Speaker Verification

This paper proposes text independent automatic speaker verification system using IMFCC (Inverse/ Reverse Mel Frequency Coefficients) and IT-EM (Information Theoretic Expectation Maximization). To perform speaker verification, feature extraction using Mel scale has been widely applied and has established better results. The IMFCC is based on inverse Mel-scale. The IMFCC effectively captures information available at the high frequency formants which is ignored by the MFCC. In this paper the fusion of MFCC and IMFCC at input level is proposed. GMMs (Gaussian Mixture Models) based on EM (Expectation Maximization) have been widely used for classification of text independent verification. However EM comes across the convergence issue. In this paper we use our proposed IT-EM which has faster convergence, to train speaker models. IT-EM uses information theory principles such as PDE (Parzen Density Estimation) and KL (Kullback-Leibler) divergence measure. IT-EM acclimatizes the weights, means and covariances, like EM. However, IT-EM process is not performed on feature vector sets but on a set of centroids obtained using IT (Information Theoretic) metric. The IT-EM process at once diminishes divergence measure between PDE estimates of features distribution within a given class and the centroids distribution within the same class. The feature level fusion and IT-EM is tested for the task of speaker verification using NIST2001 and NIST2004. The experimental evaluation validates that MFCC/IMFCC has better results than the conventional delta/MFCC feature set. The MFCC/IMFCC feature vector size is also much smaller than the delta MFCC thus reducing the computational burden as well. IT-EM method also showed faster convergence, than the conventional EM method, and thus it leads to higher speaker recognition scores.

Emotion based speaker Identification System is the process of automatically identifying speaker’s emotion based on features extracted from speech waves. This paper presents experiment with the building and testing of a Speaker’s emotion identification for Hindi speech using Mel Frequency Cepestral Coefficients and Vector Quantization techniques. We collected voice samples of Hindi speech sentences in four basic emotions to study speaker’s emotion identification and it was found that with proposed emo-voice model we are able to achieve accuracy of 73% of speaker’s emotion identification in speech out of 93% of the total speech samples provided to the system.

In this paper, a new method to deal with automatic speaker verification based on band-limited phaseonly correlation (BLPOC) is proposed. The aim of this study is to validate the use of the BLPOC function as a new limited-data automatic speaker verification technique. Although some speaker verification techniques have high accuracy, efficiency usually depends on the extraction of complex theoretical information from speech signals and the amount of the data for training the algorithms. The BLPOC function is a high-accuracy biometric technique traditionally implemented in human identification by fingerprints (through image-matching). When applying the BLPOC function in automatic speaker verification through the proposed algorithms (under limited-data conditions), a 98.24% true acceptance rate (TAR) and 87.17% true rejection rate (TRR) in a custom database (and 93.75% TAR and 67.05% TRR in the ELSDSR database) were obtained. The proposed algorithm is a theoretically simple method for automatic speaker verification whose main advantage is that it can provide identification under limited-data conditions. In this sense, the BLPOC function could be applicable in other limited-data biometric identifications by sound signals.

The aim of this paper is to present an Arabic speech database that represents Arabic native speakers from all the cities of Saudi Arabia. The database is called the Saudi Accented Arabic Voice Bank (SAAVB). Preparing the prompt sheets, selecting the right speakers and transcribing their speech are some of the challenges that faced the project team. The procedures that meet these challenges are highlighted. SAAVB consists of 1033 speakers speak in Modern Standard Arabic with a Saudi accent. The SAAVB content is analyzed and the results are illustrated. The content was verified internally and externally by IBM Cairo and can be used to train speech engines such as automatic speech recognition and speaker verification systems.

In this paper, we evaluate a recently proposed spectral envelope estimation method, stabilized weighted linear prediction (SWLP), in the feature extraction stage of a large vocabulary continuous speech recognizer (LVCSR) system. Using speech recorded in real-world noisy environments, we compare recognition error rates obtained with SWLP to those given by the conventional spectrum estimation methods in feature extraction. We use large vocabulary speech that is simultaneously recorded using a headset, lavalier and a fixed medium-distance microphone. When the recognizer is trained using a multicondition training set the results do not differ significantly. However, when the models are trained with clean speech and evaluated in noisy environments, the SWLP models perform significantly better than the conventional MFCC in all environments and in all recording settings.

It is known that the Percentage of Identification Accuracy (PIA) of Automatic Speaker Recognition (ASR) systems is increasingly vulnerable, such as noise and channel degradation in real-time. This study presents a novel class SVM and i-vector based speaker specific feature extractor (SVM-IVSSFE) algorithms based ASR system. I introduce SVM based approach at two stages within state-of-the-art ASR system: i-vector extraction estimation and at the Probabilistic Linear Discriminant (PLDA). SVM based approaches have classified more challenging subsets on the basis of training using a proper difficulty criterion. After this, the proposed training algorithm is initialized with a subset that reaches a clean set and then moves into subsets which more are challenging for training because the algorithms move forward. Based on the SVM-IVSSFE approach on noisy background I evaluated the PIA of ASR on critically degraded datasets with NIST 2010 database with standard i-vector extractor. The multi...

