Universal Background Model Research Articles

Enhanced speaker verification using an adaptive multiple low-rank representation based on the modified adaptive Gaussian mixture model framework

In this paper, a new method for the calculation of the observation-confidence value that is applied in the modified adaptive Gaussian mixture model framework is proposed for speaker verification. First, an adaptive version of the multiple low-rank representation method, for which a weighted decomposition that incorporates the prior information regarding the speech/non-speech content is considered, is proposed to find the enhanced speech and for the estimation of the frame signal-to-noise ratio (SNR) values. Then, a simple sigmoid function is applied to convert the frame SNR values into the observation-confidence values. To verify the accuracy of the system, we use utterances from the Korean movie You Came From The Stars. The experiment results show that our proposed approach achieves a greater accuracy compared with the other well-known baseline methods, such as the GMM-based universal background model, the GMM supervector-based support vector machine (SVM), the i-vector-based SVM, and the sparse representation, under the noisy environment.

HMM-Based Phrase-Independent i-Vector Extractor for Text-Dependent Speaker Verification

The low-dimensional i-vector representation of speech segments is used in the state-of-the-art text-independent speaker verification systems. However, i-vectors were deemed unsuitable for the text-dependent task, where simpler and older speaker recognition approaches were found more effective. In this work, we propose a straightforward hidden Markov model (HMM) based extension of the i-vector approach, which allows i-vectors to be successfully applied to text-dependent speaker verification. In our approach, the Universal Background Model (UBM) for training phrase-independent i-vector extractor is based on a set of monophone HMMs instead of the standard Gaussian Mixture Model (GMM). To compensate for the channel variability, we propose to precondition i-vectors using a regularized variant of within-class covariance normalization, which can be robustly estimated in a phrase-dependent fashion on the small datasets available for the text-dependent task. The verification scores are cosine similarities between the i-vectors normalized using phrase-dependent s-norm. The experimental results on RSR2015 and RedDots databases confirm the effectiveness of the proposed approach, especially in rejecting test utterances with a wrong phrase. A simple MFCC based i-vector/HMM system performs competitively when compared to very computationally expensive DNN-based approaches or the conventional relevance MAP GMM-UBM, which does not allow for compact speaker representations. To our knowledge, this paper presents the best published results obtained with a single system on both RSR2015 and RedDots dataset.

Modeling Dynamics of Expressive Body Gestures In Dyadic Interactions

Body gestures are an important non-verbal expression channel during affective communication. They convey human attitudes and emotions as they dynamically unfold during an interpersonal interaction. Hence, it is highly desirable to understand the dynamics of body gestures associated with emotion expression in human interactions. We present a statistical framework for robustly modeling the dynamics of body gestures in dyadic interactions. Our framework is based on high-level semantic gesture patterns and consists of three components. First, we construct a universal background model (UBM) using Gaussian mixture modeling (GMM) to represent subject-independent gesture variability. Next, we describe each gesture sequence as a concatenation of semantic gesture patterns which are derived from a parallel HMM structure. Then, we probabilistically compare the segments of each gesture sequence extracted from the second step with the UBM obtained from the first step, in order to select highly probabilistic gesture patterns for the sequence. The dynamics of each gesture sequence are represented by a statistical variation profile computed from the selected patterns, and are further described in a well-defined kernel space. This framework is compared with three baseline models and is evaluated in emotion recognition experiments, i.e., recognizing the overall emotional state of a participant in a dyadic interaction from the gesture dynamics. The recognition performance demonstrates the superiority of the proposed framework over the baseline models. The analysis of the relationship between the emotion recognition performance and the number of the selected segments also indicates that a few local salient events, rather than the whole gesture sequence, are sufficiently informative to trigger the human summarization of their overall global emotion perception.

