Search

Highlights: [•] We give a survey of 24 speech databases for text-dependent speaker verification. [•] The RSR2015 corpus contains 151h of speech for text-dependent speaker verification. [•] 300 balanced-gender speakers were recorded in English on mobile devices. [•] Evaluation protocols for the 3 parts of the corpus with different text constraints. [•] Survey of classifiers and results of 2 state-of-the-art systems on the RSR2015. [ABSTRACT FROM AUTHOR]

Among conventional methods for text-independent speaker recognition, Gaussian mixture model (GMM) is known for its effectiveness and scalability in modeling the spectral distribution of speech. A GMM-supervector characterizes a speaker's voice by the GMM parameters such as the mean vectors, covariance matrices and mixture weights. Besides the first-order statistics, it is generally believed that speaker's cues are partly conveyed by the second-order statistics. In this paper, we introduce a Bhattacharyya-based GMM-distance to measure the distance between two GMM distributions. Subsequently, the GMM-UBM mean interval (GUMI) concept is introduced to derive a GUMI kernel which can be used in conjunction with support vector machine (SVM) for speaker recognition. The GUMI kernel allows us to exploit the speaker's information not only from the mean vectors of GMM but also from the covariance matrices. Moreover, by analyzing the Bhattacharyya-based GMM-distance measure, we extend the Bhattacharyya-based kernel by involving both the mean and covariance statistical dissimilarities. We demonstrate the effectiveness of the new kernel on the National Institute of Standards and Technology (NIST) speaker recognition evaluation (SRE) 2006 dataset. [ABSTRACT FROM PUBLISHER]

This paper describes the NEC-TT speaker recognition system designed for the 2018 Speaker Recognition Evaluation (SRE'18) benchmarking. The NEC-TT submission was among the best-performing systems in this latest edition of SRE organized by the National Institute of Standards and Technology (NIST). It comprises multiple sub-systems based on a deep speaker embedding front-end followed by a probabilistic linear discriminant analysis (PLDA) back-end. Speaker embeddings are continuous-valued vector representations that allow easy comparison between speaker voices with simple geometric operations. The effectiveness of deep speaker embeddings relies on the quantity and diversity of the training data. To this end, we hinge on data augmentation and mixed-bandwidth training strategies to increase the number of training examples and speakers. By doing so, we not only increase the quantity of the training data but also expand the output softmax layer with a larger number of speaker classes. From a system design perspective, we adopted a two-stage pipeline consisting of a general multi-domain speaker embedding front-end followed by a domain-specific PLDA back-end. This has a significant benefit in commercial deployment since the same speaker embedding front-end could be used with multiple domain-adapted PLDA back-ends to cater to every specific deployment. This paper provides a detailed description and analysis of the design methodology, data augmentation, bandwidth extension, multi-head attention, PLDA adaptation, and other components that have contributed to good performance in NEC-TT's SRE'18 results. [ABSTRACT FROM AUTHOR]

This paper studies the relevance factor in maximum a posteriori (MAP) adaptation of Gaussian mixture model (GMM) for speaker and language recognition. Knowing that relevance factor determines how much the observed training data influence the model adaptation, thus the resulting GMM model, it is believed that more effective modeling can be achieved if the relevance factor is adaptive to the corresponding data. We therefore provide a mathematic derivation for the estimation of relevance factor. GMM supervector support vector machine (SVM) with nuisance attribute projection (NAP) (GMM–NAP–SVM) has been reported to be effective and reliable for speaker and language recognition. Being a discriminative classifier in nature, a GMM–NAP–SVM system is sensitive to the magnitude and direction of a supervector in the high dimensional space. However, when characterizing a speech utterance with GMM supervector estimated through MAP, we observe that the resulting supervector is undesirably affected by the varying duration of the utterance. We propose an adaptive relevance factor that adapts to the duration to mitigate the variability effect due to the length of utterance. We give a systematic investigation on different types of relevance factor of MAP in different applicatively platforms. We show the efficacy of the data-dependent as well as adaptive relevance factors on the National Institute of Standards and Technology (NIST) speaker recognition evaluation (SRE) 2008 and language recognition evaluation (LRE) 2009 and 2011 tasks respectively. [ABSTRACT FROM AUTHOR]

• F-ratio method was used for anti-spoofing frequency discriminative analysis. • Adaptive frequency resolution features were designed according to frequency discriminative analysis. • A new adaptive frequency resolution Relative Phase feature was proposed. • Three types of adaptive filters were proposed and compared for pros and cons. • Complementary information from phase and magnitude were combined in replay detection. Replay attacks pose the most severe threat to automatic speaker verification systems among various spoofing attacks. In this paper, we propose a novel feature extraction method that leverages both the phase-based and magnitude-based features. The proposed method fully utilizes the subband information and the complementary information from the phase and magnitude spectra. First, we conduct a discriminative performance analysis on full frequency bands via the F-ratio method. Then, variable-frequency resolution features are extracted via several techniques to capture highly discriminative information on frequency bands. Finally, complementary information from the phase and magnitude domains are fused to achieve higher performance. The results on the ASVspoof 2017 database demonstrate that our proposed frequency adaptive features attain relative error reduction rates of 83.4% and 62.3% on the development and evaluation datasets, respectively, compared to the baseline method. [ABSTRACT FROM AUTHOR]

