Speaker Verification

Recently satisfactory results have been obtained in NIST speaker recognition evaluations. These results are mainly due to accurate modeling of a very large development dataset provided by LDC. However, for many realistic scenarios the use of this development dataset is limited due to a dataset mismatch. In such cases, collection of a large enough dataset is infeasible. In this work we analyze the sources of degradation for a particular setup in the context of an i-vector PLDA system and conclude that the main source for degradation is an i-vector dataset shift. As a remedy, we introduce inter dataset variability compensation (IDVC) to explicitly compensate for dataset shift in the i-vector space. This is done using the nuisance attribute projection (NAP) method. Using IDVC we managed to reduce error dramatically by more than 50% for the domain mismatch setup

— An automatic verification of person's identity from its voice is a part of modern telecommunication services. In order to execute a verification task, a speech signal has to be transmitted to a remote server. So, a performance of the verification system can be influenced by various distortions that can occur when transmitting a speech signal through a communication channel. This paper studies an effect of the state of art wideband (WB) speech codecs on a performance of automatic speaker verification in the context of a channel/codec mismatch between enrollment and test utterances. The speaker verification system is developed on GMM-UBM method. The results show that EVS codec provides the best performance over all the investigated scenarios in this study. Moreover, deploying G.729.1 codec in a training process of the verification system provides the best equal error rate in the fully-codec mismatched scenario. Anyhow, differences between the equal error rates reported for all of the codecs involved in this scenario are mostly nonsignificant.

One characteristic that distinguishes speaker recognition (identification, verification, classification, tracking, etc.) from other biometrics is that it is designed to operate with devices and over channels that were created for other technologies and functions. That characteristic supports broad, inexpensive, and speedy deployments. The explosion of mobile devices has exacerbated the mismatch problem and the challenges for interoperability. This paper presents a detailed proposal for interoperability that supports all types of audio interchange operations while, at the same time, limiting the audio formats to a small set of widely-used, open standards. We call this proposal Standard Audio Format Encapsulation (SAFE). The SAFE proposal has been incorporated into speaker-recognition data interchange draft standards by the M1 (biometrics) committee of ANSI/INCITS and ISO/IEC JTC1/SC37 project 19794-13 (Voice data).

This article deals with a technique of voice forgery using the ALISP (Automatic Language Independent Speech Processing) approach. Such a technique allows to transform the voice of an arbitrary person (the impostor), forging the identity of another person (the client). Our goal is to demonstrate that an automatic speaker recognition system could be seriously threatened by a transformation of this kind. For this purpose, we use a speaker verification system to calculate the likelihood that the forged voice belongs to the genuine client. Experiments on NIST 2004 evaluation data show that the equal error rate for the verification task is significantly increased by our voice transformation.

Device, language and environmental mismatch adversely affect speaker verification (SV) performance. We investigate such effects empirically based on the M3 (multibiometric, multilingual and multi-device) Corpus [1]. Device mismatch (among 3G phone, PocketPC and a desktop PC plug-in microphone) brings relative performance degradation of 523%; language mismatch (between English and Cantonese) brings 284% and environmental mismatch (between office environment and recording studio) brings 109%. In particular, verification with wide-band models on narrow-band test data outperforms narrow-band models on wide-band test data. The 3G phone's SV performance is generally low, but remains stable across environments. Additionally, durational variations within two-second utterances may cause a relative change of 633% in SV performance.

Authentication is the process whereby a user proves his claim to identity. This paper aims to review existing MMBAS and multimodal biometric datasets commonly used for testing and benchmarking results. At the same time, a brief overview of fusion strategy and techniques will be discussed.

The main objectives of this work are to describe the online bus pass generation and ticket booking using QR code. Online bus pass generation is helpful to people who are suffering issues with the present technique for the generation of bus pass and renewal. This project consists of two login pages, one for user registration and the other one for admin. Users need to register by submitting their details through online. Once the registration process is done then a security code called One Time Password (OTP) code will be sent to the user's registered mail. This system is used for ticket generation, bus pass formation and renewing of the bus pass of the users. The user can login with Idno and password to perform the pass booking and renewal. Bus Ticket Checker can scan the users QR code to check the validity of bus pass.

