Speaker Authentication Using GMM-UBM

Speaker verification is desirable widely in many speech related applications, such as automatic telephone banking and biometric security system. This paper proposes an application of line spectrum frequency (LSF), formant, and support vector machine (SVM) to develop an algorithm of text-dependent speaker verification system. First, LSF and formant are extracted from the voiced password provided by the user. Then the proposed algorithm will make use of SVM to train the speaker characteristics from these speaker features and finally generate a claimed speaker model to discriminate between the speaker and other impostors. Experiments were conducted on the real speech signals and shown the performance of the proposed algorithm yields an equal error rate (EER) of 2.12% with 8-order LSFs and formant information. In addition, both of the false acceptance rate (FAR) and the false rejection rate (FRR) are also improved remarkably.

In this paper, we address the problem of score fusion in biometric authentication. Single valued metrics related to the receiver operating characteristics (ROC) curve, such as Equal Error Rate (EER) and False Rejection Rate (FRR) when False Acceptance Rate equals zero, are extensively used for evaluating biometric authentication performances. Various requirements and preferences, for example, lower EER, or smaller FRR, may be imposed on biometric authentication systems in different application scenarios. We propose a novel method of score fusion based on quasi-convex optimization to directly improve biometric authentication metrics. Experiments based on a face recognition system demonstrate the effectiveness of the proposed method.

Detection of replay spoof speech using teager energy feature cues

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

The problem of speaker recognition from a given set of speakers for any language and any context is considered. A database of Russian numerals that contains speech segments from 216 men and 177 women, each of whom spoke from 400 to 800 words, is used for recognition. Speech has been recorded on different types of microphones in different rooms at the natural noise level. Recognition is based on solutions of the inverse problem of finding the voice excitation pulse shape for each pitch period by the known speech segment. The pulse shape is defined as the inverse Fourier transform of the regularized ratio of speech signal spectra at the intervals of the open and closed glottis. Recognition is carried out by ten parameters: the pitch period, the open glottis interval duration, times when the source amplitude is maximum, minimum, or zero, the amplitude ratio for the minimum and maximum source pulses, three decomposition ratios of the source function by the principal component method, and the vowel duration. In such a recognition procedure, in the case of the utterance of a word that contains one vowel, the false reject rate (FRR) for men is 1.7---5.4%, and the false acceptance rate (FAR) is 5.4---7.1%. For women FRR = 2---5.2% and FAR = 5.2---6.3%. The recognition error decreases with an increasing number of vowels in the speech signal. At 10 vowels, for men FRR = 0.05---0.2% and FAR = 0.07---0.8%, and for women FRR = 0.09---0.2% and FAR = 0.17---2.1%.

Classification of ADHD and Non-ADHD Subjects Using a Universal Background Model.

Voice-based phishing and wire-fraud attacks have become a topical problem in the recent years due to the emergence of advanced AI-based speech synthesis models. These models can generate realistic speech signal of a known target that is difficult to differentiate from a bonafide voice of real human. This proved to be an issue by the recent security reports related to vishing attacks on bank call centers, fraud or pranking of public figures, and spoofing of voice authentication systems. Current approaches to address voice fraud issue are based on applying an Automatic Speaker Verification (ASV) system. In most cases, these systems are tuned on datasets that consist of wideband quality bonafide and spoofed voice samples. This makes ASV systems vulnerable to speech signal degradation caused by voice encoding in cellular network and Voice over IP (VoIP). However, performance evaluation of ASV (namely Equal Error Rate (EER) estimation) is almost exclusively available only for the cellular networks. Thus, performance of ASV systems for modern VoIP applications remains unclear. In this paper, we evaluate the modern ASV systems on audio compressed with codecs used in both cellular networks (AMR and GSM codecs) and VoIP applications (G.711, G.722, AAC, Lyra and Opus codecs). In addition, ASV performance was tested using popular VoIP application (Discord). Obtained results have shown that codec application results in considerable (up to two times) EER increase compared to the baseline results. Moreover, we observed up to three times increase in EER on data transmitted using Discord. We propose to apply hard samples mining to the training process in order to improve the accuracy of ASV systems on compressed voice samples. It allows to reduce EER from 21% down to 16% even for the most distorted samples obtained after aggressive voice compression by GSM codecs. Note, that improvement for real VoIP application is even higher - with EER on Discord data decrease from 35% to 20%.