In this paper, a multimodal person verification system is presented. The system is based on face and voice modalities. Fusion of information derived from each modality is performed at the matching score level using sum rule. For face verification statistical subspace tools are utilized as feature extractors. For speaker verification, mel frequency cepstral coefficients are used as features and gaussian mixture models are used for modeling. Various combination cases are tried in the experiments and the results show that for each case the combined modalities performs better than the single modality. Bu bildiride, kişiye ait yüz ve ses bilgilerinin kullanımına dayalı bir çoğul-biyometrik kimlik doğrulama sistemi önerilmiştir. Yüz doğrulama için temel bileşen analizi (TBA) ve bağımsız bileşen analizi (BBA) yöntemleri, konuşmacı doğrulama için ise mel-frekans-kepstral katsayıları (MFKK) kullanılmıştır. Her bir biyometrik özellikten (yüz, ses) elde edilen bilgi eşleşme değeri düzeyinde toplam kuralı ile tümleştirilmiştir. Deneyler sonucunda tümleştirmenin doğrulama sisteminin başarım oranını arttırdığı gözlemlenmiştir.

remains an open problem that has not been satisfactorily solved by existing recognition techniques. In this paper, we tackle this problem using a variant of the recently proposed Probabilistic Linear Discriminant Analysis (PLDA). We show that simplified versions of the PLDA model, which are regularly used in the field of speaker recognition, rely on certain assumptions that not only result in a simpler PLDA model, but also reduce the computational load of the technique and -as indicated by our experimental assessments -improve recognition performance. Moreover, we show that, contrary to the general belief that PLDA-based methods produce well calibrated verification scores, score normalization techniques can still deliver significant performance gains, but only if nonparametric score normalization techniques are employed. Last but not least, we demonstrate the competitiveness of the simplified PLDA model for face recognition by comparing our results with the state-of-the-art results from the literature obtained on the second version of the large-scale Face Recognition Grand Challenge (FRGC) database.

ABSTRACT: Previous research has shown both that listeners' ability to detect high quality voice imitation results in judicially worrying misidentification rates (Schlichting & Sullivan, 1997) and that the semantic expectation of the listeners as to the content of a message impacts upon the acceptance of a voice imitation (Zetterholm et al., this volume). This paper considers whether the results of the Zetterholm et al. paper were due to a difference in imitation quality that was not detected in the phonetic analyses of the imitations rather than ...

Spoken language identification is the process by which the language in a spoken utterance is recognized automatically. Spoken language identification is commonly used in speech translation systems, in multi-lingual speech recognition, and in speaker diarization. In the current paper, spoken language identification based on deep learning (DL) and the i-vector paradigm is presented. Specifically, a comparative study is reported, consisting of experiments on language identification using deep neural networks (DNN) and convolutional neural networks (CNN). Also, the integration of the two methods into a complete system is investigated. Previous studies demonstrated the effectiveness of using DNN in spoken language identification. However, to date, the integration of CNN and i-vectors in language identification has not been investigated. The main advantage of using CNN is that fewer parameters are required compared to DNN. As a result, CNN is cheaper in terms of memory and the computational power needed. The proposed methods are evaluated on the NIST 2015 i-vector Machine Learning Challenge task for the recognition of 50 in-set languages. Using DNN, a 3.55% equal error rate (EER) was achieved. The EER when using CNN was 3.48%. When DNN and CNN systems were fused, an EER of 3.3% was obtained. The results are very promising, and they also show the effectiveness of using CNN and i-vectors in spoken language identification. The proposed methods are compared to a baseline method based on support vector machines (SVM) and they demonstrated significantly superior performance.

In this paper, we describe the use of Artificial Neural Network (ANN) to compute the acoustic features in analysing forensic speaker verification. In the computation, there are two datasets derived from speech recording of a simulated human trafficking crime, namely Forensic Evidence Data (FED) and Comparative Evidence Data (CED). In both datasets, sound segmentation is performed and then the acoustic features (Formant Frequencies F1, F2, F3, and F4) are extracted. The acoustic feature values are computed with ANN to predict an output with a targeted sound classification /a/, /i/ and /u/. The results are interpreted as forensic evidence against sound data in recorded evidence. With a result rate of more than 80%, this method might be studied more deeply to be developed and applied in evaluating recorded sound evidence for the legal case process.

Speech processing has emerged as one of the important application area of digital signal processing. Various fields for research in speech processing are speech recognition, speaker recognition, speech synthesis, speech coding etc. Speaker recognition is one of the most useful and popular biometric recognition techniques in the world especially related to areas in which security is a major concern. This paper presents an automatic Speaker Recognition model which is Text-Dependent where the speaker is allowed to speak only fixed text. Automatic speaker recognition is the use of a machine to recognize a person from a spoken phrase, based on individual information (characteristics of voice) included in speech waves. Recognizer block employs MFCC (Mel Frequency Cepstrum Co-efficients) technique to get hybrid features for speaker identification/verification system. These features are used to train the ANN classifier in the training phase. Later in the testing phase the speaker is recognised based on the ANN classifier. The accuracy of 92% is achieved.