Hidden Markov Models based text-to-speech (HMM-TTS) synthesis is a technique for generating speech from trained statistical models where spectrum, pitch and durations of basic speech units are modelled altogether. The aim of this work is to describe a Spanish HMM-TTS system using CBR as a F0 estimator, analysing its performance objectively and subjectively. The experiments have been conducted on a reliable labelled speech corpus, whose units have been clustered using contextual factors according to the Spanish language. The results show that the CBR-based F0 estimation is capable of improving the HMM-based baseline performance when synthesizing nondeclarative short sentences and reduced contextual information is available.

The QUT-NOISE-SRE protocol is designed to mix the large QUT-NOISE database, consisting of over 10 hours of background noise, collected across 10 unique locations covering 5 common noise scenarios, with commonly used speaker recognition datasets such as Switchboard, Mixer and the speaker recognition evaluation (SRE) datasets provided by NIST. By allowing common, clean, speech corpora to be mixed with a wide variety of noise conditions, environmental reverberant responses, and signal-to-noise ratios, this protocol provides a solid basis for the development, evaluation and benchmarking of robust speaker recognition algorithms, and is freely available to down-load alongside the QUT-NOISE database. In this work, we use the QUT-NOISE-SRE protocol to evaluate a state-of-the-art PLDA i-vector speaker recognition system, demonstrating the importance of designing voice-activity-detection front-ends specifically for speaker recognition, rather than aiming for perfect coherence with the true speech/non-speech boundaries.

This paper presents new a feature transformation technique applied to improve the screening accuracy for the automatic detection of pathological voices. The statistical transformation is based on Hidden Markov Models, obtaining a transformation and classification stage simultaneously and adjusting the parameters of the model with a criterion that minimizes the classification error. The original feature vectors are built up using classic short-term noise parameters and mel-frequency cepstral coefficients. With respect to conventional approaches found in the literature of automatic detection of pathological voices, the proposed feature space transformation technique demonstrates a significant improvement of the performance with no addition of new features to the original input space. In view of the results, it is expected that this technique could provide good results in other areas such as speaker verification and/or identification.

In this paper we propose a new feature extraction algorithm based on nonlinear prediction: the Neural Predictive Coding model which is an extension of the classical LPC one. This model is applied to speaker verification by the Arithmetic-Harmonic Sphericity (AHS) method. Two different initialization methods are proposed for the coding method based on the Neural Predictive Coding (NPC): classical neural networks initialization and linear initialization. The first model obtains smaller rates. For the linear initialization, we obtain significant improvement in comparison to the most used methods (LPCC, MFCC). This study opens a new way towards different feature extraction schemes that offers better accuracy on speaker recognition tasks.

Learning good representations is of crucial importance in deep learning. Mutual Information (MI) or similar measures of statistical dependence are promising tools for learning these representations in an unsupervised way. Even though the mutual information between two random variables is hard to measure directly in high dimensional spaces, some recent studies have shown that an implicit optimization of MI can be achieved with an encoder-discriminator architecture similar to that of Genera-tive Adversarial Networks (GANs). In this work, we learn representations that capture speaker identities by maximizing the mutual information between the encoded representations of chunks of speech randomly sampled from the same sentence. The proposed encoder relies on the SincNet architecture and transforms raw speech waveform into a compact feature vector. The discriminator is fed by either positive samples (of the joint distribution of encoded chunks) or negative samples (from the product of the marginals) and is trained to separate them. We report experiments showing that this approach effectively learns useful speaker representations, leading to promising results on speaker identification and verification tasks. Our experiments consider both unsupervised and semi-supervised settings and compare the performance achieved with different objective functions.

In this paper, automatic speaker recognition system is implemented by combining feature extraction and feature matching technique. Feature extraction method that is implemented by the Mel Frequency Cepstral Coefficients (MFCC). The Vector Quantization (VQ) is done using k-means algorithm that is used for the feature matching technique due to high accuracy but its simplicity. A speaker recognition experiments was performed using a 5talkers (3 male and 2 female) with different conditions of Bangla sentence utterances. So, in this proposal achieved 96% efficiency for text dependent or independent speaker recognition system with 60 -90 seconds duration test voice sample.

