Search

International audience; In this paper, we identified seven ecological network analysis (ENA) metrics that, in our opinion, have high potential to provide useful and practical information for environmental decision-makers and stakeholders. Measurement and quantification of the network indicators requires that an ecosystem level assessment is implemented. The ENA metrics convey the status of the ecological system state variables, and mostly, the flows and relations between the various nodes of the network. The seven metrics are: 1) Average Path Length (APL), 2) Finn Cycling Index (FCI), 3) Mean Trophic level (MTL), 4) Detritivory to Herbivory ratio (D:H), 5) Keystoneness, 6) Structural Information (SI), and 7) Flow-based Information indices. The procedure for calculating each metric is detailed along with a short evaluation of their potential assessment of environmental status.

We investigated and compared various algorithms in machine learning for anomaly assessment with different feature analyses on ultrasonic signals recorded by sensor networks. The following methods were used and compared in anomaly detection modeling: hidden Markov models (HMM), support vector machines (SVM), isolation forest (IF), and reconstruction autoencoders (AEC). They were trained exclusively on sensor signals of the intact state of structures commonly used in various industries, like aerospace and automotive. The signals obtained on artificially introduced damage states were used for performance evaluation. Anomaly assessment was evaluated and compared using various classifiers and feature analysis methods. We introduced novel methodologies for two processes. The first was the dataset preparation with anomalies. The second was the detection and damage severity assessment utilizing the intact object state exclusively. The experiments proved that robust anomaly detection is practically feasible. We were able to train accurate classifiers which had a considerable safety margin. Precise quantitative analysis of damage severity will also be possible when calibration data become available during exploitation or by using expert knowledge.

We present a method for extracting health indicators from piezoelectric sensors applied in the case of microfluidic valves. Convolutional autoencoders were used to train a model on the normal operating conditions and tested on signals of different valves. The results of the model performance evaluation, as well as, the qualitative presentation of the indicator plots for each tested component, showed that the used approach is capable of detecting features that correspond to increasing component degradation. The extracted health indicators are the prerequisite and input for reliable remaining useful life prediction.

In this paper, acoustic resonance testing on glass intravenous (IV) bottles is presented. Different machine learning methods were applied to distinguish acoustic observations of bottles with defects from the intact ones. Due to the very limited amount of available specimens, the question arises whether the deep learning methods can achieve similar or even better detection performance compared with traditional methods.

Knowledge representation and processing, learning and adjusting knowledge models, communication and interaction as well as problem solving are important skills of cognitive agents. But, in order to share the learned knowledge with other communication participants, the agent must pay attention to maintaining its own functionality when interacting with the physical environment. To comply with this requirement, the meaning of perceptions and the consequences of actions must be understood. This suggests that even the non-verbal exchange of information is based on the foundations of semiotics. In this work, we show that we can model non-verbal interactions with the same linguistic means as verbal communication. We demonstrate this by solving the bidirectional translation problem of symbolic sequences into semantics through the use of minimalist grammars. To verify this approach, we turn back to the well-known example of a cognitive mouse agent living in a maze world and model the interaction and behavioral control using database semantics in a fully deterministic setting. Finally, we propose a unifying perspective for non-verbal interaction and verbal communication as well as justify its relevance for Cognitive Infocommunications.

Speech-controlled user interfaces facilitate the operation of devices and household functions to laymen. State-of-the-art language technology scans the acoustically analyzed speech signal for relevant keywords that are subsequently inserted into semantic slots to interpret the user's intent. In order to develop proper cognitive information and communication technologies, simple slot-filling should be replaced by utterance meaning transducers (UMT) that are based on semantic parsers and a \emph{mental lexicon}, comprising syntactic, phonetic and semantic features of the language under consideration. This lexicon must be acquired by a cognitive agent during interaction with its users. We outline a reinforcement learning algorithm for the acquisition of the syntactic morphology and arithmetic semantics of English numerals, based on minimalist grammar (MG), a recent computational implementation of generative linguistics. Number words are presented to the agent by a teacher in form of utterance meaning pairs (UMP) where the meanings are encoded as arithmetic terms from a suitable term algebra. Since MG encodes universal linguistic competence through inference rules, thereby separating innate linguistic knowledge from the contingently acquired lexicon, our approach unifies generative grammar and reinforcement learning, hence potentially resolving the still pending Chomsky-Skinner controversy., 13 pages, 1 figure