Detecting duplicate patient participation in clinical trials is a major challenge because repeated patients can undermine the credibility and accuracy of the trial's findings and result in significant health and financial risks. Developing accurate automated speaker verification (ASV) models is crucial to verify the identity of enrolled individuals and remove duplicates, but the size and quality of data influence ASV performance. However, there has been limited investigation into the factors that can affect ASV capabilities in clinical environments. In this paper, we bridge the gap by conducting analysis of how participant demographic characteristics, audio quality criteria, and severity level of Alzheimer's disease (AD) impact the performance of ASV utilizing a dataset of speech recordings from 659 participants with varying levels of AD, obtained through multiple speech tasks. Our results indicate that ASV performance: 1) is slightly better on male speakers than on female speakers; 2) degrades for individuals who are above 70 years old; 3) is comparatively better for nonnative English speakers than for native English speakers; 4) is negatively affected by clinician interference, noisy background, and unclear participant speech; 5) tends to decrease with an increase in the severity level of AD. Our study finds that voice biometrics raise fairness concerns as certain subgroups exhibit different ASV performances owing to their inherent voice characteristics. Moreover, the performance of ASV is influenced by the quality of speech recordings, which underscores the importance of improving the data collection settings in clinical trials.

We present a discriminative learning framework for Gaussian mixture models (GMMs) used for classification based on the extended Baum-Welch (EBW) algorithm . We suggest two criteria for discriminative optimization, namely the class conditional likelihood (CL) and the maximization of the margin (MM). In the experiments, we present results for synthetic data, broad phonetic classification, and a remote sensing application. The experiments show that CL-optimized GMMs (CL-GMMs) achieve a lower performance compared to MM-optimized GMMs (MM-GMMs), whereas both discriminative GMMs (DGMMs) perform significantly better than generatively learned GMMs. We also show that the generative discriminatively parameterized GMM classifiers still allow to marginalize over missing features, a case where generative classifiers have an advantage over purely discriminative classifiers such as support vector machines or neural networks.

This paper introduces the Voices Obscured In Complex Environmental Settings (VOICES) corpus, a freely available dataset under Creative Commons BY 4.0. This dataset will promote speech and signal processing research of speech recorded by far-field microphones in noisy room conditions. Publicly available speech corpora are mostly composed of isolated speech at close-range microphony. A typical approach to better represent realistic scenarios, is to convolve clean speech with noise and simulated room response for model training. Despite these efforts, model performance degrades when tested against uncurated speech in natural conditions. For this corpus, audio was recorded in furnished rooms with background noise played in conjunction with foreground speech selected from the Lib-riSpeech corpus. Multiple sessions were recorded in each room to accommodate for all foreground speech-background noise combinations. Audio was recorded using twelve microphones placed throughout the room, resulting in 120 hours of audio per microphone. This work is a multi-organizational effort led by SRI International and Lab41 with the intent to push forward state-of-the-art distant microphone approaches in signal processing and speech recognition.

In this paper, we address the problem of automatically detecting whether the audio and visual speech modalities in frontal pose videos are synchronous or not. This is of interest in a wide range of applications, for example spoof detection in biometrics, lip-syncing, speaker detection and diarization in multi-subject videos, and video data quality assurance. In our adopted approach, we investigate the use of deep neural networks (DNNs) for this purpose. The proposed synchrony DNNs operate directly on audio and visual features over relatively wide contexts, or, alternatively, on appropriate hidden (bottleneck) or output layers of DNNs trained for single-modal or audio-visual automatic speech recognition. In all cases, the synchrony DNN classes consist of the "in-sync" and a number of "out-of-sync" targets, the latter considered at multiples of ± 30 msec steps of overall asynchrony between the two modalities. We apply the proposed approach on two multi-subject audio-visual databases, one of high-quality data recorded in studio-like conditions, and one of data recorded by smart cell-phone devices. On both sets, and under a speaker-independent experimental framework, we are able to achieve very low equal-error-rates in distinguishing "in-sync" from "out-of-sync" data.

In this paper, we address the problem of automatically detecting whether the audio and visual speech modalities in frontal pose videos are synchronous or not. This is of interest in a wide range of applications, for example spoof detection in biometrics, lip-syncing, speaker detection and diarization in multi-subject videos, and video data quality assurance. In our adopted approach, we investigate the use of deep neural networks (DNNs) for this purpose. The proposed synchrony DNNs operate directly on audio and visual features over relatively wide contexts, or, alternatively, on appropriate hidden (bottleneck) or output layers of DNNs trained for single-modal or audio-visual automatic speech recognition. In all cases, the synchrony DNN classes consist of the "in-sync" and a number of "out-of-sync" targets, the latter considered at multiples of ± 30 msec steps of overall asynchrony between the two modalities. We apply the proposed approach on two multi-subject audio-visual databases, one of high-quality data recorded in studio-like conditions, and one of data recorded by smart cell-phone devices. On both sets, and under a speaker-independent experimental framework, we are able to achieve very low equal-error-rates in distinguishing "in-sync" from "out-of-sync" data.