This paper proposes a joint verification-localization structure based on split-band analysis of speech signal and the mixed voicing level. To address the problems in reverberant acoustic environments, a new fundamental frequency estimation algorithm is proposed based on high resolution spectral estimation. In the reconstruction of the distorted speech this information is utilized to reduce the side effect of acoustic noise on the voicing parts. A speaker verification system examines the features of the reconstructed speech in order to authorize the speaker before localization. This procedure prevents localization and beamforming for non-speech and specially the unwanted speakers in multi-speaker scenarios. The verification is implemented with the Gaussian Mixture Model and a new filtering scheme is proposed based on the voicing likelihood of each frequency band measured in the previous steps for efficient localization of the authorized speaker. The performance of the proposed VSL (Verified Speaker Localization) front-end is evaluated in various reverberant and noisy environments. The VSL is utilized in the development of distant-talking automatic speech recognition by microphone array where the system can lock on a specific source and hence the recognition quality improves noticeably.

Clustering is needed in various applications such as biometric person authentication, speech coding and recognition, image compression and information retrieval. Hundreds of clustering methods have been proposed for the task in various fields but, surprisingly, there are few extensive studies actually comparing them. An important question is how much the choice of a clustering method matters for the final pattern recognition application. Our goal is to provide a thorough experimental comparison of clustering methods for text-independent speaker verification. We consider parametric Gaussian mixture model (GMM) and non-parametric vector quantization (VQ) model using the best known clustering algorithms including iterative (K-means, random swap, expectation-maximization), hierarchical (pairwise nearest neighbor, split, split-and-merge), evolutionary (genetic algorithm), neural (self-organizing map) and fuzzy (fuzzy C-means) approaches. We study recognition accuracy, processing time, clustering validity, and correlation of clustering quality and recognition accuracy. Experiments from these complementary observations indicate clustering is not a critical task in speaker recognition and the choice of the algorithm should be based on computational complexity and simplicity of the implementation. This is mainly because of three reasons: the data is not clustered, large models are used and only the best algorithms are considered. For low-order models, choice of the algorithm, however, can have a significant effect.

Because of the differences in education background, accents, etc., different persons have their unique way of pronunciation. This paper exploits the pronunciation characteristics of speakers and proposes a new conditional pronunciation modeling (CPM) technique for speaker verification. The proposed technique aims to establish a link between articulatory properties (e.g., manners and places of articulation) and phoneme sequences produced by a speaker. This is achieved by aligning two articulatory feature (AF) streams with a phoneme sequence determined by a phoneme recognizer, and formulating the probabilities of articulatory classes conditioned on the phonemes as speaker-dependent probabilistic models. The scores obtained from the AF-based pronunciation models are then fused with those obtained from a spectral-based speaker verification system, with the frame-by-frame fused scores weighted by the confidence of the pronunciation models. Evaluations based on the SPIDRE corpus demonstrate that AF-based CPM systems can recognize speakers even with short utterances and are readily combined with spectral-based systems to further enhance the reliability of speaker verification.

Nowadays state-of-the-art speaker recognition systems obtain quite accurate results for both text-independent and text-dependent tasks as long as they are trained on a fair amount of development data from the target domain (assuming clean speech). In this work, we address the challenge of building a speaker recognition system with a small development dataset from the target domain without using out-of-domain data whatsoever. When development data is limited, the Nuisance Attribute Projector (NAP) algorithm is (in general) superior to the i-vector approach. We have investigated the relative degradation observed from the different components of the  NAP system trained on a small dataset and conclude that score normalization is a major source of degradation. We introduce a novel method for stabilizing the normalized scores. We explicitly estimate a low dimensional subspace in supervector space which accounts for high variability in score normalization parameters. We then compensate the estimated subspace. We report experiments on both text-dependent and text-independent tasks which validate our method and show large error reductions

