One of the methods consists in initialization of the adaptation statistics in order to prevent ill-conditioned transformation matrices. Another problem arises when an acoustic feature is assigned to an improper HMM state even if the reference transcription is available. Such situations can occur because of the forced alignment process used to align frames to states. Thus, it is quite handy to accumulate data statistic utilizing only reliable frames (in the sense of data likelihood). We are focusing on Maximum Likelihood Linear Transformations and the experiments were performed utilizing the feature Maximum Likelihood Linear Regression (fMLLR). Experiments are aimed to describe the behavior of the system extended by proposed methods.

This paper evaluates the performance of face and speaker verification techniques in the context of a mobile environment. The mobile environment was chosen as it provides a realistic and challenging test-bed for biometric person verification techniques to operate. For instance the audio environment is quite noisy and there is limited control over the illumination conditions and the pose of the subject for the video. To conduct this evaluation, a part of a database captured during the “Mobile Biometry” (MOBIO) European Project was used. In total there were nine participants to the evaluation who submitted a face verification system and five participants who submitted speaker verification systems. The results have shown that the best performing face and speaker verification systems obtained the same level of performance, respectively 10.9% and 10.6% of HTER., This work has been performed by the MOBIO project 7th Framework Research Programme of the European Union (EU), grant agreement number: 214324. The authors would like to thank the EU for the financial support and the partners within the consortium for a fruitful collaboration. For more information about the MOBIO consortium please visit http://www.mobioproject.org. NICTA is funded by the Australian Government as represented by the Department of Broadband, Communications and the Digital Economy as well as the Australian Research Council through the ICT Centre of Excellence program.

This paper deals with a combination of basic adaptation techniques of Hidden Markov Model used in the speech recognition. The adaptation methods approach the data only through their statistics, which have to be accumulated before the adaptation process. When performing two adaptations subsequently, the data statistics have to be accumulated twice in each of the adaptation passes. However, when the adaptation methods are chosen with care, the data statistics may be accumulated only once, as proposed in this paper. This significantly reduces the time consumption and avoids the need to store all the adaptation data. Combination of Maximum A-Posteriori Probability and feature Maximum Likelihood Linear Regression adaptation is considered. Motivation for such an approach could be the on-line adaptation, where the time consumption is of big importance.

This paper investigates the combination of discriminative adaptation techniques. The discriminative Maximum A-Posteriori (DMAP) adaptation and discriminative feature Maximum Likelihood Linear Regression (DfMLLR) are examined. Since each of the methods is proposed for distinct amount of adaptation data it is useful to combine them in order to preserve the systems performance in situations with varying amount of adaptation data. Generally, DfMLLR and DMAP are executed subsequently (DMAP preceded by DfMLLR) demanding to approach the data twice. Since both methods address the data through the same statistics an one-pass-combination was proposed in order to decrease the time consumption. The one-pass-combination utilizes the advantage of DfMLLR method to transform directly the feature vectors. However, instead of feature vectors the statistics are transformed, what allows to use already computed statistics for the DMAP pass without the need to process the data once again. All the approaches are compared also to their non-discriminative alternatives.

The article introduces an expert system for the speaker verification task. Our main purpose was to design a tool for the combination of various speaker verification systems proposed for various operating conditions. First of all, the essential ideas are explained that made us design the expert system. Next section describes the structure of a rule-based expert system and subsequently an oriented graph is proposed for the representation of the topology of the system. The expert rules exploited by the system are derived automatically from the input data and we have implemented also a certainty factor to acquire more reliable decisions. The experiments show that the proposed system has the capability to significantly improve the verification results in trials with various operating conditions.

In the paper Factor Analysis (FA) and Nuisance Attribute Projection (NAP) are reviewed, analyzed and compared. Since nowadays FA become a part of most state-of-the-art recognition systems (used e.g. in the concept of i-vectors or PLDA models) it is of relevance to examine different insights into the problem. NAP was chosen as a counterpart to FA as an advanced PCA like method often utilized in speaker recognition systems along with FA. It is demonstrated how can be both FA and NAP expressed as solutions of Least Squares (LS), the consequences of the LS formulation are discussed, and it is shown in what extent do solutions of NAP and FA overlap.