I-vector based speaker recognition using advanced channel compensation techniques

This paper investigates advanced channel compensation techniques for the purpose of improving i-vector speaker verification performance in the presence of high intersession variability using the NIST 2008 and 2010 SRE corpora. The performance of four channel compensation techniques: (a) weighted maximum margin criterion (WMMC), (b) source-normalized WMMC (SN-WMMC), (c) weighted linear discriminant analysis (WLDA), and; (d) source-normalized WLDA (SN-WLDA) have been investigated. We show that, by extracting the discriminatory information between pairs of speakers as well as capturing the source variation information in the development i-vector space, the SN-WLDA based cosine similarity scoring (CSS) i-vector system is shown to provide over 20% improvement in EER for NIST 2008 interview and microphone verification and over 10% improvement in EER for NIST 2008 telephone verification, when compared to SN-LDA based CSS i-vector system. Further, score-level fusion techniques are analyzed to combine the best channel compensation approaches, to provide over 8% improvement in DCF over the best single approach, (SN-WLDA), for NIST 2008 interview/ telephone enrolment-verification condition. Finally, we demonstrate that the improvements found in the context of CSS also generalize to state-of-the-art GPLDA with up to 14% relative improvement in EER for NIST SRE 2010 interview and microphone verification and over 7% relative improvement in EER for NIST SRE 2010 telephone verification.

Improving short utterance based I-vector speaker recognition using source and utterance-duration normalization techniques

A significant amount of speech is typically required for speaker verification system development and evaluation, especially in the presence of large intersession variability. This paper introduces a source and utterance duration normalized linear discriminant analysis (SUN-LDA) approaches to compensate session variability in short-utterance i-vector speaker verification systems. Two variations of SUN-LDA are proposed where normalization techniques are used to capture source variation from both short and full-length development i-vectors, one based upon pooling (SUN-LDA-pooled) and the other on concatenation (SUN-LDA-concat) across the duration and source-dependent session variation. Both the SUN-LDA-pooled and SUN-LDA-concat techniques are shown to provide improvement over traditional LDA on NIST 08 truncated 10sec-10sec evaluation conditions, with the highest improvement obtained with the SUN-LDA-concat technique achieving a relative improvement of 8% in EER for mis-matched conditions and over 3% for matched conditions over traditional LDA approaches.

Improving the PLDA based speaker verification in limited microphone data conditions

A significant amount of speech data is required to develop a robust speaker verification system, but it is difficult to find enough development speech to match all expected conditions. In this paper we introduce a new approach to Gaussian probabilistic linear discriminant analysis (GPLDA) to estimate reliable model parameters as a linearly weighted model taking more input from the large volume of available telephone data and smaller proportional input from limited microphone data. In comparison to a traditional pooled training approach, where the GPLDA model is trained over both telephone and microphone speech, this linear-weighted GPLDA approach is shown to provide better EER and DCF performance in microphone and mixed conditions in both the NIST 2008 and NIST 2010 evaluation corpora. Based upon these results, we believe that linear-weighted GPLDA will provide a better approach than pooled GPLDA, allowing for the further improvement of GPLDA speaker verification in conditions with limited development data.

The heterogeneous cluster ensemble method using hubness for clustering text documents

We propose a cluster ensemble method to map the corpus documents into the semantic space embedded in Wikipedia and group them using multiple types of feature space. A heterogeneous cluster ensemble is constructed with multiple types of relations i.e. document-term, document-concept and document-category. A final clustering solution is obtained by exploiting associations between document pairs and hubness of the documents. Empirical analysis with various real data sets reveals that the proposed meth-od outperforms state-of-the-art text clustering approaches.

Sequential decision fusion of multibiometrics applied to text-dependent speaker verification for controlled errors