This paper investigates the effects of limited speech data in the context of speaker verification using a probabilistic linear dis-criminant analysis (PLDA) approach. Being able to reduce the length of required speech data is important to the development of automatic speaker verification system in real world applications. When sufficient speech is available, previous research has shown that heavy-tailed PLDA (HTPLDA) modeling of speakers in the i-vector space provides state-of-the-art performance, however, the robustness of HTPLDA to the limited speech resources in development, enrolment and verification is an important issue that has not yet been investigated. In this paper , we analyze the speaker verification performance with regards to the duration of utterances used for both speaker evaluation (enrolment and verification) and score normalization and PLDA modeling during development. Two different approaches to total-variability representation are analyzed within the PLDA approach to show improved performance in short-utterance mis-matched evaluation conditions and conditions for which insufficient speech resources are available for adequate system development. The results presented within this paper using the NIST 2008 Speaker Recognition Evaluation dataset suggest that the HT-PLDA system can continue to achieve better performance than Gaussian PLDA (GPLDA) as evaluation utterance lengths are decreased. We also highlight the importance of matching durations for score normalization and PLDA modeling to the expected evaluation conditions. Finally, we found that a pooled total-variability approach to PLDA modeling can achieve better performance than the traditional concatenated total-variability approach for short utterances in mismatched evaluation conditions and conditions for which insufficient speech resources are available for adequate system development.

This paper proposes the addition of a weighted median Fisher discriminator (WMFD) projection prior to length-normalised Gaussian probabilistic linear discriminant analysis (GPLDA) modelling in order to compensate the additional session variation. In limited microphone data conditions, a linear-weighted approach is introduced to increase the influence of microphone speech dataset. The linear-weighted WMFD-projected GPLDA system shows improvements in EER and DCF values over the pooled LDA-and WMFD-projected GPLDA systems in interview interview condition as WMFD projection extracts more speaker discriminant information with limited number of ses-sions/ speaker data, and linear-weighted GPLDA approach estimates reliable model parameters with limited microphone data.

Speaker verification might be considered a binary classification problem in that the objective is to determine whether or not an utterance is from the individual whose identity is claimed. Several factors make speaker verification different from a standard binary problem. Speaker verification is challenging because of the open nature of the problem, if the utterances of an individual are examples of the class to be recognised then the non- class examples cover everything else. It is also challenging due to the the format of the data to be classified, the data consists of sentences whose lengths depend on its phonetic content and the speaking rate of the underlying speaker.
One class classifiers have emerged as a set of techniques for situations where labelled data exists for only one of the classes in a two-class problem. A related problem arises where non-class examples exist, but the non-class distribution cannot be characterised as in speaker verification. The approach taken by one class classifiers is to develop a classifier that characterises the target class, and thus can distinguish it from all counter-examples.
Traditional speaker verification systems relied on one class classifiers in order to make a decision on the validity of the claim. A popular approach used Gaussian mixture models to create a score for variable length utterances which could then be thresholded. More recently these underlying one class classifiers have been successfully used to either project variable length utterances into a fixed dimensional space or provide a characterisation which can be compared against. This thesis investigates the use of one class classifiers in speaker verification, first by casting the problem as a one class problem and then by using one class classifiers to pre-process variable length utterances so that they can be used by any standard binary classifier.
This thesis has found that by using one class classifiers not in their traditional setting but as tool to model or characterise utterances they can be harnessed to enable binary learners to perform discriminative learning of variable length utterances.

This paper introduces Locally Recurrent Probabilistic Neural Networks (LRPNN) as an extension of the well-known Probabilistic Neural Networks (PNN). A LRPNN, in contrast to a PNN, is sensitive to the context in which events occur, and therefore, identification of time or spatial correlations is attainable. Besides the definition of the LRPNN architecture a fast three-step training method is proposed. The first two steps are identical to the training of traditional PNNs, while the third step is based on the Differential Evolution optimization method. Finally, the superiority of LRPNNs over PNNs on the task of text-independent speaker verification is demonstrated.