• We describe the protocol and design of the ASVspoof Challenge 2019 database • We detail the speech synthesis and voice conversion algorithms used in the database • We detail the carefully controlled simulation to generate replay spoofing speech • We evaluate of baseline countermeasure and ASV systems on the database • Human assessment found that one spoofing system can fool human listeners Automatic speaker verification (ASV) is one of the most natural and convenient means of biometric person recognition. Unfortunately, just like all other biometric systems, ASV is vulnerable to spoofing, also referred to as "presentation attacks." These vulnerabilities are generally unacceptable and call for spoofing countermeasures or "presentation attack detection" systems. In addition to impersonation, ASV systems are vulnerable to replay, speech synthesis, and voice conversion attacks. The ASVspoof challenge initiative was created to foster research on anti-spoofing and to provide common platforms for the assessment and comparison of spoofing countermeasures. The first edition, ASVspoof 2015, focused upon the study of countermeasures for detecting of text-to-speech synthesis (TTS) and voice conversion (VC) attacks. The second edition, ASVspoof 2017, focused instead upon replay spoofing attacks and countermeasures. The ASVspoof 2019 edition is the first to consider all three spoofing attack types within a single challenge. While they originate from the same source database and same underlying protocol, they are explored in two specific use case scenarios. Spoofing attacks within a logical access (LA) scenario are generated with the latest speech synthesis and voice conversion technologies, including state-of-the-art neural acoustic and waveform model techniques. Replay spoofing attacks within a physical access (PA) scenario are generated through carefully controlled simulations that support much more revealing analysis than possible previously. Also new to the 2019 edition is the use of the tandem detection cost function metric, which reflects the impact of spoofing and countermeasures on the reliability of a fixed ASV system. This paper describes the database design, protocol, spoofing attack implementations, and baseline ASV and countermeasure results. It also describes a human assessment on spoofed data in logical access. It was demonstrated that the spoofing data in the ASVspoof 2019 database have varied degrees of perceived quality and similarity to the target speakers, including spoofed data that cannot be differentiated from bona fide utterances even by human subjects. It is expected that the ASVspoof 2019 database, with its varied coverage of different types of spoofing data, could further foster research on anti-spoofing. [ABSTRACT FROM AUTHOR]

• Security evaluation of automatic speaker verification for arbitrarily large population. • Bayesian generative model of non-target scores. • From random-impostor evaluation to worst-case impostor false alarm rate. • Analysis of i-vector and x-vector systems up to 100,000 simulated impostors. How secure automatic speaker verification (ASV) technology is? More concretely, given a specific target speaker, how likely is it to find another person who gets falsely accepted as that target? This question may be addressed empirically by studying naturally confusable pairs of speakers within a large enough corpus. To this end, one might expect to find at least some speaker pairs that are indistinguishable from each other in terms of ASV. To a certain extent, such aim is mirrored in the standardized ASV evaluation benchmarks, for instance, the series of speaker recognition evaluation (SRE) organized by the National Institute of Standards and Technology (NIST). Nonetheless, arguably the number of speakers in such evaluation benchmarks represents only a small fraction of all possible human voices, making it challenging to extrapolate performance beyond a given corpus. Furthermore, the impostors used in performance evaluation are usually selected randomly. A potentially more meaningful definition of an impostor — at least in the context of security-driven ASV applications — would be closest (most confusable) other speaker to a given target. We put forward a novel performance assessment framework to address both the inadequacy of the random-impostor evaluation model and the size limitation of evaluation corpora by addressing ASV security against closest impostors on arbitrarily large datasets. The framework allows one to make a prediction of the safety of given ASV technology, in its current state, for arbitrarily large speaker database size consisting of virtual (sampled) speakers. As a proof-of-concept, we analyze the performance of two state-of-the-art ASV systems, based on i-vector and x-vector speaker embeddings (as implemented in the popular Kaldi toolkit), on the recent VoxCeleb 1, and 2 corpora, containing a total of 7365 speakers. We fix the number of target speakers to 1000, and generate up to N = 100 , 000 virtual impostors sampled from the generative model. The model-based false alarm rates are in a reasonable agreement with empirical false alarm rates and, as predicted, increase substantially (values up to 98%) with N = 100 , 000 impostors. Neither the i-vector or x-vector system is immune to increased false alarm rate at increased impostor database size, as predicted by the model. [ABSTRACT FROM AUTHOR]

Presentation attack (spoofing) detection (PAD) typically operates alongside biometric verification to improve reliablity in the face of spoofing attacks. Even though the two sub-systems operate in tandem to solve the single task of reliable biometric verification, they address different detection tasks and are hence typically evaluated separately. Evidence shows that this approach is suboptimal. We introduce a new metric for the joint evaluation of PAD solutions operating in situ with biometric verification. In contrast to the tandem detection cost function proposed recently, the new tandem equal error rate (t-EER) is parameter free. The combination of two classifiers nonetheless leads to a set of operating points at which false alarm and miss rates are equal and also dependent upon the prevalence of attacks. We therefore introduce the concurrent t-EER, a unique operating point which is invariable to the prevalence of attacks. Using both modality (and even application) agnostic simulated scores, as well as real scores for a voice biometrics application, we demonstrate application of the t-EER to a wide range of biometric system evaluations under attack. The proposed approach is a strong candidate metric for the tandem evaluation of PAD systems and biometric comparators.