Speaker recognition systems achieved significant improvements over the last decade, especially due to the performance of the i-vectors. Despite the achievements, mismatch between training and test data affects the recognition performance considerably. In this paper, a solution is offered to increase robustness against additive noises by inserting model compensation techniques within the i-vector extraction scheme. For stationary noises, the model compensation techniques produce highly robust systems. Parallel Model Compensation and Vector Taylor Series are considered as state-of-the-art model compensation techniques. Applying these methods to the first order statistics, a noisy total variability space training is aimed, which will reduce the mismatch resulted by additive noises. All other parts of the conventional i-vector scheme remain unchanged, such as total variability matrix training, reducing the i-vector dimensionality, scoring the i-vectors. The proposed method was tested wi...

Speaker Recognition systems exhibit a decrease in performance when the input speech is not in optimal circumstances, for example when the user is under emotional or stress conditions. The objective of this paper is measuring the effects of stress on speech to ultimately try to mitigate its consequences on a speaker recognition task. On this paper, we develop a stress-robust speaker identification system using data selection and augmentation by means of the manipulation of the original speech utterances. An extensive experimentation has been carried out for assessing the effectiveness of the proposed techniques. First, we concluded that the best performance is always obtained when naturally stressed samples are included in the training set, and second, when these are not available, their substitution and augmentation with synthetically generated stress-like samples, improves the performance of the system.

Speaker verification is crucial in biometric security systems, but performance degradation occurs under mismatched noise and recording conditions. This paper explores acoustic feature learning techniques to enhance robustness in speaker verification systems. We analyze state-of-the-art methods, recent advances from 2023, and novel feature learning strategies, including deep learning models. Empirical evaluations demonstrate the effectiveness of selected acoustic features under diverse noise scenarios. The study provides comparative analyses through tables and graphs, offering insights into optimal feature learning strategies.

Human listeners are able to understand speech in the presence of a noisy background. How to simulate this perceptual ability remains a great challenge. This paper describes a preliminary evaluation of intelligibility of the output of a monaural speech segregation system. The system performs speech segregation in two stages. The first stage segregates voiced speech using supervised learning of harmonic features, and the second stage segregates unvoiced speech by subtracting noise energy that is estimated from voiced intervals and onset/offset based segmentation. Objective evaluation in terms of the match to ideal binary time-frequency masks shows substantial improvements. Tests with human subjects indicate that the system improves intelligibility for young listeners when the input SNR is very low, but does not aid elderly listeners. This preliminary evaluation identifies aspects of the system that should be improved in order to produce consistent improvement in intelligibility in noisy environments.

The ASVSpoof challenges goal is to evaluate countermeasures to spoof attacks on automatic speaker verification systems. We first analyze in more details the results of the baseline systems provided by the organization and unveil several weaknesses for some types of attack. In particular for the physical access (PA) task, replay attacks with low reverberation time and/or high quality of the replay device are problematic. Based on this observation, we propose several improvements. Firstly, a specific learning targeting the problematic types of attack. Secondly, a new type of feature enhancing the reverberation. Thirdly, a Deep Neural Network with more modelling capability. On the development set of the PA task, each proposed improvements show results amelioration for the targeted types of attack. Furthermore, the ensemble systems based on this proposed improvements show great overall results amelioration compared to the baseline (0.140 vs 0.193 min t-DCF). However, the amelioration is less encouraging on the evaluation set (0.225 vs 0.245 min t-DCF), thus raising the question of over-fitting as the development set and the train set are similar.

In this paper, we present the winning BUT submission for the text-dependent task of the SdSV challenge 2020. Given the large amount of training data available in this challenge, we explore successful techniques from text-independent systems in the text-dependent scenario. In particular, we trained x-vector extractors on both in-domain and out-of-domain datasets and combine them with i-vectors trained on concatenated MFCCs and bottleneck features, which have proven effective for the text-dependent scenario. Moreover, we proposed the use of phrase-dependent PLDA backend for scoring and its combination with a simple phrase recognizer, which brings up to 63% relative improvement on our development set with respect to using standard PLDA. Finally, we combine our different i-vector and x-vector based systems using a simple linear logistic regression score level fusion, which provides 28% relative improvement on the evaluation set with respect to our best single system.

In this paper, we present the winning BUT submission for the text-dependent task of the SdSV challenge 2020. Given the large amount of training data available in this challenge, we explore successful techniques from text-independent systems in the text-dependent scenario. In particular, we trained x-vector extractors on both in-domain and out-of-domain datasets and combine them with i-vectors trained on concatenated MFCCs and bottleneck features, which have proven effective for the text-dependent scenario. Moreover, we proposed the use of phrase-dependent PLDA backend for scoring and its combination with a simple phrase recognizer, which brings up to 63% relative improvement on our development set with respect to using standard PLDA. Finally, we combine our different i-vector and x-vector based systems using a simple linear logistic regression score level fusion, which provides 28% relative improvement on the evaluation set with respect to our best single system.

This paper presents the approach used in the BirdCLEF 2020 Competition. The objective of the competition is to try to recognize bird species through its sings and calls among 960 species in soundscapes. We use a MultiScale CNN + Triplet Loss to learn the melspectrogram characteristics that differentiate the species from each other. The CNN create a 128-D embedding used to classify the species. The approach achieves third place on the competition with c-mAP of 0.009752 and r-mAP of 0.008. We also present some changes on train parameters done after the competition that achieve our best evaluation with c-mAP 0.062877 and r-mAP of 0.108.