Speaker recognition is the process of automatically recognizing who is speaking on the basis of individual information included in speech waves. Speaker recognition is basically divided into two-classification: speaker verification and speaker identification .It is the method of automatically identify who is speaking on the basis of individual information integrated in speech waves. The speaker identification is the process of determining which registered speaker provides a given speech. On the other hand, speaker verification is the process of rejecting or accepting the identity claim of a speaker.Speaker identificatiton has mainly two phases.Feature extraction and Feature matching.For feature extraction,MFCC method is used for generating the acoustic vectors.For feature matching,HMM and SVM ,two classifiers and analyzed the aacuracy of the above two classifiers under various noise levels.

This paper investigates advanced channel compensation techniques for the purpose of improving i-vector speaker verification performance in the presence of high intersession variability using the NIST 2008 and 2010 SRE corpora. The performance of four channel compensation techniques: (a) weighted maximum margin criterion (WMMC), (b) source-normalized WMMC (SN-WMMC), (c) weighted linear discriminant analysis (WLDA), and; (d) source-normalized WLDA (SN-WLDA) have been investigated.

We show that, by extracting the discriminatory information between pairs of speakers as well as capturing the source variation information in the development i-vector space, the SN-WLDA based cosine similarity scoring (CSS) i-vector system is shown to provide over 20% improvement in EER for NIST 2008 interview and microphone verification and over 10% improvement in EER for NIST 2008 telephone verification, when compared to SN-LDA based CSS i-vector system. Further, score-level fusion techniques are analyzed to combine the best channel compensation approaches, to provide over 8% improvement in DCF over the best single approach, (SN-WLDA), for NIST 2008 interview/ telephone enrolment-verification condition. Finally, we demonstrate that the improvements found in the context of CSS also generalize to state-of-the-art GPLDA with up to 14% relative improvement in EER for NIST SRE 2010 interview and microphone verification and over 7% relative improvement in EER for NIST SRE 2010 telephone verification.

Substantial progress has been achieved in voice-based biometrics in recent times but a variety of challenges still remain for speech research community. One such obstacle is reliable speaker authentication from speech signals degraded by lossy compression. Compression is commonplace in modern telecommunications, such as mobile telephony, VoIP services, teleconference, voice messaging or gaming. In this study, the authors investigate the effect of lossy speech compression on text-independent speaker verification. Voice biometrics performance is evaluated on clean speech signals distorted by the state-of-the-art narrowband (NB) as well as wideband (WB) speech codecs. The tests are performed in both channel-matched and channel-mismatched scenarios. The test results show that coded WB speech improves voice authentication precision by 1–3% of equal error rate over coded NB speech, even at the lowest investigated bitrates. It is also shown that the enhanced voice services codec does not provide better results than the other codecs involved in this study.

In the past few years, discriminative approaches to perform speaker detection have shown good results and an increasing interest. Among these methods, SVM based systems have lots of advantages, especially their ability to deal with a high dimension feature space. Generative systems such as UBM-GMM systems show the greatest performance among other systems in speaker verification tasks. Combination of generative and discriminative approaches is not a new idea and has been studied several times by mapping a whole speech utterance onto a fixed length vector. This paper presents a straight-forward, cost friendly method to combine the two approaches with the use of a UBM model only to drive the experiment. We show that the use of the TFLLR kernel, while closely related to a reduced form of the Fisher mapping, implies a performance that is close to a standard GMM/UBM based speaker detection system. Moreover, we show that a combination of both outperforms the systems taken independently.

The Internet provides a convenient platform for cyber criminals to anonymously conduct their illegitimate activities, such as phishing and spamming. As a result, in recent years, authorship analysis of anonymous e-mails has received some attention in the cyber forensic and data mining communities. In this paper, we study the problem of authorship verification: given a set of e-mails written by a suspect along with an e-mail dataset collected from the sample population, we want to determine whether or not an anonymous e-mail is written by the suspect. To address the problem of authorship verification of textual documents and employ detection measures that are more suited in the context of forensic investigation, we borrow the NIST's speaker recognition evaluation (SRE) framework. Our experimental results on real world e-mail dataset suggest that the employed framework addresses the e-mail authorship verification problem with a matching success as in case of speaker verification. The proposed framework produces an average equal error rate of 15-20% and minDCF equal to 0.0671 (with 10-fold cross validation technique) in correctly verifying the author of a malicious e-mail.