Text-independent speaker identification using robust statistics estimation

It is well-known that the performance of Gaussian mixture model-based text-independent speaker identification systems deteriorates significantly with the presence of noise and spectral distortion in the training and testing utterances. In this paper, we propose a novel GMM-based speaker identification system based on two robust-statistics estimation methods: the minimum volume ellipsoid method, and the minimum covariance determinant method. Compared to the traditional maximum likelihood estimation method, the proposed methods are less sensitive to outliers in the feature-vector space caused by additive noise and spectral distortion. Moreover, in the testing phase, we propose a simple distance metric to be used for comparing the unknown testing utterance against the speakers’ models. Furthermore, we derive a more robust version of the i-vector extractor, named robust i-vector, which utilizes our proposed robust estimation methods for estimating the parameters of the base universal background model. The proposed classification system has been applied to the NIST 2000 speaker recognition evaluation and the COSINE database. It has also been compared against state-of-the-art techniques such as the GMM/UBM method, the super-vectors method, and the i-vector methods. Experimental results show that the proposed classification system provides up to 16% relative improvement in the identification performance over the i-vector methods for short utterances in the NIST 2000 database and up to 8% when the utterances of the NIST 2000 database are contaminated by different types of artificial noise for signal-to-noise ratio ranging from 0 to 20 dB. For the COSINE database, the robust i-vector estimation provides an absolute improvement of up to 8%. Finally, the real time factor of the proposed distance metric for testing is 55% higher than the RT of the regular ML scoring.

Identification Features Analysis in Speech Data Using Gmm-Ubm Speaker Verification System

This paper is devoted to feature selection and evaluation in an automatic text-independent speaker verification task. In order to solve this problem a speaker verification system based on the Gaussian mixture model and the universal background model (GMM-UBM system) was used. The application sphere and challenges of modern systems of automatic speaker identification were considered. Overview of the modern speaker recognition methods and main speech features used in speaker identification is provided. Features extraction process used in this article was examined. Reviewed speech features were used for speaker verification including mel-cepstral coefficients (MFCC), line spectral pairs (LSP), perceptual linear prediction cepstral coefficients (PLP), short-term energy, formant frequencies, fundamental frequency, voicing probability, zero crossing rate (ZCR), jitter and shimmer. The experimental evaluation of the GMM-UBM system using different speech features was conducted on a 50 speaker set and a result is presented. Feature selection was done using the genetic algorithm and the greedy adding and deleting algorithm. Equal error rate (EER) equals 0,579 % when using 256 component Gaussian mixture model and the obtained feature vector. Comparing to standard 14 MFCC vector, 42,1 % of EER improvement was acquired.

Spectral Features for Synthetic Speech Detection

Recent advancements in voice conversion (VC) and speech synthesis research make speech-based biometric systems highly prone to spoofing attacks. This can provoke an increase in false acceptance rate in such systems and requires countermeasure to mitigate such spoofing attacks. In this paper, we first study the characteristics of synthetic speech vis-a-vis natural speech and then propose a set of novel short-term spectral features that can efficiently capture the discriminative information between them. The proposed features are computed using inverted frequency warping scale and overlapped block transformation of filter bank log energies. Our study presents a detailed analysis of antispoofing performance with respect to the variations in the warping scale for inverted frequency and block size for the block transform. For performance analysis, Gaussian mixture model (GMM) based synthetic speech detector is used as a classifier on a stand-alone basis and also, integrated with automatic speaker verification (ASV) systems. For ASV systems, standard mel-frequency cepstral coefficients are used as feature while GMM with universal background model and i-vector are used as classifiers. The experiments are conducted on ten different kinds of synthetic data from ASVspoof 2015 corpus. The results show that the countermeasures based on the proposed features outperform other spectral features for both known and unknown attacks. An average equal error rate (EER) of 0.00% has been achieved for nine attacks that use VC or SS speech and the best performance of 7.12% EER is arrived at the remaining natural speech concatenation-based spoofing attack.