Abobi SM, Mildenberger TK, Kolding J, Wolff M. 2019. Assessing the exploitation status of main fisheries resources in Ghana’s reservoirs based on reconstructed catches and a length-based bootstrapping stock assessment method. Lake Reserv Manage. 35:415–434. The cichlid species Oreochromis niloticus, Sarotherondon galilaeus, and Coptodon zillii, which are among the most exploited resources in the small-scale fisheries of the Tono, Bontanga, and Golinga reservoirs in northern Ghana, were assessed based on length frequency samples. Growth, mortality, exploitation status, stock size, and relative yield per recruit reference points were determined using bootstrapping fish stock assessment (BFSA), a novel framework that allows for the estimation of uncertainties around the life-history parameters and reference levels (e.g., L∞, K, and F0.1). The results suggest that the 3 species studied are heavily exploited in all 3 reservoirs, but with no alarming signs of overexploitation. The fishing effort at Golinga is comparatively low as a result of insignificant fishing during the agriculture season, which relates to low exploitation rates. Sarotherondon galilaeus and C. zillii have the highest and lowest biomass (t/km2) respectively in all the 3 reservoirs. The small shallow reservoir (Golinga) has the highest biomass of the target resources per unit area. According to a second assessment approach, based on length-based indicators, all species at Bontanga and O. niloticus and S. galilaeus populations at Golinga have spawning stock biomasses below 40% of the unfished biomass. This points to a situation of a possible ongoing recruitment overfishing of those species in the 2 reservoirs and suggests that a further increase in fishing effort should be prevented. Further monitoring of these fisheries will be needed for the improvement of assessments and thus management advice.

In future power systems, electrical storage will be the key technology for balancing feed-in fluctuations. With increasing share of renewables and reduction of system inertia, the focus of research expands towards short-term grid dynamics and collective phenomena. Against this backdrop, Kuramoto-like power grids have been established as a sound mathematical modeling framework bridging between the simplified models from nonlinear dynamics and the more detailed models used in electrical engineering. However, they have a blind spot concerning grid components, which cannot be modeled by oscillator equations, and hence do not allow to investigate storage-related issues from scratch. We remove this shortcoming by bringing together Kuramoto-like and algebraic load-flow equations. This is a substantial extension of the current Kuramoto framework with arbitrary grid components. Based on this concept, we provide a solid starting point for the integration of flexible storage units enabling to address current problems like smart storage control, optimal siting and rough cost estimations. For demonstration purpose, we here consider a wind power application with realistic feed-in conditions. We show how to implement basic control strategies from electrical engineering, give insights into their potential with respect to frequency quality improvement and point out their limitations by maximum capacity and finite-time response., Comment: 12 pages, 6 figures

Database design of real-world scenarios requires complex data structures in order to adequately model complex real-world objects. Complex data structures can be constructed by a recursive use of elementary data types and data type constructors. The mathematics behind quantum mechanics provides us an interesting theory combining concepts from linear algebra, probability calculus, and logic. In order to make the mathematics of quantum mechanics available for database structures and states we develop a mapping of concepts from type theory of databases to the mathematics of quantum mechanics.

We present a model, inspired by quantum field theory, of the so-called inner stage of technical cognitive agents. The inner stage represents all knowledge of the agent. It allows for planning of actions and for higher cognitive functions like coping and fantasy. By the example of a cognitive mouse agent living in a maze wold, we discuss learning, action planning, and attention in a fully deterministic setting and assuming a totally observable world. We explain the relevance of our approach to cognitive infocommunications.