The automatic identification of person’s identity from their voice is a part of modern telecommunication services. In order to execute the identification task, speech signal has to be transmitted to a remote server. So a performance of the recognition/identification system can be influenced by various distortions that occur when transmitting speech signal through a communication channel. This paper studies an effect of telecommunication channel, particularly commonly used narrowband (NB) speech codecs in current telecommunication networks, on a performance of automatic speaker recognition in the context of a channel/codec mismatch between enrollment and test utterances. An influence of speech coding on speaker identification is assessed by using the reference GMM-UBM method. The results show that the partially mismatched scenario offers better results than the fully matched scenario when speaker recognition is done on speech utterances degraded by the different NB codecs. Moreover, deploying EVS and G.711 codecs in a training process of the recognition system provides the best success rate in the fully mismatched scenario. It should be noted here that the both EVS and G.711codecs offer the best speech quality among the codecs deployed in this study. This finding also fully corresponds with the finding presented by Janicki & Staroszczyk in [1] focusing on other speech codecs.

Probabilistic linear discriminant analysis (PLDA) has shown to be effective for modeling channel variability in the i-vector space for text-independent speaker verification. Speaker verification is a binary hypothesis testing. Given a test segment, the verification score could be computed as the log-likelihood ratio between a speaker-adapted PLDA and the universal PLDA model. This work proposes to infer the channel factor specific to each test segment and to include the channel estimate in the PLDA models, which essentially shifts the scoring function to better match that of the test channel. We also explore the influence of covariance adaptation in both speaker and channel adaptations. Experimental results on NIST SRE'08 and SRE'10 dataset confirm that the proposed channel adaptation can be effective when the covariance is kept un-adapted, while the covariance adaptation is necessary in the speaker adaptation.

A method is described for designing speaker recognition features that are robust to telephone handset distortion. The approach transforms features such as mel-cepstral features, log spectrum, and prosody-based features with a non-linear arti®cial neural network. The neural network is discriminatively trained to maximize speaker recognition performance speci®cally in the setting of telephone handset mismatch between training and testing. The algorithm requires neither stereo recordings of speech during training nor manual labeling of handset types either in training or testing. Results on the 1998 National Institute of Standards and Technology (NIST) Speaker Recognition Evaluation corpus show relative improvements as high as 28% for the new multilayered perceptron (MLP)-based features as compared to a standard mel-cepstral feature set with cepstral mean subtraction (CMS) and handset-dependent normalizing impostor models. Ó : S 0 1 6 7 -6 3 9 3 ( 9 9 ) 0 0 0 7 7 -1 syst eme de reconnaissance du locuteur, sp eci®quement dans des conditions o u des types de t el ephone di erents sont utilis es lors de l'entra õnement et de la v eri®cation. L'algorithme ne requiert, ni enregistrement st er eo de la session d'entra õnement, ni etiquettage manuel des types de t el ephone utilis es a l'entra õnement et a la v eri®cation. Les r esultats sur le corpus 1998 NIST Speaker Recognition Evaluation montrent une am elioration relative atteignant 28% avec les nouvelles caract eristiques bas ees sur le r eseau de neurones. Le syst eme de r ef erence utilise des vecteurs de caract eristiques bas es le MEL-cepstre avec soustraction du cepstre moyen ainsi que des mod eles d'imposteurs d ependant du type de t el ephone. Ó

This paper presents the QUT speaker recognition system, as a competing system in the Speakers In The Wild (SITW) speaker recognition challenge. Our proposed system achieved an overall ranking of second place, in the main core-core condition evaluations of the SITW challenge. This system uses an i-vector/PLDA approach, with domain adaptation and a deep neu-ral network (DNN) trained to provide feature statistics. The statistics are accumulated by using class posteriors from the DNN, in place of GMM component posteriors in a typical GMM-UBM i-vector/PLDA system. Once the statistics have been collected, the i-vector computation is carried out as in a GMM-UBM based system. We apply domain adaptation to the extracted i-vectors to ensure robustness against dataset variability , PLDA modelling is used to capture speaker and session variability in the i-vector space, and the processed i-vectors are compared using the batch likelihood ratio. The final scores are calibrated to obtain the calibrated likelihood scores, which are then used to carry out speaker recognition and evaluate the performance of the system. Finally, we explore the practical application of our system to the core-multi condition recordings of the SITW data and propose a technique for speaker recognition in recordings with multiple speakers.