Speaker Verification (SV) is a task to verify the claimed identity of the claimant using his/her voice sample. Though there exists an ample amount of research in SV technologies, the development concerning a multilingual conversation is limited. In a country like India, almost all the speakers are polyglot in nature. Consequently, the development of a Multilingual SV (MSV) system on the data collected in the Indian scenario is more challenging. With this motivation, the Indic-Multilingual Speaker Verification (I-MSV) Challenge 2022 has been designed for understanding and comparing the stateof-the-art SV techniques. For the challenge, approximately 100 hours of data spoken by 100 speakers has been collected using 5 different sensors in 13 Indian languages. The data is divided into development, training, and testing sets and has been made publicly available for further research. The goal of this challenge is to make the SV system robust to language and sensor variations between enrollment and testing. In the challenge, participants were asked to develop the SV system in two scenarios, viz. constrained and unconstrained. The best system in the constrained and unconstrained scenario achieved a performance of 2.12% and 0.26% in terms of Equal Error Rate (EER), respectively.

Automatic speaker verification (ASV) systems are vulnerable to spoofing attacks. We propose a spoofing-robust ASV system optimized directly for the recently introduced architecture-agnostic detection cost function (a-DCF), which allows targeting a desired trade-off between the contradicting aims of user convenience and robustness to spoofing. We combine a-DCF and binary cross-entropy (BCE) with a novel straightforward threshold optimization technique. Our results with an embedding fusion system on ASVspoof2019 data demonstrate relative improvement of 13% over a system trained using BCE only (from minimum a-DCF of 0.1445 to 0.1254). Using an alternative non-linear score fusion approach provides relative improvement of 43% (from minimum a-DCF of 0.0508 to 0.0289).

Research has shown that handset selectors can be used to assist telephone-based speech/speaker recognition. Most handset selectors, however, simply select the most likely handset from a set of known handsets even for speech coming from an 'unseen' handset. This paper proposes a divergence-based handset selector with out-of-handset (OOH) rejection capability to identify the 'unseen' handsets. This is achieved by measuring the Jensen difference between the selector's output and a constant vector with identical elements. The resulting handset selector is combined with a feature-based channel compensation algorithm for telephonebased speaker verification. Utterances whose handsets were identified as 'unseen' are either transformed by a global bias vector or normalized by cepstral mean subtraction (CMS). On the other hand, if the handset can be identified (considered as 'seen'), its corresponding transformation parameters will be used to transform the utterances. Experiments based on ten handsets of the HTIMIT corpus show that using the transformation parameters of the 'seen' handsets to transform the utterances with correctly identified handsets and processing those utterances with 'unseen' handsets by CMS achieve the best result.

In speaker verification, a claimant may produce two or more utterances. In our previous study , we proposed to compute the optimal weights for fusing the scores of these utterances based on their score distribution and our prior knowledge about the score statistics estimated from the mean scores of the corresponding client speaker and some pseudo-impostors during enrollment. As the fusion weights depend on the prior scores, in this paper, we propose to adapt the prior scores during verification based on the likelihood of the claimant being an impostor. To this end, a pseudo-imposter GMM score model is created for each speaker. During verification, the claimant's scores are fed to the score model to obtain a likelihood for adapting the prior score. Experimental results based on the GSM-transcoded speech of 150 speakers from the HTIMIT corpus demonstrate that the proposed prior score adaptation approach provides a relative error reduction of 15% when compared with our previous approach where the prior scores are non-adaptive.

In telephone-based speaker verification, the channel conditions can be varied significantly from sessions to sessions. Therefore, it is desirable to estimate the channel conditions online and compensate the acoustic distortion without prior knowledge of the channel characteristics. Because no a priori knowledge is used, the estimation accuracy depends greatly on the length of the verification utterances. This paper extends the Blind Stochastic Feature Transformation (BSFT) algorithm that we recently proposed to handle the short-utterance scenario. The idea is to estimate a set of prior transformation parameters from a development set in which a wide variety of channel conditions exists in the verification utterances. The prior transformations are then incorporated into the online estimation of the BSFT parameters in a Bayesian (maximum a posteriori) fashion. The resulting transformation parameters are therefore dependent on both the prior transformations and the verification utterances. For short (long) utterances, the prior transformations play a more (less) important role. We referred the extended algorithm to as Bayesian BSFT (BBSFT) and applied it to the 2001 NIST SRE task. Results show that Bayesian BSFT outperforms BSFT for utterances shorter than or equal to 4 seconds.

In speaker verification, a claimant may produce two or more utterances. Typically, the scores of the speech patterns extracted from these utterances are averaged and the resulting mean score is compared with a decision threshold. Rather than simply computing the mean score, we propose to compute the optimal weights for fusing the scores based on the score distribution of the independent utterances and our prior knowledge about the score statistics. More specifically, we use enrollment data to compute the mean scores of client speakers and impostors and consider them to be the prior scores. During verification, we set the fusion weights for individual speech patterns to be a function of the dispersion between the scores of these speech patterns and the prior scores. Experimental results based on the GSM-transcoded speech of 150 speakers from the HTIMIT corpus demonstrate that the proposed fusion algorithm can increase the dispersion between the mean speaker scores and the mean impostor scores. Compared with a baseline approach where equal weights are assigned to all scores, the proposed approach provides a relative error reduction of 19%.

