SpinRCP is an integrated development environment for the Spin model checker used for verifying the correctnesses of concurrent and distributed systems. Using SpinRCP, it is easy to enter, edit, examine, and check the syntax of models which represent the systems to be analyzed, to check redundancies in models, to specify the required properties of models, to graphically represent processes and never claims derived from specified properties in the form of nondeterministic final state machines, to enter or select various simulation and verification parameters, to perform verification and random, guided, or interactive simulations and to transform a Spin simulation trail into a standard Message Sequence Chart (MSC). SpinRCP is implemented in Java as an Eclipse Rich Client Platform (RCP) product.
B.03 Paper at an international scientific conference
COBISS.SI-ID: 18075414Translations in statistical machine translation (SMT) are generated on the basis of statistical models, the parameters of which are derived from the analysis of aligned bilingual text corpora. Different models’ parameters provide various translations, which are evaluated by the BiLingual Evaluation Understudy (BLEU) metric. The problem of finding a suitable translation can be regarded as an optimization problem and some optimization can be done using the decoder itself – the optimization of models parameters. The main goal of this paper was to build SMT systems for the language pairings English-Slovenian and Slovenian-English, and improve their translation qualities using a global optimization algorithm – Differential Evolution (DE) algorithm. Experiments were performed using English and Slovenian JRC-ACQUIS Multilingual Parallel Corpora. The results show improvement in the translation quality.
B.03 Paper at an international scientific conference
COBISS.SI-ID: 18361366This article describes a large-scale evaluation of the use of Statistical Machine Translation for professional subtitling. The work was carried out within the FP7 EU-funded project SUMAT and involved two rounds of evaluation: a quality evaluation and a measure of productivity gain/loss. We present the SMT systems built for the project and the corpora they were trained on, which combine professionally created and crowd-sourced data. Evaluation goals, methodology and results are presented for the eleven translation pairs that were evaluated by professional subtitlers. Overall, a majority of the machine translated subtitles received good quality ratings. The results were also positive in terms of productivity, with a global gain approaching 40%.
B.03 Paper at an international scientific conference
COBISS.SI-ID: 17878806The aim of the research work is to search for common guidelines for the future development of speech databases for less resourced languages in order to make them the most useful for both main fields of their use, linguistic research and speech technologies. We compare two standards for creating speech databases, one followed when developing the Slovene speech database for automatic speech recognition – BNSI Broadcast News, the other followed when developing the Slovene reference speech corpus GOS, and outline possible common guidelines for future work. We also present an add-on for the GOS corpus, which enables its usage for automatic speech recognition.
B.03 Paper at an international scientific conference
COBISS.SI-ID: 17960982This thesis focuses on quality assessment of multimodal services in contemporary telecommunication systems. It addresses quality degradations which affect user experience. Depending on their origin, they can be categorized as source or network impairments. Their impact can be measured with subjective or objective methods. Since multimodal services can be bi-directional systems, it is necessary to have control over input and output modalities of the system. This leads to intermodal influences between the modalities as a consequence of human perception. Furthermore, the users' focus on Regions-of-Interest (ROI) gives degradations in those particular regions greater impact on the overall quality, which we can use for differentiated quality assessment. The aim of this thesis is to propose a model for quality assessment of multimodal services and develop the concept of the quality evaluator, which takes the above mentioned facts into account. Therefore, the thesis is divided into three sections. In the first section, the impact of quality degradations on the input modality is determined. In the second, a suitable multimodal database comprising HD recordings is established. This section also presents subjective and objective assessment of output modality, where subjective mean opinion score (subMOS) and objective mean opinion score (objMOS) were conducted. Based on the results, a new model of multimodal quality assessment is proposed. The last section addresses differential quality evaluation based on ROI. As part of the evaluation of the effect of quality degradations on the input modality, a voice-driven IVR service with a built-in speech recognition module (ASR) is analyzed. Assessment begins by measuring objMOS values of the samples from the SpeechDat(II) database. Samples were degraded by transcoding and packet loss. There were substantial differences between the speech codecs used, even when the exact same codec was used with different configurations. Generally, deterioration was greater for codecs with lower bandwidth. The voice signal degraded to such an extent that it was necessary to use a more robust modality, i.e. DTMF dialing. After an analysis of the results, a classifier of input modality based on the Gaussian Mixture Models (GMM) was proposed. When training the classifier, different classification parameters were conducted. Test phase confirmed the successful operation of the classifier regarding the input modality with various packet loss scenarios. For the purpose of assessing the impact of degradations on the quality of output modality, a specifically designed multimodal database was established. It comprised audio (AAC at 48 kbps), video (H.264/AVC at a resolution of 1920x1080 pixels) and combined audio and video clips for a total of 240 samples, used in various packet loss scenarios. After that, subjective tests with 20 subjects were conducted, which gave reference data for objective quality assessment. Objective quality was measured separately for audio and video modalities. To assess the audio modality, standardized PESQ speech quality metric was used, and to assess the video modality NQM video metric was applied. Then, using the regression method, a linear model for evaluating the quality of multimodal services was proposed, which takes into account the type of modality, type of scene, amount of degradation and unimodal objMOS scores. Correlation yields 0.892. The differential quality evaluation consists of two stages. First, a ROI face detector was used, based on the Viola-Jones object detection algorithm with weak Haar-like feature-based cascade classifiers. Then, using good detection results, an analysis of the optimization possibilities due to differential quality assessment of visual modality is presented. This investigation proposed evaluating the quality of ROI regions with a more complex algorithm (NQM) since those regions have higher visual attention, and using a simpler quality metric (PSNR) for the
D.09 Tutoring for postgraduate students
COBISS.SI-ID: 277694720