Non-linear equations describing the time evolution of frequencies and voltages in power grids exhibit fixed points of stable grid operation. The dynamical behaviour after perturbations around these fixed points can be used to characterise the stability of the grid. We investigate both probabilities of return to a fixed point and times needed for this return after perturbation of single nodes. Our analysis is based on an IEEE test grid and the second-order swing equations for voltage phase angles $\theta_j$ at nodes $j$ in the synchronous machine model. The perturbations cover all possible changes $\Delta\theta$ of voltage angles and a wide range of frequency deviations in a range $\Delta f=\pm1$~Hz around the common frequency $\omega=2\pi f=\dot\theta_j$ in a synchronous fixed point state. Extensive numerical calculations are carried out to determine, for all node pairs $(j,k)$, the return times $t_{jk}(\Delta\theta,\Delta \omega)$ of node $k$ after a perturbation of node $j$. We find that for strong perturbations of some nodes, the grid does not return to its synchronous state. If returning to the fixed point, the times needed for the return are strongly different for different disturbed nodes and can reach values up to 20 seconds and more. When homogenising transmission line and node properties, the grid always returns to a synchronous state for the considered perturbations, and the longest return times have a value of about 4 seconds for all nodes. The neglect of reactances between points of power generation (internal nodes) and injection (terminal nodes) leads to an underestimation of return probabilities., Comment: 19 pages, 9 figures

We developed a hardware-based cognitive user interface to help inexperienced and little technology-affine people to get easy access to smart home devices. The interface is able to interact (via speech, gestures, or touchscreen) with the user. By learning from the user’s behavior, it can adapt to each individual. In contrast to most commercial products, our solution keeps all data required for operation internally and is connected to other UCUI devices only via an encrypted wireless network. By design, no data ever leave the system to file servers of third-party service providers. In this way, we ensure the privacy protection of the user.

We present a Matlab toolbox, called “FockBox”, handling Fock spaces and objects associated with Fock spaces: scalars, ket and bra vectors, and linear operators. We give brief application examples from computational linguistics, semantics processing, and quantum logic, demonstrating the use of the toolbox.

We present a miniaturized universal hardware module for acoustic pattern recognition in various types of multichannel sensor signals. The module implements configurable signal analysis (signal transforms, filter banks, statistical transforms) and a GMM-HMM recognizer. The main hardware components are a XC7A75T FPGA performing almost all the computations, a TMS320C6746 digital signal processor organizing the data flow and executing the automata search (decoding), 3×1 Gibit DDR2 memory and 4 Gibit flash memory. The module can be plugged into a host device and be connected to signal acquisition hardware. It can process input signals with up to approx. 250 kHz bandwidth for continuous input and up to approx. 3 MHz for burst input.

We present a miniaturized universal hardware module for acoustic pattern recognition in various types of multi-channel sensor signals. The module implements a configurable signal analysis (signal transforms, filter banks, statistical transforms) and GMM-HMM recognizer. The main hardware components are a XC7A75T FPGA performing almost all the computations, a TMS320C6746 digital signal processor organizing the data flow and executing the automata search (decoding), 3×1 Gibit DDR2 memory and 4 Gibit flash memory. The module can be plugged into a host device and be connected to signal acquisition hardware. It can process input signals with up to approx. 250 kHz bandwidth for continuous input and up to approx. 3 MHz for burst input.

The design of a cognitive agent requires a behaviour control of actions and observations for exploring an unknown world. Typically, observations are influenced by a certain degree of randomness which can be modeled as probabilities. In our scenario we let a mouse explore a maze with walls, boundaries and a random portal. All observations are stored and managed in a so-called inner stage. As a decision problem, we want to be able to plan actions and to predict their resulting observations. In our approach we develop models of the inner stage based on concepts of probabilistic databases and their mapping to denormalized density matrices which are known from quantum mechanics. Density matrices provide a compact representation of the powerful but unwieldy many-world-semantics. We show that density matrices make the many-world-semantics feasible and are well suited to model the inner stage. We propose algorithms for learning and predicting action results.

1. The R package TropFishR is a new analysis toolbox compiling single-species stock assessment methods specifically designed for data-limited fisheries analysis using length-frequency data. 2. It includes methods for (i) estimating biological stock characteristics such as growth and mortality parameters, (ii) exploring technical aspects of the fisheries (e.g. exploitation rate and selectivity characteristics), (iii) assessingsize and composition of a fish stock bymeans of virtual population analysis (VPA), and (iv) assessing stock status with yield prediction and production models.3. This paper introduces the package and demonstrates the functionality of a selection of its core methods.4. TropFishR modernises traditional stock assessment methods by easing application and development and by combining it with advanced statistical approaches

Current technical cognitive systems cannot react on unforeseen situations. In particular, they fail if their repertoire of actions is insufficient to solve a problem or to overcome a barrier. In this paper we propose a novel ‘coping apparatus’ for cognitive agents which is capable of conceiving entirely new actions, simulating their prospective effect in the world and learning from their actual effect. The apparatus also provides the agent with ‘fantasy’ which allows to imagine unseen world states and to develop a strategy how to try to get there.