Fusion techniques have been widely used in multi-modal biometric authentication systems. While these techniques are mainly applied to combine the outputs of modality-dependent classifiers, they can also be applied to fuse the decisions or scores from a single modality. The idea is to consider the multiple samples extracted from a single modality as independent but coming from the same source. In this chapter, we propose a single-source, multi-sample data-dependent fusion algorithm for speaker verification. The algorithm is data-dependent in that the fusion weights are dependent on the verification scores and the prior score statistics of claimed speakers and background speakers. To obtain the best out of the speaker's scores, scores from multiple utterances are sorted before they are probabilistically combined. Evaluations based on 150 speakers from a GSM-transcoded corpus are presented. Results show that data-dependent fusion of speaker's scores is significantly better than the conventional score averaging approach. It was also found that the proposed fusion algorithm can be further enhanced by sorting the score sequences before they are probabilistically combined.

In many biometric systems, the scores of multiple samples (e.g. utterances) are averaged and the average score is compared against a decision threshold for decision making. The average score, however, may not be optimal because the distribution of the scores is ignored. To address this limitation, we have recently proposed a fusion model that incorporates the score distribution by making the fusion weights dependent on the dispersion between the framebased scores and the prior score statistics obtained from training data. As the fusion weights are data-dependent, the positions of scores in the score sequences become detrimental to the final fused scores. In this paper, we propose to enhance the fusion model by sorting the score sequences before fusion takes place. The fusion model was evaluated on a speaker verification task where each claimant utters two utterances in a verification session. Results demonstrate that fusion of sorted scores has the effect of maximizing the dispersion between the client scores and the impostor scores, making the verification process more reliable. Compared with our previous work where no sorting is applied, the new approach reduces the equal error rate by 11%.

This paper proposes an articulatory feature-based conditional pronunciation modeling (AFCPM) technique for speaker verification. The technique models the pronunciation behaviors of speakers by creating a link between the actual phones produced by the speakers and the state of articulations during speech production. Speaker models consisting of conditional probabilities of two articulatory classes are adapted from a set of universal background models (UBMs) using MAP adaptation technique. This adaptation approach aims to prevent over-fitting the speaker models when the amount of speaker data is insufficient for a direct estimation. Experimental results show that the adaptation technique can enhance the discriminating power of speaker models by establishing a tighter coupling between speaker models and the UBMs. Results also show that fusing the scores derived from an AFCPM-based system and a conventional spectral-based system achieves a significantly lower error rate than that of the individual systems. This suggests that AFCPM and spectral features are complementary to each other.

Acoustic mismatch between the training and recognition conditions presents one of the serious challenges faced by speaker recognition researchers today. The goal of channel compensation is to achieve performance approaching that of a "matched condition" system while avoiding the need for a large amount of training data. It is important to ensure that the channel compensation algorithms in these systems compensate the channel variation instead of speaker variation. This paper addresses the problem of unsupervised compensation in which the features of a test utterance are transformed to fit the clean speaker model and gender-dependent background model. Specifically, a feature-based transformation is estimated based on the statistical difference between a test utterance and a composite acoustic model formed by combining the speaker and background models. By transforming the features to fit both models, the transformation is implicitly constrained. Experimental results based on the 2001 NIST evaluation set show that the proposed transformation approach achieves significant improvement in both equal error rate and minimum detection cost as compared to cepstral mean subtraction, Znorm and short-time Gaussianization.

This paper proposes a speaker verification system based on articulatory feature-based conditional pronunciation modeling (AFCPM). The system captures the pronunciation characteristics of speakers by modeling the linkage between the actual phones produced by the speakers and the state of articulations during speech production. The speaker models, which consist of conditional probabilities of two articulatory classes, are adapted from a set of universal background models (UBMs) via MAP adaptation. This creates a direct coupling between the speaker and background models, which prevents over-fitting the speaker models when the amount of speaker data is limited. Experimental results demonstrate that MAP adaptation not only enhances the discriminative power of the speaker models but also improves their robustness against handset mismatches. Results also show that fusing the scores derived from an AFCPM-based system and a conventional spectral-based system achieves an error rate that is significantly lower than that can be achieved by the individual systems. This suggests that AFCPM and spectral features are complementary to each other.

This paper presents an approach that uses articulatory features (AFs) derived from spectral features for telephone-based speaker verification. To minimize the acoustic mismatch caused by different handsets, handset-specific normalization is applied to the spectral features before the AFs are extracted. Experimental results based on 150 speakers using 10 different handsets show that AFs contain useful speaker-specific information for speaker verification and the use of handset-specific normalization significantly lowers the error rates under the handset mismatched conditions. Results also demonstrate that fusing the scores obtained from an AF-based system with those obtained from a spectral feature-based (MFCC) system helps lower the error rates of the individual systems.

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.