Speaker Model Clustering to Construct Background Models for Speaker Verification

Abstract Conventional speaker recognition systems use the Universal Background Model (UBM) as an imposter for all speakers. In this paper, speaker models are clustered to obtain better imposter model representations for speaker verification purpose. First, a UBM is trained, and speaker models are adapted from the UBM. Then, the k-means algorithm with the Euclidean distance measure is applied to the speaker models. The speakers are divided into two, three, four, and five clusters. The resulting cluster centers are used as background models of their respective speakers. Experiments showed that the proposed method consistently produced lower Equal Error Rates (EER) than the conventional UBM approach for 3, 10, and 30 seconds long test utterances, and also for channel mismatch conditions. The proposed method is also compared with the i-vector approach. The three-cluster model achieved the best performance with a 12.4% relative EER reduction in average, compared to the i-vector method. Statistical significance of the results are also given.

A study on the effects of using short utterance length development data in the design of GPLDA speaker verification systems

This paper studies the performance degradation of Gaussian probabilistic linear discriminant analysis (GPLDA) speaker verification system, when only short-utterance data is used for speaker verification system development. Subsequently, a number of techniques, including utterance partitioning and source-normalised weighted linear discriminant analysis (SN-WLDA) projections are introduced to improve the speaker verification performance in such conditions. Experimental studies have found that when short utterance data is available for speaker verification development, GPLDA system overall achieves best performance with a lower number of universal background model (UBM) components. As a lower number of UBM components significantly reduces the computational complexity of speaker verification system, that is a useful observation. In limited session data conditions, we propose a simple utterance-partitioning technique, which when applied to the LDA-projected GPLDA system shows over 8% relative improvement on EER values over baseline system on NIST 2008 truncated 10---10 s conditions. We conjecture that this improvement arises from the apparent increase in the number of sessions arising from our partitioning technique and this helps to better model the GPLDA parameters. Further, partitioning SN-WLDA-projected GPLDA shows over 16% and 6% relative improvement on EER values over LDA-projected GPLDA systems respectively on NIST 2008 truncated 10---10 s interview-interview, and NIST 2010 truncated 10---10 s interview-interview and telephone-telephone conditions.

Dialect Identification of Assamese Language using Spectral Features

Objective: Accurate dialect identification technique helps in improving the speech recognition systems that exist in most of the present day electronic devices and is also expected to help in providing new services in the field of e-health and telemedicine which is especially important for older and homebound people. Methods: In this paper we have developed the speech corpora needed for the dialect identification purpose and described a method to identify Assamese language (an Indian language) and two of its dialects, viz., Kamrupi and Goalparia. Findings: Research work done on dialect identification is relatively much less than that on language identification for which one of the reasons being dearth of sufficient database on dialects. As mentioned, we have developed the database and then Mel-Frequency Cepstral Coefficient has been used to extract the spectral features of the collected speech data. Two modeling techniques, namely, Gaussian Mixture Model and Gaussian Mixture Model with Universal Background Model have been used as the modeling techniques to identify the dialects and language. Novelty: So far, standard speech database for Assamese dialects that can be used for speech processing research has not been made. In this paper, we not only describe a method to identify dialects of the Assamese language, but have also developed the speech corpora needed for the dialect identification purpose. Keywords: Assamese, Gaussian Mixture Model, Gaussian Mixture Model with Universal Background Model, Goalparia, Kamrupi, Mel-Frequency Cepstral Coefficient