Reliability of the performance of biometric identity verification systems remains a significant challenge. Individual biometric samples of the same person (identity class) are not identical at each presentation and performance degradation arises from intra-class variability and inter-class similarity. These limitations lead to false accepts and false rejects that are dependent. It is therefore difficult to reduce the rate of one type of error without increasing the other. The focus of this dissertation is to investigate a method based on classifier fusion techniques to better control the trade-off between the verification errors using text-dependent speaker verification as the test platform. A sequential classifier fusion architecture that integrates multi-instance and multisample fusion schemes is proposed. This fusion method enables a controlled trade-off between false alarms and false rejects. For statistically independent classifier decisions, analytical expressions for each type of verification error are derived using base classifier performances. As this assumption may not be always valid, these expressions are modified to incorporate the correlation between statistically dependent decisions from clients and impostors. The architecture is empirically evaluated by applying the proposed architecture for text dependent speaker verification using the Hidden Markov Model based digit dependent speaker models in each stage with multiple attempts for each digit utterance. The trade-off between the verification errors is controlled using the parameters, number of decision stages (instances) and the number of attempts at each decision stage (samples), fine-tuned on evaluation/tune set. The statistical validation of the derived expressions for error estimates is evaluated on test data. The performance of the sequential method is further demonstrated to depend on the order of the combination of digits (instances) and the nature of repetitive attempts (samples). The false rejection and false acceptance rates for proposed fusion are estimated using the base classifier performances, the variance in correlation between classifier decisions and the sequence of classifiers with favourable dependence selected using the 'Sequential Error Ratio' criteria. The error rates are better estimated by incorporating user-dependent (such as speaker-dependent thresholds and speaker-specific digit combinations) and class-dependent (such as clientimpostor dependent favourable combinations and class-error based threshold estimation) information. The proposed architecture is desirable in most of the speaker verification applications such as remote authentication, telephone and internet shopping applications. The tuning of parameters - the number of instances and samples - serve both the security and user convenience requirements of speaker-specific verification. The architecture investigated here is applicable to verification using other biometric modalities such as handwriting, fingerprints and key strokes.

Algorithm clustering for multi-algorithm processor design

An Application Specific Instruction-set Processor (ASIP) is a specialized processor tailored to run a particular application/s efficiently. However, when there are multiple candidate applications in the application’s domain it is difficult and time consuming to find optimum set of applications to be implemented. Existing ASIP design approaches perform this selection manually based on a designer’s knowledge. We help in cutting down the number of candidate applications by devising a classification method to cluster similar applications based on the special-purpose operations they share. This provides a significant reduction in the comparison overhead while resulting in customized ASIP instruction sets which can benefit a whole family of related applications. Our method gives users the ability to quantify the degree of similarity between the sets of shared operations to control the size of clusters. A case study involving twelve algorithms confirms that our approach can successfully cluster similar algorithms together based on the similarity of their component operations.

XML clustering and its application to XML transformation

The continuous growth of the XML data poses a great concern in the area of XML data management. The need for processing large amounts of XML data brings complications to many applications, such as information retrieval, data integration and many others. One way of simplifying this problem is to break the massive amount of data into smaller groups by application of clustering techniques. However, XML clustering is an intricate task that may involve the processing of both the structure and the content of XML data in order to identify similar XML data. This research presents four clustering methods, two methods utilizing the structure of XML documents and the other two utilizing both the structure and the content. The two structural clustering methods have different data models. One is based on a path model and other is based on a tree model. These methods employ rigid similarity measures which aim to identifying corresponding elements between documents with different or similar underlying structure. The two clustering methods that utilize both the structural and content information vary in terms of how the structure and content similarity are combined. One clustering method calculates the document similarity by using a linear weighting combination strategy of structure and content similarities. The content similarity in this clustering method is based on a semantic kernel. The other method calculates the distance between documents by a non-linear combination of the structure and content of XML documents using a semantic kernel. Empirical analysis shows that the structure-only clustering method based on the tree model is more scalable than the structure-only clustering method based on the path model as the tree similarity measure for the tree model does not need to visit the parents of an element many times. Experimental results also show that the clustering methods perform better with the inclusion of the content information on most test document collections. To further the research, the structural clustering method based on tree model is extended and employed in XML transformation. The results from the experiments show that the proposed transformation process is faster than the traditional transformation system that translates and converts the source XML documents sequentially. Also, the schema matching process of XML transformation produces a better matching result in a shorter time.

Quality based frame selection for face clustering in news video

Clustering identities in a broadcast video is a useful task to aid in video annotation and retrieval. Quality based frame selection is a crucial task in video face clustering, to both improve the clustering performance and reduce the computational cost. We present a frame work that selects the highest quality frames available in a video to cluster the face. This frame selection technique is based on low level and high level features (face symmetry, sharpness, contrast and brightness) to select the highest quality facial images available in a face sequence for clustering. We also consider the temporal distribution of the faces to ensure that selected faces are taken at times distributed throughout the sequence. Normalized feature scores are fused and frames with high quality scores are used in a Local Gabor Binary Pattern Histogram Sequence based face clustering system. We present a news video database to evaluate the clustering system performance. Experiments on the newly created news database show that the proposed method selects the best quality face images in the video sequence, resulting in improved clustering performance.