To improve the reliability of telephone-based speaker verification systems, channel compensation is indispensable. However, it is also important to ensure that the channel compensation algorithms in these systems surpress channel variations and enhance interspeaker distinction. This paper addresses this problem by a blind feature-based transformation approach in which the transformation parameters are determined online without any a priori knowledge of channel characteristics. Specifically, a composite statistical model formed by the fusion of a speaker model and a background model is used to represent the characteristics of enrollment speech. Based on the difference between the claimant's speech and the composite model, a stochastic matching type of approach is proposed to transform the claimant's speech to a region close to the enrollment speech. Therefore, the algorithm can estimate the transformation online without the necessity of detecting the handset types. Experimental results based on the 2001 NIST evaluation set show that the proposed transformation approach achieves significant improvement in both equal error rate and minimum detection cost as compared to cepstral mean subtraction, Znorm, and short-time Gaussianization.

The ETSI has recently published a front-end processing standard for distributed speech recognition systems. The key idea of the standard is to extract the spectral features of speech signals at the front-end terminals so that acoustic distortion caused by communication channels can be avoided. This paper investigates the effect of extracting spectral features from different stages of the front-end processing on the performance of distributed speaker verification systems. A technique that combines handset selectors with stochastic feature transformation is also employed in a back-end speaker verification system to reduce the acoustic mismatch between different handsets. Because the feature vectors obtained from the back-end server are vector quantized, the paper proposes two approaches to adding Gaussian noise to the quantized feature vectors for training the Gaussian mixture speaker models. In one approach, the variances of the Gaussian noise are made dependent on the codeword distance. In another approach, the variances are a function of the distance between some unquantized training vectors and their closest code vector. The HTIMIT corpus was

With the rise of crimes in Automated Teller Machines, the security of the ATM is at stake. The Traditional Security Methods such as passwords or pins had always been a cause of worry to the users because of it getting lost, stolen or forgotten. The biometric authentication uses one's unique features as password. The paper deals with one of the oldest biometric traits, Fingerprint Recognition. The paper moreover adds a sophisticated voice input system and a Password Breaking scheme. The end results of a highly secure and sophisticated ATM system enhances the efficacy of the system, providing customer satisfaction in banking sectors.

High-level characteristics such as word usage, pronunciation, phonotactics, prosody, etc., have seen a resurgence for automatic speaker recognition over the last several years. With the availability of many conversation sides per speaker in current corpora, high-level systems now have the amount of data needed to sufficiently characterize a speaker. Although a significant amount of work has been done in finding novel high-level features, less work has been done on modeling these features. We describe a method of speaker modeling based upon support vector machines. Current high-level feature extraction produces sequences or lattices of tokens for a given conversation side. These sequences can be converted to counts and then frequencies of -gram for a given conversation side. We use support vector machine modeling of these n-gram frequencies for speaker verification. We derive a new kernel based upon linearizing a log likelihood ratio scoring system. Generalizations of this method are shown to produce excellent results on a variety of high-level features. We demonstrate that our methods produce results significantly better than standard log-likelihood ratio modeling. We also demonstrate that our system can perform well in conjunction with standard cesptral speaker recognition systems.

This paper introduces a modeling flow for predicting waveforms as a function of parameters, variables in the system generating the waveforms. In order to achieve this goal, a neural network is involved. The model is developed using early-stage simulation data from the automotive industry. Usually, a large amount of data is necessary in order to properly create such a model and successfully train a neural network, and this can be problematic. To address this issue, a model which can be trained with just a handful of characteristics is proposed. The results obtained show that our model can predict waveforms based on input factors with high accuracy.

This thesis attempts to solve the problem of authorship verification. Authorship verification is a subdomain of authorship analysis and its origins lie in stylometry analysis. However most of the research in authorship analysis is based on authorship identification where authorship verification is rather unexplored. With the increase of digital documents and authors it is very difficult to employ authorship identification solutions. Hence in such cases authorship verification solutions are in necessity. This research focuses on utilizing digital documents with 1000 words, written in English to solve the problem of authorship verification: coming into conclusion about the authorship of a text in dispute by analyzing texts written by some candidate author. To solve this problem three machine learning models were designed employing two feature sets, comprising of linguistic features which are suggested to characterize the writing style of a person, one comprising of stylometric features and other consisting of word frequency based features. One-class support vector machine and two-class support vector machine are used as machine learning models to tackle this problem. Results suggest one-class support vector machine with selected stylometric features does not tackle the problem very well while two-class classification model with stylometric features trained for known author class and unknown author class shows potential in solving the problem if the unknown author class can be properly represented. One-class support vector machine with word frequency based features, shows promising results in solving the authorship verification problem. By conducting this research, I have developed an immense interest in stylometry analysis and natural language processing and gained my first experiences in research world. Hence I would like to thank my supervisor, Dr. A. R. Weerasinghe in introducing me to the project and advising me in any way needed and trusting my capabilities. I would also like to thank Dr. H. Ekanayake for effectively coordinating the final year project in computer science and helping students in case of problematic situations. I am immensely grateful for my batchmates who also helped me in many ways and giving suggestions to improve the project. I would also like to be thankful towards my family who gave me great support and courage to carry out the research studies and providing with a suitable academic environment. Lastly my appreciation goes to everyone who helped during this attempt. Document -A digital file containing text entirely written by one person. Known document -A document, with prior knowledge of the person who has written it. Unknown document -A document, with no knowledge of the person who has written it. Known author -The author who is suspected to have written the unknown document. Unknown author -The author of the unknown document, if the unknown document is written by a different person than the known author. Target class -The class that contain all documents with known authorship from a suspected author. Outlier class -The class which contains all documents from other authors than the suspected author.

