Xavier Sevillano

Towards High-Quality Next-Generation Text-to-Speech Synthesis: A Multidomain Approach by Automatic Domain Classification

This paper is a contribution to the recent advancements in the development of high-quality next generation text-to-speech (TTS) synthesis systems. Two of the hottest research topics in this area are oriented towards the improvement of speech expressiveness and flexibility of synthesis. In this context, this paper presents a new TTS strategy called multidomain TTS (MD-TTS) for synthesizing among different domains. Although the multidomain philosophy has been widely applied in spoken language systems, few research efforts have been conducted to extend it to the TTS field. To do so, several proposals are described in this paper. First, a text classifier (TC) is included in the classic TTS architecture in order to automatically conduct the selection of the most appropriate domain for synthesizing the input text. In contrast to classic topic text classification tasks, the MD-TTS TC should not only consider the contents of text but also its structure. To this end, this paper introduces a new text modeling scheme based on an associative relational network, which represents texts as a directional weighted word-based graph. The conducted experiments validate the proposal in terms of both objective (TC efficiency) and subjective (perceived synthetic speech quality) evaluation criteria.

Positional and confidence voting-based consensus functions for fuzzy cluster ensembles

Consensus clustering, i.e. the task of combining the outcomes of several clustering systems into a single partition, has lately attracted the attention of researchers in the unsupervised classification field, as it allows the creation of clustering committees that can be applied with multiple interesting purposes, such as knowledge reuse or distributed clustering. However, little attention has been paid to the development of algorithms, known as consensus functions, especially designed for consolidating the outcomes of multiple fuzzy (or soft) clustering systems into a single fuzzy partition-despite the fact that fuzzy clustering is far more informative than its crisp counterpart, as it provides information regarding the degree of association between objects and clusters that can be helpful for deriving richer descriptive data models. For this reason, this paper presents a set of fuzzy consensus functions capable of creating soft consensus partitions by fusing a collection of fuzzy clusterings. Our proposals base clustering combination on a cluster disambiguation process followed by the application of positional and confidence voting techniques. The modular design of these algorithms makes it possible to sequence their constituting steps in different manners, which allows to derive versions of the proposed consensus functions optimized from a computational standpoint. The proposed consensus functions have been evaluated in terms of the quality of the consensus partitions they deliver and in terms of their running time on multiple benchmark data sets. A comparison against several representative state-of-the-art consensus functions reveals that our proposals constitute an appealing alternative for conducting fuzzy consensus clustering, as they are capable of yielding high quality consensus partitions at a low computational cost.

DYNAMAP – Development of low cost sensors networks for real time noise mapping

Abstract The Environmental Noise Directive (END) requires that regular updating of noise maps is implemented every five years to check and report about the changes occurred during the reference period. The updating process is usually achieved using a standardized approach, consisting in collating and processing information through acoustic models to produce the updated maps. This procedure is time consuming and costly, and has a significant impact on the budget of the authorities responsible for providing the maps. Furthermore, END requires that simplified and easy-to-read noise maps are made available to inform the public about noise levels and actions to be undertaken by local and central authorities to reduce noise impacts. To make the updating of noisemaps easier and more cost effective, there is a need for integrated systems that incorporate real-time measurement and processing to assess the acoustic impact of noise sources. To that end, a dedicated project, named DYNAMAP (DYNamic Acoustic MAPping), has been proposed and co-financed in the framework of the LIFE 2013 program, with the aim to develop a dynamic noise mapping system able to detect and represent in real time the acoustic impact of road infrastructures. In this paper, after a comprehensive description of the project idea, objectives and expected results, the most important steps to achieve the ultimate goal are described.

BordaConsensus: a new consensus function for soft cluster ensembles

Consensus clustering is the task of deriving a single labeling by applying a consensus function on a cluster ensemble. This work introduces BordaConsensus, a new consensus function for soft cluster ensembles based on the Borda voting scheme. In contrast to classic, hard consensus functions that operate on labelings, our proposal considers cluster membership information, thus being able to tackle multiclass clustering problems. Initial small scale experiments reveal that, compared to state-of-the-art consensus functions, BordaConsensus constitutes a good performance vs. complexity trade-off.