This paper proposes an automatic stream-weight and threshold estimation method for noise-robust speaker verification using multi-stream HMMs integrating segmental and prosodic information. The proposed method simultaneously optimizes stream-weights and a decision threshold by combining the linear discriminant analysis (LDA) and Adaboost techniques. Experiments were conducted using Japanese connected digit speech contaminated by white noise with various SNRs. In this experiment, a target ratio of false acceptance rate (FAR) and false rejection rate (FRR) was set by 1:1 so as to adjust them to approach an equal error rate (EER). Experimental results show that the proposed method effectively estimates stream-weights and thresholds so that FARs and FRRs are adjusted to EERs in most of the SNR conditions

In recent years, speaker verification technologies have received an extensive amount of attention. Designing and developing machines that could communicate with humans are believed to be one of the primary motivations behind such developments. Speaker verification technologies are applied to numerous fields such as security, Biometrics, and forensics. In this paper, the authors study the effects of different languages on the performance of the automatic speaker verification (ASV) system. The MirasVoice speech corpus (MVSC), a bilingual English and Farsi speech corpus, is used in this study. This study collects results from both an I-vector based ASV system and a GMM-UBM based ASV system. The experimental results show that a mismatch between the enrolled data used for training and verification data can lead to a significant decrease in the overall system efficiency. This study shows that it is best to use an i-vector based framework with data from the English language used in the enrollment phase to improve the robustness of the ASV systems. The achieved results in this study indicate that this can narrow the degradation gap caused by the language mismatch.

Radio frequency identification (RFID) tag counterfeiting issue is prevalent throughout the world, especially in the access control and information technology sectors. Counterfeit detection is vital in digital forensic practices, especially for counterfeit RFID tag analysis. Hence, electronic fingerprint matching method is proposed to be used as a detection mechanism to detect counterfeit tags. The electronic fingerprint matching method is presented in the digital forensic investigation model that consists of seven phases.The received and backscatter powers of tag are proposed to be used as unique electronic fingerprint in the fingerprint matching method. Two statistical tests, namely, t-test and ANOVA test, are used in the fingerprint matching method. Five fingerprint matching methods are presented and are categorized based on the power response of tag and the statistical test used. Method V which uses three way ANOVA test to analyze backscatter power of tag has the most accurate results. This is because Method V has the highest area under curve (AUC) (0.999) and lowest equal error rate (EER) (0.01) values. Besides that, the false acceptance rate (FAR) and false rejection rate (FRR) obtained are 0.1 % and 1.3 %, respectively. Therefore, the proposed Method V has been proven able to detect counterfeit tags efficiently.KeywordsCounterfeitEPCglobal Class-1 Generation-2Fingerprinting Matching MethodRFID

A text-dependent speaker verification system has been implemented for forensic application since 2008 when speech recording is allowed as legal evidence in Indonesian court. Due to the laborious procedures of the current system, an automatic text-independent speaker verification system is being developed in order to accelerate the forensic speaker verification process. The new forensic system is developed using a database spoken in Indonesian. The database is built on two scenarios, interview and conversation. This is done to simulate the real forensic condition for speaker verification in Indonesia. The system adopts the very famous Gaussian Mixture Model (GMM) and Universal Background Model (UBM) framework in speaker verification field and using Mel-Frequency Cepstral Coefficients (MFCC) as the speech feature. Along with the application of zero-normalization, the new forensic speaker verification system has successfully reached an Equal Error Rate (EER) of 4.66 % that is a better performance than the previous developed system of Indonesian forensic speaker verification.