State-of-the-art speech synthesis systems achieve a high overall quality. However, synthesized speech still lacks naturalness. To produce more natural and colloquial synthetic speech, our research focuses on integration of effects present in spontaneous speech. Conventional speech synthesis systems do not consider the probability of a word in its context. Recent investigations on corpora of natural speech showed that words that are very likely to occur in a given context are pronounced less accurately and faster than improbable ones. In this paper three approaches are introduced to model this effect found in spontaneous speech. The first algorithm changes the speaking rate directly by shortening or lengthening the syllables of a word depending on the language model probability of that word. Since probable words are not only pronounced faster but also less accurately this approach was extended by selecting appropriate pronunciation variants of a word according to the language model probability. This second algorithm changes the local speaking rate indirectly by controlling the grapheme-phoneme conversion. In a third stage, a pronunciation sequence model was used to select the appropriate variants according to their sequence probability. In listening experiments test participants were asked to rate the synthesized speech in the categories colloquial impression and naturalness. Our approaches achieved a significant improvement in the category colloquial impression. However, no significantly higher naturalness could be observed. The observed effects will be discussed in detail.

This paper presents an analysis-by-synthesis approach for acoustic model adaptation. Using artificial speech data for speech recognition systems adaptation, has the potential to address the problem of data sparseness, to avoid speech recordings in real conditions and to provide the capability of performing large number of development cycles for Automatic Speech Recognition (ASR) systems in shorter time. The proposed adaptation framework uses unified ASR and synthesis system to produce artificial adaptation speech signals. In order to confirm the usability of the proposed approach, several experiments were performed where the artificial speech data was coded-decoded by different speech and waveform coders and the acoustic model used for synthesis was adapted for each coder. The recognition results show that the proposed method could be used successfully in the process of speech recognition systems performance assessment and improvement, not only for coded speech effects evaluation and adaptation, but also for other environment conditions.

This paper presents a cross-language development method for speech recognition and synthesis applications for Macedonian language. Unified system for speech recognition and synthesis trained on German language data was used for acoustic model bootstrapping and adaptation. Both knowledge-based and data-driven approaches for source and target language phoneme mapping were used for initial transcription and labeling of small amount of recorded speech. The recognition experiments on the source language acoustic model with target language dataset showed significant recognition performance degradation. Acceptable performance was achieved after Maximum a posteriori (MAP) model adaptation with limited amount of target language data, allowing suitable use for small to medium vocabulary speech recognition applications. The same unified system was used again to train new separate acoustic model for HMM based synthesis. Qualitative analysis showed, despite the low quality of the available recordings and sub-optimal phoneme mapping, that HMM synthesis produces perceptually good and intelligible synthetic speech.

This paper describes an approach to intelligent signal processing. First we propose a general signal model which applies to speech, music, biological, and technical signals. We formulate this model mathematically using a unification of hidden Markov models and finite state machines. Then we name tasks for intelligent signal processing systems and derive a hierarchical architecture which is capable of solving them. We show the close relationship of our approach to cognitive dynamic systems. Finally we give a number of application examples.

Obesity is a growing healthcare challenge in present days. Objective automated methods of food intake monitoring are necessary to face this challenge in future. A method for non-invasive monitoring of human food intake behavior by the evaluation of chewing and swallowing sounds has been developed. A wearable food intake sensor has been created by integrating in-ear microphone and a reference microphone in a hearing aid case. A concept for food intake monitoring requiring low computational cost is presented. After the detection of food intake activity periods, signal recognition algorithms based on Hidden Markov Models distinguish several types of food based on the sound properties of their chewing sounds. Algorithms are developed using manual labeled records of the food intake sounds of 40 participants.