The SRI speaker recognition system for the 2008 NIST speaker recognition evaluation (SRE) incorporates a variety of models and features, both cepstral and stylistic. We highlight the improvements made to specific subsystems and analyze the performance of various subsystem combinations in different data conditions. We show the importance of language and nativeness conditioning, as well as the role of ASR for speaker verification.

Focused on the issue that the robustness of traditional Mel Frequency Cepstral Coefficient (MFCC) feature degrades drastically in speaker verification in noisy environments, a kind of suitable extraction method for low SNR environments based on Gaussian Mixture Model-Universal Background Model (GMM-UBM) and improved Power Normalized Cepstral Coefficient (PNCC) is proposed. First, the PNCC feature is extracted after the Voice Activity Detection (VAD), which uses long term analysis to remove the effect of background noise. Then, Cepstral Mean Variance Normalization (CMVN), Feature Warping and other methods are used to improve PNCC. Finally, GMM-UBM-MAP is set as the baseline system for speaker verification test with TIMIT speech database, the robustness of four different features (MFCC, GFCC, PNCC and improved PNCC) are analyzed and compared in different noisy conditions. The experimental results indicate that MFCC has achieved the highest recognition rate under the environment of clean speech. By mixing the test speech with sine noises, the improved PNCC is more robust against different low-SNR noises than other original features and its Equal Error Rate (EER) reduce significantly in low-SNR noise environments.

In the last few years, the use of i-vectors along with a generative back-end has become the new standard in speaker recognition. An i-vector is a compact representation of a speaker utterance extracted from a low dimensional total variability subspace. Although current speaker recognition systems achieve very good results in clean training and test conditions, the performance degrades considerably in noisy environments. The compensation of the noise effect is actually a research subject of major importance. As far as we know, there was no serious attempt to treat the noise problem directly in the i-vectors space without relying on data distributions computed on a prior domain. This paper proposes a full-covariance Gaussian modeling of the clean i-vectors and noise distributions in the i-vectors space then introduces a technique to estimate a clean i-vector given the noisy version and the noise density function using MAP approach. Based on NIST data, we show that it is possible to improve up to 60% the baseline system performances. A noise adding tool is used to help simulate a real-world noisy environment at different signal-to-noise ratio levels.

In this paper we describe the major elements of MIT Lincoln Laboratory's Gaussian mixture model (GMM)-based speaker verification system used successfully in several NIST Speaker Recognition Evaluations (SREs). The system is built around the likelihood ratio test for verification, using simple but effective GMMs for likelihood functions, a universal background model (UBM) for alternative speaker representation, and a form of Bayesian adaptation to derive speaker models from the UBM. The development and use of a handset detector and score normalization to greatly improve verification performance is also described and discussed. Finally, representative performance benchmarks and system behavior experiments on NIST SRE corpora are presented.

In speech and audio applications, short-term signal spectrum is often represented using mel-frequency cepstral coefficients (MFCCs) computed from a windowed discrete Fourier transform (DFT). Windowing reduces spectral leakage but variance of the spectrum estimate remains high. An elegant extension to windowed DFT is the so-called multitaper method which uses multiple time-domain windows (tapers) with frequencydomain averaging. Multitapers have received little attention in speech processing even though they produce low-variance features. In this paper, we propose the multitaper method for MFCC extraction with a practical focus. We provide, firstly, detailed statistical analysis of MFCC bias and variance using autoregressive process simulations on the TIMIT corpus.