QuickSpot: a video analytics solution for on-street vacant parking spot detection

Vehicles searching for a vacant parking spot on the street can amount to as much as 40 % of the traffic in certain city areas, thus largely affecting mobility in urban environments. For this reason, it would be desirable to create integrated smart traffic management systems capable of providing real-time information to drivers about the location of available vacant parking spots. A scalable solution would consist in exploiting the existing and widely-deployed video surveillance camera networks, which requires the development of computer vision algorithms for detecting vacant parking spots. Following this idea, this work introduces QuickSpot, a car-driven video analytics solution for on-street vacant parking spot detection designed as a motion detection, object tracking and visual recognition pipeline. One of the main features of QuickSpot is its simplified setup, as it can be trained on external databases to learn the appearances of the objects it is capable of recognizing (pedestrians and vehicles). To test its performance under different daytime lighting conditions, we have recorded, edited, annotated and made available to the research community the QuickSpotDB video database for the vacant parking spot detection problem. In the conducted experiments, we have evaluated the trade-off between the accuracy and the computational complexity of QuickSpot with an eye to its practical applicability. The results show that QuickSpot detects parking spot status with an average accuracy close to 99 % at a 1-second rate regardless of the illumination conditions, outperforming in an indirect comparison the other car-driven approaches reported in the literature.

Reliability in ICA-Based Text Classification

This paper introduces a novel approach for improving the reliability of ICA-based text classifiers, attempting to make the most of the independent components of the text data. In this framework, two issues are adressed: firstly, a relative relevance measure for category assignment is presented. And secondly, a reliability control process is included in the classifier, avoiding the classification of documents belonging to none of the categories defined during the training stage. The experiments have been conducted on a journalistic-style text corpus in Catalan, achieving encouraging results in terms of rejection accuracy. However, similar results are obtained when comparing the proposed relevance measure to the classic magnitude-based technique for category assignment.

Audio and video cues for geo-tagging online videos in the absence of metadata

Tagging videos with the geo-coordinates of the place where they were filmed (i.e. geo-tagging) enables indexing online multimedia repositories using geographical criteria. However, millions of non geo-tagged videos available online are invisible to the eyes of geo-oriented applications, which calls for the development of automatic techniques for estimating the location where a video was filmed. The most successful approaches to this problem largely rely on exploiting the textual metadata associated to the video, but it is not rare to encounter videos with no title, description nor tags. This work focuses on this adverse scenario and proposes a purely audiovisual approach to geo-tagging. Using a subset of the MediaEval 2011 Placing task data set, we evaluate the ability of several visual and acoustic features for estimating the videos location, and demonstrate that the optimally configured version of the proposed system outperforms the only audiovisual participant in the MediaEval 2011 Placing task.

Single-Cell Telomere-Length Quantification Couples Telomere Length to Meristem Activity and Stem Cell Development in Arabidopsis

Summary Telomeres are specialized nucleoprotein caps that protect chromosome ends assuring cell division. Single-cell telomere quantification in animals established a critical role for telomerase in stem cells, yet, in plants, telomere-length quantification has been reported only at the organ level. Here, a quantitative analysis of telomere length of single cells in Arabidopsis root apex uncovered a heterogeneous telomere-length distribution of different cell lineages showing the longest telomeres at the stem cells. The defects in meristem and stem cell renewal observed in tert mutants demonstrate that telomere lengthening by TERT sets a replicative limit in the root meristem. Conversely, the long telomeres of the columella cells and the premature stem cell differentiation plt1,2 mutants suggest that differentiation can prevent telomere erosion. Overall, our results indicate that telomere dynamics are coupled to meristem activity and continuous growth, disclosing a critical association between telomere length, stem cell function, and the extended lifespan of plants.

Geo-tagging online videos using semantic expansion and visual analysis

The association of geographical tags to multimedia resources enables browsing and searching online multimedia repositories using geographical criteria, but millions of already online but non geo-tagged videos and images remain invisible to the eyes of this type of systems. This situation calls for the development of automatic geo-tagging techniques capable of estimating the location where a video or image was taken. This paper presents a bimodal geo-tagging system for online videos based on extracting and expanding the geographical information contained in the textual metadata and on visual similarity criteria. The performance of the proposed system is evaluated on the MediaEval 2011 Placing task data set, and compared against the participants in that workshop.

ICA-based hierarchical text classification for multi-domain text-to-speech synthesis

In the framework of multi-domain text-to-speech synthesis, it is essential (i) to design a hierarchically structured database for allowing several domains in the same speech corpus and (ii) to include a text classification module that, at run time, assigns the input sentences to a domain or set of domains from the database. We present a hierarchical text classifier based on independent component analysis (ICA), which is capable of (i) organizing the contents of the corpus in a hierarchical manner and (ii) classifying the texts to be synthesized according to the learned structure. The document organization and classification performance of our ICA-based hierarchical classifier are evaluated in several encouraging experiments conducted on a journalistic-style text corpus for speech synthesis in Catalan.