In this paper we describe the compression of diphone inventories used by the acoustic synthesis of a concatenative synthesis system. The inventory compression is based on a codebook drawn from the Gaussian mean vectors of phoneme HMMs. There are two encoding/synthesis schemes, a speaker dependent and a speaker independent one. The advantage of the latter is the potential common use of the HM-models by a recognizer and a synthesizer. We describe the steps to encode the inventories as well as the acoustic synthesis using them. Using the proposed method a diphone inventory with 1175 units can be compressed down to 19 kB. We will show that the synthesis quality with HMM-encoded inventories matches the quality of synthesis with AMR- or SPEEX-encoded inventories at noticeably smaller inventory sizes.

In this paper we propose a universal strategy for the automatic interpretation of sensor signals. We focus on acoustic signals. However, any time series may be used. We assume that changes in an object's state cause a typical and reproducible change in the characteristics of the acquired sensor signal. In such cases we can train pattern recognizers basing on Hidden-Markov-Models or Support Vector Machines with data recordings of different object states and use these classifiers to assess the state of identical or similar objects. Our approach assumes that the sensor signals consist of elementary signal events and some kind of syntax defining their temporal relation (much like a musical score defines the temporal relation between notes). It is capable of automatically determining both, the elementary events and their syntax from the training data. We present experimental results from seven different applications from the fields of non-destructive testing, bio and music signal processing.

We propose an automatic region growing technique for the segmentation of the cerebral cortex and white matter in MRI data. Our method exploits general anatomical knowledge and uses an iterative multi resolution scheme for the estimation of intensity distributions to compensate for artifacts within the data. We present a comparison to segmentation results created by the neuroimaging software Brainvoyager QX and show advantages of our approach based on a qualitative and quantitative evaluation.

Pitch marking is a major task in speech processing. Thus, an accurate detection of pitch marks (PM) is required. In this paper, we propose a hybrid method for pitch marking that combines outputs of two different speech signal based pitch marking algorithms (PMA). We use a finite state machine (FSM) to represent and combine the pitch marks. The hybrid PMA is implemented in four stages: preprocessing, alignment, selection and postprocessing. In the alignment stage, the preprocessed pitch marks are shifted to a local minimum of the speech signal and the confidence score for every pitch mark is calculated. The confidence scores are used as transition weights for the FSM. The PMA outputs are combined into a single sequence of pitch marks. The more accurate pitch marks with the highest confidence score are chosen in the selection stage. A PM reference database contains 10 minutes speech including manually adjusted PM. The evaluation results indicate that the proposed hybrid method outperforms the single PMAs but also other current state-of-the-art algorithms which have been evaluated on a second reference database containing 44 speakers.

In this preliminary study we investigate the application of statistical classifiers for structural health monitoring of materials commonly used in airplanes. Our approach is based on the propagation of ultrasonic guided waves through materials like aluminum or carbon fiber reinforced plastic (CFRP). When the material gets damaged, the sound propagation changes. There are two ways of detecting these changes: we can use a physical model of the wave propagation or we can use a statistical approach. In this paper we focus on the latter. We present results using classifiers based on Hidden Markov Models (HMM) and Support Vector Machines (SVM). We compare these results to acoustic travel time tomography as a representative of the physical model based methods.

During the last few years, a framework for the development of algorithms for speech analysis and synthesis was implemented. The algorithms are connected to common databases on the different levels of a hierarchical structure. This framework which is called UASR (Unified Approach for Speech Synthesis and Recognition) and some related experiments and applications are described. Special focus is directed to the suitability of the system for processing nonverbal signals. This part is related to the analysis methods which are addressed in the COST 2102 initiative now. A potential application field in interaction research is discussed.

The institutional effect is used to explain and predict MBA students ' decisions to adopt e-learning in their studies. A composite model including four constructs, namely, perceived usefulness, perceived ease of use, sense of popularity and convenience, were formed and tested in the empirical study. This study found that "normatively appropriate " and "taken-for- granted structure" of institutions were the most important principles for adult students when making e- learning decisions.

Decision making and classification methods are very important in the field of structural health monitoring and life cycle prediction. We want to introduce an approach basing on sequence classifiers which can be used to several applications without any explicit knowledge of structures. To illustrate the concept we explain the method by means of a special example. So we can demonstrate our approach detailed, but although not too abstract.