Secure buildings are currently protected from unauthorized access by a variety of devices. Nowadays, there are many kinds of devices to guarantee the building security such as PIN pads, keys both conventional and electronic, identity cards, cryptographic and dual control procedures. However, these conventional devices can be stolen, forgotten, lost, guessed or impersonated with accuracy. Biometric system based on behavioral and/or physiological characteristics of person becomes popular as an alternative method to overcome the problem of conventional method. In this paper, voice-based biometric system is introduced for access control of building security. The ability to verify the identity of a person by analyzing his/her speech, or speaker verification, is an attractive and relatively unobtrusive means of providing security for admission into an important or secured place. An individualpsilas voice cannot be stolen, lost, forgotten, guessed, or impersonated with accuracy. In the proposed system, the access may be authorized simply by means of an enrolled user speaking into a microphone attached to the system. The proposed system then will decide whether to accept or reject the userpsilas identity claim or possibly to report insufficient confidence and request additional input before making the decision. Two approaches are adopted and evaluated to model the authorized persons, namely classical Gaussian mixture model (GMM) and artificial neural network (ANN). Experimental result confirms that in terms of false acceptance rate (FAR) and false rejection rate (FRR), the proposed voice-based access control with ANN model is better than that with GMM.

The usage of Automatic Speaker Verification (ASV) in educational contexts has grown as a result of measurements of the speech signal in the claimed identity's native tongue. Even though resource-rich languages have received a lot of attention, low-resource languages, like Punjabi, still require work to be done on the development of ASV systems. In this study, an effort has been put forth to create an ASV system based on an i-vector for children while authenticating the student's dialects in Punjab region. Additionally, it has been determined how changing the Universal Background Model (UBM) size and i-vector dimensionality for the proposed system would affect the performance measure i.e. Equal Error Rate (EER). Finally, the EER values for the Malwa and Majha regions were lower than expected, at 8.47% and 9.26%, respectively. However, the output shows a large EER of 14.25% in Doaba region in comparison to that of Malwa and Majha regions.

Replay, which is to playback a pre-recorded speech sample, presents a genuine risk to automatic speaker verification technology. In this study, we evaluate the vulnerability of text-dependent speaker verification systems under the replay attack using a standard benchmarking database, and also propose an anti-spoofing technique to safeguard the speaker verification systems. The key idea of the spoofing detection technique is to decide whether the presented sample is matched to any previous stored speech samples based a similarity score. The experiments conducted on the RSR2015 database showed that the equal error rate (EER) and false acceptance rate (FAR) increased from both 2.92 % to 25.56 % and 78.36 % respectively as a result of the replay attack. It confirmed the vulnerability of speaker verification to replay attacks. On the other hand, our proposed spoofing countermeasure was able to reduce the FARs from 78.36 % and 73.14 % to 0.06 % and 0.0 % for male and female systems, respectively, in the face of replay spoofing. The experiments confirmed the effectiveness of the proposed anti-spoofing technique.

This article explores a method for speaker verification to appropriately set the threshold value used to judge the identities of individual speakers and also considers methods to update speaker models and the robustness of the speaker models so as to make the models more robust with respect to utterance variations, using a small amount of data for updating that was recently uttered. The speaker model is represented by the hidden Markov model (HMM). In the model updating, the parameters of the speaker HMM are estimated from the data for updating and the current parameter values. For setting the threshold, the new threshold value for each speaker is set with the initial value as a value that is passed to an FA rate higher than the equal error rate (a value for which the false rejection rate and the false acceptance rate are equal), which is calculated from the data for updating and which steadily approaches the value that was passed to the equal error rate in concert with updating of the speaker HMM. The results of evaluating this method using text-dependent and text-prompted speaker verification experiments with twenty speakers shows that the average error rate fell by roughly 40% for the text-independent type and by roughly 80% for the text-prompted type as compared to when the model and threshold value are not updated. © 1999 Scripta Technica, Syst Comp Jpn, 30(13): 96–105, 1999