Radio Frequency Traffic Classification Over WLAN

Network traffic classification is the process of analyzing traffic flows and associating them to different categories of network applications. Network traffic classification represents an essential task in the whole chain of network security. Some of the most important and widely spread applications of traffic classification are the ability to classify encrypted traffic, the identification of malicious traffic flows, and the enforcement of security policies on the use of different applications. Passively monitoring a network utilizing low-cost and low-complexity wireless local area network (WLAN) devices is desirable. Mobile devices can be used or existing office desktops can be temporarily utilized when their computational load is low. This reduces the burden on existing network hardware. The aim of this paper is to investigate traffic classification techniques for wireless communications. To aid with intrusion detection, the key goal is to passively monitor and classify different traffic types over WLAN to ensure that network security policies are adhered to. The classification of encrypted WLAN data poses some unique challenges not normally encountered in wired traffic. WLAN traffic is analyzed for features that are then used as an input to six different machine learning (ML) algorithms for traffic classification. One of these algorithms (a Gaussian mixture model incorporating a universal background model) has not been applied to wired or wireless network classification before. The authors also propose a ML algorithm that makes use of the well-known vector quantization algorithm in conjunction with a decision tree—referred to as a TRee Adaptive Parallel Vector Quantiser. This algorithm has a number of advantages over the other ML algorithms tested and is suited to wireless traffic classification. An average F-score (harmonic mean of precision and recall) > 0.84 was achieved when training and testing on the same day across six distinct traffic types.

Detection of Double Compressed AMR Audio Using Stacked Autoencoder

The adaptive multi-rate (AMR) audio codec adopted by many portable recording devices is widely used in speech compression. The use of AMR speech recordings as evidence in court is growing. Nowadays, it is easy to tamper with digital speech recordings, which makes audio forensics increasingly important. The detection of double compressed audio is one of the key issues in audio forensics. In this paper, we propose a framework for detecting double compressed AMR audio based on the stacked autoencoder (SAE) network and the universal background model—Gaussian mixture model (UBM-GMM). Instead of hand-crafted features, we used the SAE to learn the optimal features automatically from the audio waveforms. Audio frames are used as network input and the last hidden layer’s output constitutes the features of a single frame. For an audio clip with many frames, the features of all the frames are aggregated and classified by UBM-GMM. Experimental results show that our method is effective in distinguishing single/double compressed AMR audio and outperforms the existing methods by achieving a detection accuracy of 98% on the TIMIT database. Exhaustive experiments demonstrate the effectiveness and robustness of the proposed method.

Speaker diarization system using HXLPS and deep neural network

In general, speaker diarization is defined as the process of segmenting the input speech signal and grouped the homogenous regions with regard to the speaker identity. The main idea behind this system is that it is able to discriminate the speaker signal by assigning the label of the each speaker signal. Due to rapid growth of broadcasting and meeting, the speaker diarization is burdensome to enhance the readability of the speech transcription. In order to solve this issue, Holoentropy with the eXtended Linear Prediction using autocorrelation Snapshot (HXLPS) and deep neural network (DNN) is proposed for the speaker diarization system. The HXLPS extraction method is newly developed by incorporating the Holoentropy with the XLPS. Once we attain the features, the speech and non-speech signals are detected by the Voice Activity Detection (VAD) method. Then, i-vector representation of every segmented signal is obtained using Universal Background Model (UBM) model. Consequently, DNN is utilized to assign the label for the speaker signal which is then clustered according to the speaker label. The performance is analysed using the evaluation metrics, such as tracking distance, false alarm rate and diarization error rate. The outcome of the proposed method ensures the better diarization performance by achieving the lower DER of 1.36% based on lambda value and DER of 2.23% depends on the frame length.

New transformed features generated by deep bottleneck extractor and a GMM\u2013UBM classifier for speaker age and gender classification

Speaker age and gender classification is one of the most challenging problems in speech signal processing. Recently with developing technologies, identifying speaker age and gender information has become a necessity for speaker verification and identification systems such as identifying suspects in criminal cases, improving human–machine interaction, and adapting music for awaiting people queue. Despite the intensive studies that have been conducted to extract descriptive and distinctive features, the classification accuracies are still not satisfactory. In this work, a model for generating bottleneck features from a deep neural network and a Gaussian Mixture Model–Universal Background Model (GMM–UBM) classifier are proposed for speaker age and gender classification problem. Deep neural network with a bottleneck layer is trained in an unsupervised manner for calculating the initial weights between layers. Then, it is trained and tuned in a supervised manner to generate transformed mel-frequency cepstral coefficients (T-MFCCs). The GMM–UBM is used to build a GMM model for each class, and the models are used to classify speaker age and gender. Age-annotated database of German telephone speech (aGender) is used to evaluate the proposed classification system. The newly generated T-MFCCs have shown potential to achieve significant classification improvements in speaker age and gender classification by using the GMM–UBM classifier. The proposed classification system achieved an overall accuracy of 57.63%. The highest accuracy is calculated as 72.97% for adult female speakers.