This paper reports on a study of applying an HMM-based labeler along with a tailored feature extraction to Korotkoff sounds. These sounds can be heard through a stethoscope during the auscultatory blood pressure measurement usually done at medical practices. While this method works well when the patient is at rest, interfering noise from muscles and joints cause major problems when the subject is doing any activities like sports or fitness exercises. We propose a signal processing and classification method to overcome these difficulties and present first promising results.

In previous works (see for instance S. Werner et al. (2004)) we introduced different duration control methods in speech synthesis. The most outstanding approach is to control the grapheme to phoneme conversion (and thus indirectly control the speaking rate) by selecting (reduced) pronunciation variants according to a pronunciation variant sequence model. Listeners would only accept long synthesized utterances if the listening effort is nearly the same as the one when listening to natural speech. To evaluate the quality of the variant synthesis compared to the canonical one (as the state-of-the-art system), we performed a listening test with two different synthesis systems. The variant synthesis applying a pronunciation variant sequence model achieved a significant lower listening effort and a higher overall rate (MOS) compared to the canonical synthesis. We also show that the listening effort can act as a quality parameter for a speech sample. The rating for the listening effort is correlated with the rating of the naturalness and intelligibility of synthesized speech sample

In the past, several approaches have been proposed for voice conversion in TTS systems. Mostly, conversion is done by modification of the spectral properties and pitch to match a certain target voice. This conversion causes distortions that deteriorate the quality of the synthesized speech. In this paper we investigate a very simple and straightforward method for voice conversion. It generates a new voice from the source speaker instead of generating a certain target speaker's voice. For application in TTS systems it is often sufficient to synthesize new voices that sound sufficiently different to be distinguishable from each other. This is done by applying a spectral warping technique that is commonly used for speaker normalization in speech recognition systems called vocal tract length normalization (VTLN). Due to the low requirements of resources this method is especially suited for embedded systems.

Integration of pronunciation modeling into speech synthesis makes synthetic speech more natural and colloquial. Pronunciation variation as one observable effect in spontaneous speech is a step towards spontaneous speech synthesis. In the previous works (see Proc. ICASSP, vol.1, p.417-20, Orlando, FL, USA, 2002 and Proc. ICASSP, Hong Kong, PR China, Apr. 2003) we introduced different duration control methods in speech synthesis. These methods are based on the observation that words, which are very likely to occur in a given context are pronounced faster and less accurately than improbable ones (see D. Jurafsky et al., Proc. ICASSP, vol.2, p.801-4, Salt Lake City, USA, 2001). Therefore we use the probability of a word in its context to either control directly the local speaking rate, or to select appropriate pronunciation variants in order to change the local speaking rate. Extending these methods by a pronunciation sequence model, we involve knowledge about how well two subsequent variants fit together. Using this proposed algorithm we could further improve the natural and colloquial listening impressions.

In this paper we propose a voice activation method based on prosodic keyword verification. In current voice activation systems features like the fundamental frequency contour have not been considered so far. Normally a continuous listening word spotter is used to detect a certain predefined keyword. We conducted an experiment which shows that people emphasize this keyword when they address a recognizer. To capture the prosodic information we trained an HMM on the fundamental frequency and energy contour of the keyword. The prosodic model is used to verify the keyword hypotheses of a phonetic recognizer. We investigated the performance of the prosodic model to distinguish between the same keyword spoken in command and non-command phrases. The introduction of the prosodic information significantly reduced the false alarm rate whereas the detection rate was only slightly degraded.

Non-speech acoustic signals are widely used as the input of systems for non-destructive testing. In this rapidly growing field, the signals have an increasing complexity leading to the fact that powerful models are required. Methods like DTW and HMM, which are established in speech recognition, have been successfully used but are not sufficient in all cases. We propose the application of generalized structured Markov graphs (SMG). We describe a task independent structure learning technique which automatically adapts the models to the structure of the test signals. We demonstrate that our solution outperforms hand-tuned HMM structures in terms of class discrimination by two case studies using data from real applications.

