Gaussian Mixture Clustering and Language Adaptation for the Development of a New Language Speech Recognition System

The porting of a speech recognition system to a new language is usually a time-consuming and expensive process since it requires collecting, transcribing, and processing a large amount of language-specific training sentences. This work presents techniques for improved cross-language transfer of speech recognition systems to new target languages. Such techniques are particularly useful for target languages where minimal amounts of training data are available. We describe a novel method to produce a language-independent system by combining acoustic models from a number of source languages. This intermediate language-independent acoustic model is used to bootstrap a target-language system by applying language adaptation. For our experiments, we use acoustic models of seven source languages to develop a target Greek acoustic model. We show that our technique significantly outperforms a system trained from scratch when less than 8 h of read speech is available     

SQL extension for spatio-temporal data

An SQL extension is formalized for the management of spatio-temporal data, i.e. of spatial data that evolves with respect to time. The extension is dedicated to applications such as topography, cartography, and cadastral systems, hence it considers discrete changes both in space and in time. It is based on the rigid formalization of data types and of SQL constructs. Data types are defined in terms of time and spatial quanta. The SQL constructs are defined in terms of a kernel of few relational algebra operations, composed of the well-known operations of the 1NF model and of two more, Unfold and Fold. In conjunction with previous work, it enables the uniform management of 1NF structures that may contain not only spatio-temporal but also either purely temporal or purely spatial or conventional data. The syntax and semantics of the extension is fully consistent with the {SQL:2003} standard.     

Reverse nearest neighbors in large graphs

A reverse nearest neighbor (RNN) query returns the data objects that have a query point as their nearest neighbor (NN). Although such queries have been studied quite extensively in Euclidean spaces, there is no previous work in the context of large graphs. In this paper, we provide a fundamental lemma, which can be used to prune the search space while traversing the graph in search for RNN. Based on it, we develop two RNN methods; an eager algorithm that attempts to prune network nodes as soon as they are visited and a lazy technique that prunes the search space when a data point is discovered. We study retrieval of an arbitrary number k of reverse nearest neighbors, investigate the benefits of materialization, cover several query types, and deal with cases where the queries and the data objects reside on nodes or edges of the graph. The proposed techniques are evaluated in various practical scenarios involving spatial maps, computer networks, and the DBLP coauthorship graph.     

Indoor robot gardening: design and implementation

This paper describes the architecture and implementation of a distributed autonomous gardening system with applications in urban/indoor precision agriculture. The garden is a mesh network of robots and plants. The gardening robots are mobile manipulators with an eye-in-hand camera. They are capable of locating plants in the garden, watering them, and locating and grasping fruit. The plants are potted cherry tomatoes enhanced with sensors and computation to monitor their well-being (e.g. soil humidity, state of fruits) and with networking to communicate servicing requests to the robots. By embedding sensing, computation, and communication into the pots, task allocation in the system is de-centrally coordinated, which makes the system scalable and robust against the failure of a centralized agent. We describe the architecture of this system and present experimental results for navigation, object recognition, and manipulation as well as challenges that lie ahead toward autonomous precision agriculture with multi-robot teams.     

The <Emphasis Type="Italic">B</Emphasis><Superscript><Emphasis Type="Italic">dual</Emphasis></Superscript>-Tree: indexing moving objects by space filling curves in the dual space

Existing spatiotemporal indexes suffer from either large update cost or poor query performance, except for the B ^x -tree (the state-of-the-art), which consists of multiple B ⁺-trees indexing the 1D values transformed from the (multi-dimensional) moving objects based on a space filling curve (Hilbert, in particular). This curve, however, does not consider object velocities, and as a result, query processing with a B ^x -tree retrieves a large number of false hits, which seriously compromises its efficiency. It is natural to wonder “can we obtain better performance by capturing also the velocity information, using a Hilbert curve of a higher dimensionality?”. This paper provides a positive answer by developing the B ^dual -tree, a novel spatiotemporal access method leveraging pure relational methodology. We show, with theoretical evidence, that the B ^dual -tree indeed outperforms the B ^x -tree in most circum- stances. Furthermore, our technique can effectively answer progressive spatiotemporal queries, which are poorly supported by B ^x -trees.     

Probabilistic motion-compensated prediction in distributed video coding

Distributed video coding (DVC) constitutes an original coding framework to meet the stringent requirements imposed by uplink-oriented and low-power mobile video applications. The quality of the side information available to the decoder and the efficiency of the employed channel codes are primary factors determining the success of a DVC system. This contribution introduces two novel techniques for probabilistic motion compensation in order to generate side information at the Wyner-Ziv decoder. The employed DVC scheme uses a base layer, serving as a hash to facilitate overlapped block motion estimation at the decoder side. On top of the base layer, a supplementary Wyner-Ziv layer is coded in the DCT domain. Both proposed probabilistic motion compensation techniques are driven by the actual correlation channel statistics and reuse information contained in the hash. Experimental results report significant rate savings caused by the novel side information generation methods compared to previous techniques. Moreover, the compression performance of the presented DVC architecture, featuring the proposed side-information generation techniques, delivers state-of-the-art compression performance.     

Automatic speech recognition performance in different room acoustic environments with and without dereverberation preprocessing

The performance of recent dereverberation methods for reverberant speech preprocessing prior to Automatic Speech Recognition (ASR) is compared for an extensive range of room and source-receiver configurations. It is shown that room acoustic parameters such as the clarity (C50) and the definition (D50) correlate well with the ASR results. When available, such room acoustic parameters can provide insight into reverberant speech ASR performance and potential improvement via dereverberation preprocessing. It is also shown that the application of a recent dereverberation method based on perceptual modelling can be used in the above context and achieve significant Phone Recognition (PR) improvement, especially under highly reverberant conditions.     

Piecewise linear interval maps both expanding and contracting

The dynamics of piecewise linear interval maps is studied with two branches, one expanding and one contracting. It is proved that such a map either has a periodic attractor or it is eventually expanding. In the latter case there exists an absolutely continuous invariant measure.