Improving short utterance i-vector speaker verification using utterance variance modelling and compensation techniques

This paper proposes techniques to improve the performance of i-vector based speaker verification systems when only short utterances are available. Short-length utterance i-vectors vary with speaker, session variations, and the phonetic content of the utterance. Well established methods such as linear discriminant analysis (LDA), source-normalized LDA (SN-LDA) and within-class covariance normalisation (WCCN) exist for compensating the session variation but we have identified the variability introduced by phonetic content due to utterance variation as an additional source of degradation when short-duration utterances are used. To compensate for utterance variations in short i-vector speaker verification systems using cosine similarity scoring (CSS), we have introduced a short utterance variance normalization (SUVN) technique and a short utterance variance (SUV) modelling approach at the i-vector feature level. A combination of SUVN with LDA and SN-LDA is proposed to compensate the session and utterance variations and is shown to provide improvement in performance over the traditional approach of using LDA and/or SN-LDA followed by WCCN. An alternative approach is also introduced using probabilistic linear discriminant analysis (PLDA) approach to directly model the SUV. The combination of SUVN, LDA and SN-LDA followed by SUV PLDA modelling provides an improvement over the baseline PLDA approach. We also show that for this combination of techniques, the utterance variation information needs to be artificially added to full-length i-vectors for PLDA modelling.

Improving PLDA speaker verification with limited development data

This paper analyses the probabilistic linear discriminant analysis (PLDA) speaker verification approach with limited development data. This paper investigates the use of the median as the central tendency of a speaker’s i-vector representation, and the effectiveness of weighted discriminative techniques on the performance of state-of-the-art length-normalised Gaussian PLDA (GPLDA) speaker verification systems. The analysis within shows that the median (using a median fisher discriminator (MFD)) provides a better representation of a speaker when the number of representative i-vectors available during development is reduced, and that further, usage of the pair-wise weighting approach in weighted LDA and weighted MFD provides further improvement in limited development conditions. Best performance is obtained using a weighted MFD approach, which shows over 10% improvement in EER over the baseline GPLDA system on mismatched and interview-interview conditions.

An efficient hybrid evolutionary optimization algorithm based on PSO and SA for clustering

The K-means algorithm is one of the most popular techniques in clustering. Nevertheless, the performance of the K-means algorithm depends highly on initial cluster centers and converges to local minima. This paper proposes a hybrid evolutionary programming based clustering algorithm, called PSO-SA, by combining particle swarm optimization (PSO) and simulated annealing (SA). The basic idea is to search around the global solution by SA and to increase the information exchange among particles using a mutation operator to escape local optima. Three datasets, Iris, Wisconsin Breast Cancer, and Ripley’s Glass, have been considered to show the effectiveness of the proposed clustering algorithm in providing optimal clusters. The simulation results show that the PSO-SA clustering algorithm not only has a better response but also converges more quickly than the K-means, PSO, and SA algorithms.

The ranking based constrained document clustering method and its application to social event detection

With the growing size and variety of social media files on the web, it’s becoming critical to efficiently organize them into clusters for further processing. This paper presents a novel scalable constrained document clustering method that harnesses the power of search engines capable of dealing with large text data. Instead of calculating distance between the documents and all of the clusters’ centroids, a neighborhood of best cluster candidates is chosen using a document ranking scheme. To make the method faster and less memory dependable, the in-memory and in-database processing are combined in a semi-incremental manner. This method has been extensively tested in the social event detection application. Empirical analysis shows that the proposed method is efficient both in computation and memory usage while producing notable accuracy.

Short utterance PLDA speaker verification using SN-WLDA and variance modelling techniques

This paper proposes a combination of source-normalized weighted linear discriminant analysis (SN-WLDA) and short utterance variance (SUV) PLDA modelling to improve the short utterance PLDA speaker verification. As short-length utterance i-vectors vary with the speaker, session variations and phonetic content of the utterance (utterance variation), a combined approach of SN-WLDA projection and SUV PLDA modelling is used to compensate the session and utterance variations. Experimental studies have found that a combination of SN-WLDA and SUV PLDA modelling approach shows an improvement over baseline system (WCCN[LDA]-projected Gaussian PLDA (GPLDA)) as this approach effectively compensates the session and utterance variations.

Document clustering algorithms, representations and evaluation for information retrieval

This thesis presents new methods for classification and thematic grouping of billions of web pages, at scales previously not achievable. This process is also known as document clustering, where similar documents are automatically associated with clusters that represent various distinct topic. These automatically discovered topics are in turn used to improve search engine performance by only searching the topics that are deemed relevant to particular user queries.