Recently we have investigated the use of state-of-the-art text-dependent speaker verification algorithms for user authentication and obtained satisfactory results mainly by using a fair amount of text-dependent development data from the target domain. In this work we investigate the ability to build high accuracy text-dependent systems using no data at all from the target domain. Instead of using target domain data, we use resources such as TIMIT, Switchboard, and NIST data. We introduce several techniques addressing both lexical mismatch and channel mismatch. These techniques include synthesizing a universal background model according to lexical content, automatic filtering of irrelevant phonetic content, exploiting information in residual supervectors (usually discarded in the i-vector framework), and inter dataset variability modeling. These techniques reduce verification error significantly, and also improve accuracy when target domain data is available

Speaker verification is a challenging problem in speaker recognition where the objective is to determine whether a segment of speech in fact comes from a specific individual. In supervised machine learning terms this is a challenging problem as, while examples belonging to the target class are easy to gather, the set of counter-examples is completely open. In this paper we cast this as a one-class classification problem and evaluate a variety of state-of-the-art one-class classification techniques on a benchmark speech recognition dataset. We show that of the one-class classification techniques, Gaussian Mixture Models shows the best performance on this task.

In this paper we describe a system we have developed for automatic broadcast-quality video indexing that successfully combines results from the fields of speaker verification, acoustic analysis, very large vocabulary speech recognition, content based sampling of video, information retrieval, natural language processing, dialogue systems, and MPEG2 delivery over IP. Our audio classification and anchorperson detection (in the case of news material) classifies video into news versus commercials using acoustic features and can reach 97% accuracy on our test data set. The processing includes very large vocabulary speech recognition (over 230K-word vocabulary) for synchronizing the closed caption stream with the audio stream. Broadcast news corpora are used to generate language models and acoustic models for speaker identification. Compared with conventional discourse segmentation algorithms based on only text information, our integrated method operates more efficiently with more accurate...

The performance of speaker verification (SV) systems degrades rapidly in noise rendering them unsuitable for security-critical applications in mobile phones, where false acceptance rates (FAR) of ∼ 10 −4 are required. However, less demanding applications for which equal error rates (EER) comparable to word error rates (WER) of speech recognizers are acceptable could benefit from the SV technology.

This paper describes a GMM-based speaker verification system that uses speaker-dependent background models transformed by speaker-specific maximum likelihood linear transforms to achieve a sharper separation between the target and the non- target acoustic region. The effect of tying, or coupling, Gaus- sian components between the target and the background model is studied and shown to be a relevant factor with

This paper investigates the use of the dimensionality-reduction techniques weighted linear discriminant analysis (WLDA), and weighted median fisher discriminant analysis (WMFD), before probabilistic linear discriminant analysis (PLDA) modeling for the purpose of improving speaker verification performance in the presence of high inter-session variability. Recently it was shown that WLDA techniques can provide improvement over traditional linear discriminant analysis (LDA) for channel compensation in i-vector based speaker verification systems. We show in this paper that the speaker discriminative information that is available in the distance between pair of speakers clustered in the development i-vector space can also be exploited in heavy-tailed PLDA modeling by using the weighted dis-criminant approaches prior to PLDA modeling. Based upon the results presented within this paper using the NIST 2008 Speaker Recognition Evaluation dataset, we believe that WLDA and WMFD projections before PLDA modeling can provide an improved approach when compared to uncompensated PLDA modeling for i-vector based speaker verification systems.

Support vector machines (SVMs), and kernel classifiers in general, rely on the kernel functions to measure the pairwise similarity between inputs. This paper advocates the use of discrete representation of speech signals in terms of the probabilities of discrete events as feature for speaker verification and proposes the use of Bhattacharyya coefficient as the similarity measure for this type of inputs to SVM. We analyze the effectiveness of the Bhattacharyya measure from the perspective of feature normalization and distribution warping in the SVM feature space. Experiments conducted on the NIST 2006 speaker verification task indicate that the Bhattacharyya measure outperforms the Fisher kernel, term frequency log-likelihood ratio (TFLLR) scaling, and rank normalization reported earlier in literature. Moreover, the Bhattacharyya measure is computed using a data-independent square-root operation instead of datadriven normalization, which simplifies the implementation. The effectiveness of the Bhattacharyya measure becomes more apparent when channel compensation is applied at the model and score levels. The performance of the proposed method is close to that of the popular GMM supervector with a small margin.