Structure of pauses in speech in the context of speaker verification and classification of speech type

Statistics of pauses appearing in Polish as a potential source of biometry information for automatic speaker recognition were described. The usage of three main types of acoustic pauses (silent, filled and breath pauses) and syntactic pauses (punctuation marks in speech transcripts) was investigated quantitatively in three types of spontaneous speech (presentations, simultaneous interpretation and radio interviews) and read speech (audio books). Selected parameters of pauses extracted for each speaker separately or for speaker groups were examined statistically to verify usefulness of information on pauses for speaker recognition and speaker profile estimation. Quantity and duration of filled pauses, audible breaths, and correlation between the temporal structure of speech and the syntax structure of the spoken language were the features which characterize speakers most. The experiment of using pauses in speaker biometry system (using Universal Background Model and i-vectors) resulted in 30 % equal error rate. Including pause-related features to the baseline Mel-frequency cepstral coefficient system has not significantly improved its performance. In the experiment with automatic recognition of three types of spontaneous speech, we achieved 78 % accuracy, using GMM classifier. Silent pause-related features allowed distinguishing between read and spontaneous speech by extreme gradient boosting with 75 % accuracy.

A classification method for crowded situation using environmental sounds based on Gaussian mixture model-universal background model

This paper presents a method to classify a situation of being crowded with people using environmental sounds that are collected by smartphones. A final goal of the research is to estimate “crowd-density” using only environmental sounds collect by smartphones. Advantages of the approach are (1) acoustic singles can be collected and processed at low cost, (2) because many people carry smartphones, crowd-density can be obtained not only from many places, but also at any time. As the first step, in this paper, we tried to classify “a situation of being crowded with people.” We collected environmental sounds using smartphones both in residential area and downtown area. The total duration of the collected data is 77,900 seconds. The sound of “crowded with people” is defined as buzz-buzz where more than one person talked at the same time. Two kinds of classifiers were trained on the basis of the GMM-UBM (Gaussian Mixture Model and Universal Background Model) method. The one was trained with acoustic features that are generally used in speech recognition, and the other was trained with additional parameter of sound power. Experiment results showed that the parameter of sound power improves the F-measure to 0.60 from 0.58.

Similarity Search of Acted Voices for Automatic Voice Casting

This paper presents a large-scale similarity search of professionally acted voices for computer-aided voice casting. The proposed voice casting system explores Gaussian mixture model-based acoustic models and multilabel recognition of perceived paralinguistic content (speaker states and speaker traits, e.g., age/gender, voice quality, emotion) for the voice casting of professionally acted voices. First, acoustic models (universal background model, super-vector, i-vector) are constructed to model the acoustic space of voices, from which the similarity between voices can be measured directly in the acoustic space. Second, multiple binary classification of speaker traits and states is added to the acoustic models in order to represent the vocal signature of a voice, which is then used to measure the similarity between voices in the paralinguistic space. Finally, a similarity search is processed in order to determine the set of target actors that are the most similar to the voice of a source actor. In a subjective experiment conducted in the real-context of cross-language voice casting, the multilabel scoring system significantly outperforms the acoustic scoring system. This constitutes a proof of concept for the role of perceived para-linguistic categories in the perception of voice similarity.

