Ludek Müller

Extension of HVS semantic parser by allowing left-right branching

The hidden vector state (HVS) parser is a popular method for semantic parsing. It is used in the language understanding module of the statistical based spoken dialog system. This paper presents an extension of the HVS semantic parser. It enables the parser to generate broader class of semantic trees. This modification can be used to improve the performance of the parser by generating not only the right-branching trees (like original HVS parser) but also limited left-branching trees and their combinations. The extension retains simplicity and properties of the original HVS parser. We tested the method on Czech human-human train timetable corpus. The modified HVS parser yields statistically significant improvement. The accuracy of the system increased from 50.4% to 58.3% absolutely.

Automatic recognition of bird individuals on an open set using as-is recordings

Abstract The most common method used to determine the identity of an individual bird is the capture-mark-recapture technique. The method has several major disadvantages, e.g. some species are difficult to capture/recapture and the capturing process itself may cause significant stress in animals leading even to injuries of more vulnerable species. Some studies introduce systems based on methods used for human identification. An automatic system for recognition of bird individuals (ASRBI) described in this article is based on a Gaussian mixture model (GMM) and a universal background model (GMM-UBM) method extended by an advanced voice activity detection (VAD) algorithm. It is focused on recognizing the bird individuals on an open set, i.e. any number of unknown birds may appear anytime during the identification process as is common in nature. The introduced ASRBI processes the recordings just as if they were recorded by an ornithologist: with durations from seconds to minutes, containing noise and unwanted sounds, as well as masking of the singer, etc. Thanks to the VAD algorithm, the proposed system is fully automatic, no manual pre-processing of recordings is needed, neither by cutting off the songs nor syllables. The overall achieved identification accuracy is 78.5%, the lowest 60.3% and the highest 95.7%. In total, 90% of all experiments reach at least 70% accuracy. The result suggests the application of the GMM-UBM with VAD is feasible for individual identification on the open set processing real-life recordings. The described method is capable of reducing both the time consumption and human intervention in animal monitoring projects.

On complementarity of state-of-the-art speaker recognition systems

In this paper recent methods used in the task of Speaker Recognition (SR) are reviewed and their complementarity is analysed. At first, methods based on Supervectors (SVs) related to Gaussian Mixture Models (GMMs) and Support Vector Machines (SVMs) used as a discriminative model are described along with the Nuisance Attribute Projection (NAP). NAP was proposed to suppress undesirable influences of high channel variabilities between several sessions of a speaker. Next, recent methods focusing on the extraction of so called i-vectors (low dimensional representations of GMM based SVs) are discussed. The space in which i-vectors lie is denoted the Total Variability Space (TVS) since it contains both between-speaker and session/channel variabilities. Once i-vectors have been extracted a Probabilistic Linear Discriminant Analysis (PLDA) model is trained in the TVS. In the training phase of PLDA the TVS is decomposed to a channel and a speaker subspace, hence each i-vector is supposed to be composed from a speaker identity component and a channel component. The complementarity of PLDA and SVM based modelling techniques is examined utilizing the linear logistic regression as a fusion tool used to combine the verification scores of individual systems leading to significant reductions in error rates of the SR system. The results are presented on the NIST SRE 2008 and NIST SRE 2010 corpora.

Trends in gene expression changes during adipogenesis in human adipose derived mesenchymal stem cells under dichlorodiphenyldichloroethylene exposure

BackgroundsExposure to lipophilic environmental pollutants has been explored as a risk factor of development of diabetes mellitus in obese. Adipose tissue is a reservoir of bioaccumulative lipophilic contaminants including p,pʹ-dichlorodiphenyldichloroethylene (DDE). Our aim was to analyze the effect of DDE (in concentrations 0.1 μM, 1 μM, and 10 μM) on adipocyte differentiation and insulin signalling pathway on in vitro adipogenic model of human adipose derived mesenchymal stem cells (hADMSC).MethodsThe effect of DDE was monitored by analysis of expression (RT qPCR, Western blotting) of genes involved in adipocyte differentiation and insulin signalling pathway including lipid metabolism on days 0, 4, 10, 21, 28 of differentiation.ResultsThe main observation was significant increase of INSR, LIPE, FASN, SREBP1, OCT4 and AKT2 expression under influence of DDE. We did not record any increase of the active form of Akt.ConclusionOur findings suggest that DDE exposure changes the differentiation of adipocytes, enhances the lipid metabolism and so may play a role in the development of obesity and metabolic diseases by affecting the insulin signalling pathway. The influence of DDE seems to be similar to the effect of insulin itself. However, further studies to elucidate the mechanism of action of DDE are necessary.

Mashona Mole-Rat Automatic Individual Identification Based on the Mating Calls

In this study was tested mole-rat vocalization for presence of diverse individually distinctive features. An automatic system based on the GMM-UBM was used for individual recognition. The system distinguishes the recordings of the five mole-rats females. The overall achieved identification accuracy is 76.7%, the lowest 59.2%, and the highest 83.5%. The overall percentage is thus high enough to prove that the mating calls of the Mashona mole-rat can carry information about mole-rat individuality. Our results showed that studied vocal signals in the Mashona mole-rats are individually specific which indicates the possibility of individual vocal recognition in this species.

The Influence of a Filter Shape in Telephone-Based Recognition Module Using PLP Parameterization

This paper concerns an influence of a filter shape and a benefit of the Hertz-Bark transformation to the word error rate (WER) obtained in a telephone-based speech recognition application working with the Perceptually-based Linear Predictive (PLP) parameterization. Five various shapes of filters (rectangular, narrow and wide trapezium, triangular and the classical PLP filter shape [1]) were compared and an effect of a nonlinear frequency transformation between Hertz and generalized Bark axis was explored. Experiments with 100 speakers and with the vocabulary size of 475 words were performed. During all experiments only the zero-gram language model was used to see better an influence of particular variables to changes of the WER.

Various approaches in classification of technical processes

Within nuclear reactor diagnostics, it is necessary to automatically recognise the operations which have occurred These operations (called transient events) have to be recognised from the measured data stream. Because it is not possible to determine the boundaries between the transient events, the entire sequence of events has to be recognised. Also, the duration of the similar processes of the same class differs in different realisations. Several methods have been used to solve this problem. Three of them are discussed in this article: human recognition, a program based on a pattern recognition method (so-called Hidden Markov Models) and a program using pattern recognition for generation of hypotheses and a rule based expert system for hypotheses testing.