We investigate the deployment possibilities of speech enabled services in a Web based e-learning environment. The integration of speech technology is realized with a client/server architecture. Therefore, the services speech synthesis, speech recognition, and speaker verification are installed at a central SpeechServer. The client uses a Java applet (SpeechApplet), which is integrated in an HTML page. It takes the user's input (e.g. speech or text input) and activates the according service at the SpeechServer. The SpeechApplet is easy to integrate into existing Web pages and aims at a simple JavaScript interface for communication between the Web page and the applet. In this paper we introduce this system, explain different modules, and discuss first evaluation results of these technologies.

State of the art speech synthesis systems achieve a high overall quality. However, the synthesized speech still lacks naturalness. To make speech synthesis more natural and colloquial we are trying to integrate effects that are observable in spontaneous speech. In a previous paper we introduced a new approach for duration control in speech synthesis that uses the probability of a word in its context to control the local speaking rate within the utterance. This idea is based on the observation that words that are very likely to occur in a given context are pronounced faster than improbable ones. Since probable words are not only pronounced faster but also less accurate we extend this approach by selecting appropriate pronunciation variants to realize the change in the local speaking rate.

Speech synthesis systems basing on concatenation of natural speech segments achieve a high quality in terms of naturalness and intelligibility. However, in many applications such systems are not easy to apply because of the huge demand for storage capacity. Speech synthesis systems based on HMMs could be an alternative to concatenative speech synthesis systems but do not yet achieve the quality needed for use in applications. In one of our research projects we investigate the possibility of combining speech synthesis and speech recognition to a unified system using the same databases and similar algorithms for synthesis and recognition. In this context we examine the suitability of stochastic Markov graphs instead of HMMs to improve the performance of such synthesis systems. The paper describes the training procedure we used to train the SMGs, explains the synthesis process and introduces an algorithm for state selection and state duration modeling. We focus particularly on issues which arise using SMGs instead of HMMs.

Presented is a trainable data-driven method of deriving numerals from number strings. The concept is based on learning a graph model for numeral grammars and a graph search which is capable of extracting numeral words from the grammar according to given number strings. Because this method separates code and data it is universal and applicable to every language. No a priori knowledge about formal descriptions of numeral grammars is required. Due to the underlying graph concept, the algorithm is able to automatically generate grammars for number to numeral translations with a minimal effort. The only required input information is a set of pairs of number strings and appropriate numerals. This method is useful for an easy implementation of knowledge bases for further languages in the framework of a multi-lingual speech synthesis system.

In the framework of the German Verbmobil project we developed a procedure for the automatic, data-driven generation of pronunciation dictionaries for speech recognition systems. In most recognizers, only simple dictionaries containing the canonical pronunciation form are used. They represent the correct pronunciation, but in most cases the canonical pronunciation does not match the actual realization of the word. To solve this problem we chose an approach to derive pronunciation variants automatically from a speech database. The training algorithm is based on a canonical dictionary which is compiled into a graph representation in a first stage. Pronunciation variants are then learned from a training sample consisting of speech signal and its orthographic transcription. The authors focus on the experimental results obtained in the Verbmobil framework and introduce methods to evaluate pronunciation dictionaries generated by the training procedure.

Education in spoken language engineering may be supported very effectively by Web based methods. The paper describes at first some problems which arise if Web based tutorials are to feature online speech input and interactive usage. A framework is defined which is aimed at handling these problems. The architecture and the interfaces are explained in more detail. Finally, we describe a tutorial on speech synthesis which was designed using this framework.

In the framework of the German Verbmobil project we developed a procedure for the automatic, data-driven generation of pronunciation dictionaries for speech recognition systems. In most recognizers only simple dictionaries containing the canonical pronunciation form are used. They represent the correct pronunciation, but in most cases the canonical pronunciation does not match the actual realization of the word. To solve this problem we chose an approach to derive pronunciation variants automatically from a speech database. The training algorithm bases on a canonical dictionary which is compiled into a graph representation in a first stage. Pronunciation variants are then learned from a training sample consisting of speech signal and its orthographic transcription. In this paper we will focus on the experimental results obtained in the Verbmobil framework and introduce methods to evaluate pronunciation dictionaries generated by the training procedure.