Noise Robust Speaker Identification Using RASTA–MFCC Feature with Quadrilateral Filter Bank Structure

This paper motivates the use of Relative Spectra---Mel Frequency Cepstral Coefficients (RASTA---MFCC) feature extracted from the newly designed Quadrilateral filter bank structure and Gaussian Mixture Model---Universal Background Model (GMM---UBM) for improved text independent speaker identification under noisy environment. Unlike neural network model which requires retraining of entire database when a new sample is added to it, GMM---UBM model does not require retraining of entire database which leads to easier and faster processing. RASTA---MFCC is found to be more robust to noisy environment compared with traditional MFCC method. MFCC is an efficient feature for identifying the speaker as it has speaker specific information capturing ability. RASTA processing of speech improves the performance of recognizer in the presence of convolution and additive noise. This work combines the better of these two processes to yield RASTA---MFCC feature which is robust to noise and also proposes a new Quadrilateral filter bank structure which approximates the response of cochlear membrane of human ear to effectively capture the feature vectors. The proposed Quadrilateral filter bank structure with RASTA---MFCC feature and GMM---UBM modeling for speaker identification demonstrates supremacy over triangular and Gaussian filter banks and offers a speaker identification accuracy of 97.67 % for the MEPCO noisy speech database with 50 speakers.

Applications of UBMs and I-vectors in EEG subject verification.

The processing of electroencephalograms (EEGs) is a growing field where mature speech processing techniques are able to rapidly progress development and understanding of the associated neuroscience. I-vectors and Joint Factor Analysis (JFA), along with their foundational universal background models (UBMs) have progressed to a level of understanding that makes them prime for transition to the EEG community. To prove the capability of these techniques they are tested against two contrasting EEG data sets, PhysioNet's EEG Motor Movement/Imagery Dataset and the Temple University Hospital EEG Corpus, to highlight the effectiveness of the techniques with minimal domain knowledge modifications. The initial results, presented as equal error rates as low as 20%, support the development of these techniques as a viable approach to addressing subject verification within and across subjects.

Text-Dependent Speaker Recognition With Random Digit Strings

In this paper, we explore joint factor analysis (JFA) for text-dependent speaker recognition with random digit strings. The core of the proposed method is a JFA model by which we extract features. These features can either represent overall utterances or individual digits, and are fed into a trainable backend to estimate likelihood ratios. Within this framework, several extensions are proposed. First is a logistic regression method for combining log-likelihood ratios that correspond to individual mixture components. Second is the extraction of phonetically aware Baum--Welch statistics, by using forced alignment instead of the typical posterior probabilities that are derived by the universal background model. We also explore a digit-string-dependent way to apply score normalization that exhibits a notable improvement compared to the standard one. By fusing six JFA features, we attained 2.01% and 3.19% equal error rates on male and female, respectively, on the challenging RSR2015 (part III) dataset.

On-line signature verification using Gaussian Mixture Models and small-sample learning strategies

This paper addresses the problem of training on-line signature verification systems when the number of training samples is small, facing the real-world scenario when the number of available signatures per user is limited. The paper evaluates nine different classification strategies based on Gaussian Mixture Models (GMM), and the Universal Background Model (UBM) strategy, which are designed to work under small-sample size conditions. The GMM’s learning strategies include the conventional Expectation-Maximisation algorithm and also a Bayesian approach based on variational learning. The signatures are characterised mainly in terms of velocities and accelerations of the users’ handwriting patterns. The results show that for a genuine vs. impostor test, the GMM-UBM method is able to keep the accuracy above 93%, even when only 20% of samples are used for training (5 signatures). Moreover, the combination of a full Bayesian UBM and a Support Vector Machine (SVM) (known as GMM-Supervector) is able to achieve 99% of accuracy when the training samples exceed 20. On the other hand, when simulating a real environment where there are not available impostor signatures, once again the combination of a full Bayesian UBM and a SVM, achieve more than 77% of accuracy and a false acceptance rate lower than 3%, using only 20% of the samples for training.