Usually the mel-frequency cepstral coefficients are estimated either from a periodogram or from a windowed periodogram. We state a general estimator which also includes multitaper estimators. We propose approximations of the variance and bias of the estimate of each coefficient. By using Monte Carlo computations, we demonstrate that the approximations are accurate. Using the proposed formulas, the peak matched multitaper estimator is shown to have low mean square error (squared bias + variance) on speech-like processes. It is also shown to perform slightly better in the NIST 2006 speaker verification task as compared to the Hamming window conventionally used in this context.

Speaker verification is the process of accepting or rejecting claimed identity in terms of its sound features. A speaker verification system can be used for numerous security systems, including bank account accessing, getting to security points, criminology and etc. When a speaker verification system wants to check the identity of individuals remotely, it confronts problems such as noise effect on speech signal and also identity falsification with speech synthesis. In this system, we have proposed a new speaker verification system based on Multi Model GMM, called SV-MMGMM, in which all speakers are divided into seven different age groups, and then an isolated GMM model for each group is created; instead of one model for all speakers. In order to evaluate, the proposed method has been compared with several speaker verification systems based on Naïve, SVM, Random Forest, Ensemble and basic GMM. Experimental results show that the proposed method has so better efficiency than others.

This paper addresses the problem of speaker verification in two speaker conversations, proposing a set of confidence measures to assess the quality of a given speaker segmentation. In addition we study how these measures can be used to estimate the performance of a state of the art speaker verification system. Our approach for speaker segmentation is based on the eigenvoice paradigm. We present a novel PCA based initialization in the speaker factor space along with a modification of the speaker turn duration distribution that improves the performance of previously reported Joint Factor Analysis based speaker segmentation systems. Three confidence measures are analyzed on the output of the proposed segmentation system for the summed-channel telephone data of the NIST Speaker Recognition Evaluation 2008, showing that they constitute a good measure to estimate not only the segmentation accuracy but the performance of a speaker verification system when it faces two speaker conversations.

Audio deepfakes, a subset of deepfake technology, employ machine learning or deep learning to create deceptive audio content by synthesizing authentic recordings. Such deepfakes not only fosters the dissemination of misinformation but also empowers identity theft while compromising individual privacy. Discerning between counterfeit and authentic audio content poses escalating challenges for digital forensic analysts. The proposed paper develops a robust deep learning model that harnesses fusion approach with a spectrum of diverse audio spectral features to effectively detect deepfake audios. By employing a fusion strategy, the developed model ensembles predictions from two pre-trained networks, CIFAR-10 and ResNet50. Additionally, it capitalizes on a diverse array of spectral audio features-Mel Frequency Cepstral Coefficients (MFCC), Constant-Q Cepstral Coefficients (CQCC), Mel-Spectrogram, and Spectral Centroid, for extraction of crucial details from raw audio data. Accuracy of proposed model is assessed on more recent and widely used FoR dataset having three sub-datasets of 195,000 audio samples. Experimental results reveal that proposed model achieves superior performance, boasting an accuracy of 99.12%, precision of 97.54%, recall of 98.44%, and F1 score of 98.12% when utilizing MFCC feature as compared to other audio features. Moreover, the model undergoes an accuracy-centric quantitative assessment, surpassing eight state-of-the-art audio detection models, including DNN, DeepSonar, STN, TCN, SVM, CNN, KNN, and RF.

In the last years, the i-vector approach became the state-of-theart in speaker recognition systems. As in previous approaches, i-vector -based systems suffer greatly in presence of additive noise, especially in low SNR cases. In this paper, we will describe a statistical framework allowing to estimate a clean i-vector given the noisy one or to integrate, directly, statistical knowledges about the noise and clean i-vectors in the scoring phase. The proposed procedure is essentially based on a method which enables to produce statistical knowledge about the noise effect in the i-vector domain. The work presented here is based on the hypothesis that the noise effect is Gaussian and additive in the i-vector space. To validate our approach, experiments were carried out on NIST 2008 data (det7). Significant improvement was observed compared to the baseline system and to the "muti-style" backend training technique.

Les progrès de performance en vérification du locuteur ces quinze dernières années sont incontestables. Les systèmes sont de plus en plus sûrs dans le sens où les taux EER ou DCF diminuent d'année en année. Pourtant, il est nécessaire de déterminer dans quelles circonstances les systèmes d'identification du locuteur sont fiables. Des études ont été menées pour analyser les performances en fonction du locuteur. Dans cet article, nous nous intérrogeons sur la variation des performances observée en fonction du signal de parole utilisé pour représenter le locuteur. Des variations très importantes des valeurs d'EER sont obtenues pour deux systèmes état de l'art. Nous proposons également une méthode pour mesurer la variation de performance propre au système. Les valeurs d'EER varient alors d'un point.