Supervised learning using Hidden Markov Models has been used to train acoustic models for automatic speech recognition for several years. Typically clean transcriptions form the basis for this training regimen. However, results have shown that using sources of readily available transcriptions, which can be erroneous at times (e.g., closed captions) do not degrade the performance significantly. This work analyzes the effects of mislabeled data on recognition accuracy. For this purpose, the training is performed using manually corrupted training data and the results are observed on three different databases: TIDigits, Alphadigits and SwitchBoard. For Alphadigits, with 16% of data mislabeled, the performance of the system degrades by 12% relative to the baseline results. For a complex task like SWITCHBOARD, at 16% mislabeled training data, the performance of the system degrades by 8.5% relative to the baseline results. The training process is more robust to mislabeled data because the Gaussian mixtures that are used to model the underlying distribution tend to cluster around the majority of the correct data. The outliers (incorrect data) do not contribute significantly to the reestimation process.

Projektet har gjorts på uppdrag av Ölandsfärjan AB för att tillverka elscheman för huvudinstrumenttavlan ombord på deras fartyg M/S Solsund. Anledningen till detta var att det inte fanns någon dokumentation om huvudinstrumenttavlan och detta försvårade arbetet med felsökning och underhåll. Därför önskades elscheman som kunde användas för att se skåpets funktioner och uppbyggnad.Uppdraget var att skapa ett layoutschema som beskriver indikeringslampor och knappars fysiska positioner för huvudinstrumenttavlans framsida. Det skulle även skapa ett kretsschema över förbrukardelen och ett kretsschema för kraftförsörjningsdelen som beskriver funktioner och utgående matningar. Kraftförsörjningen består av två generatorer och möjlighet till landanslutning. Det gjordes även separata kretsschema över generatorernas övervakningsskåp som innehöll effektbrytarnas utlösningsskydd. För samtliga scheman gjordes en gemensam apparatlista. För att skapa elscheman så påbörjades uppdraget med detektering av huvudinstrumenttavlan. Arbetsmetoden för detekteringen utfördes genom att visuellt följa eller resistansmäta kablarna. Avslutningsvis ritades den slutliga produkten i datorprogrammet Easy EL. / This project was commissioned by Ölandsfärjan AB to manufacture wiring diagrams for the main switchboard on board their ship M/S Solsund. The reason for this was that there was no documentation found on the main switchboard and this complicated the process of troubleshooting and maintenance. Therefore, Ölandsfärjan AB required wiring diagrams, which could be used to see cabinet functions and the contexture of the wiring.  The task was to create a layout diagram that would be able to provide a description for the physical positions of indicator lights and buttons on the front of the main switchboard. The task was also to create a circuit diagram of the consuming section and a circuit diagram of power supply section of the main switchboard, the purpose for this was to reveal the functions and output feeds. The power supply consists of two generators and also the opportunity for shore connection. Separate circuit diagrams of the generators' control boxes were also constructed for the circuit breaker tripping protection devices. For all diagrams a common device list was made. The project began to create wiring diagrams of the main switchboard. The work methods for the detection were performed by visually monitoring the cables or measuring the cables' resistance. Finally, the diagrams were made in the computer program Easy EL. / Main Switchboard Drawing

This project describes the design of the system for remote temperature monitoring of stored goods for horse-breeding. The main task of this system is the early alert before the fire developed to spontaneous combustion and continuous monitoring of this commodities .. The control panel communicates with several autonomous temperature sensors. The measured values are displayed on the display and transmitted over the Ethernet network to a server where monitored temperatures and their graphical waveforms can be viewed.

This thesis describes design of a power and data network in an intelligent house. The aim is to create a project of wiring system for an intelligent house, which is being controlled by system Foxtrot. At introduction, in the second chapter, the division of buildings in terms of heat loss amount is described. Furthermore, the chapter deals with low-powered buildings and describes division of low-powered buildings in terms of heating needs. The third chapter deals with the system Foxtrot and describes individual parts of this system. Designed object is presented in the last chapter. Moreover, the chapter focusses on design of a whole project describing wiring system using the elements of system Foxtrot.

This thesis describes the reconstruction network 10 kV HV networks in a teaching hospital in the transition to a new 22 kV voltage level according to valid regulations. The paper describes the current state of technological equipment and its gradual replacement with new technology at 22 kV voltage level in each power units so as not to limit the operation of the hospital. For this purpose, the new backup system for rotating spare resources. The thesis deals with new input distribution, which will serve as a transfer point between the customer and distributor of electricity as well as advance the existing technological equipment.

The main goal of this work is to make complete project of a computer network for Less & Timber s.r.o. The project will contain all information, from the cable traces, connection units, sockets and their count.

The goal of this thesis is to advance the use of recurrent neural network language models (RNNLMs) for large vocabulary continuous speech recognition (LVCSR). RNNLMs are currently state-of-the-art and shown to consistently reduce the word error rates (WERs) of LVCSR tasks when compared to other language models. In this thesis we propose various advances to RNNLMs. The advances are: improved learning procedures for RNNLMs, enhancing the context, and adaptation of RNNLMs. We learned better parameters by a novel pre-training approach and enhanced the context using prosody and syntactic features. We present a pre-training method for RNNLMs, in which the output weights of a feed-forward neural network language model (NNLM) are shared with the RNNLM. This is accomplished by first fine-tuning the weights of the NNLM, which are then used to initialise the output weights of an RNNLM with the same number of hidden units. To investigate the effectiveness of the proposed pre-training method, we have carried out text-based experiments on the Penn Treebank Wall Street Journal data, and ASR experiments on the TED lectures data. Across the experiments, we observe small but significant improvements in perplexity (PPL) and ASR WER. Next, we present unsupervised adaptation of RNNLMs. We adapted the RNNLMs to a target domain (topic or genre or television programme (show)) at test time using ASR transcripts from first pass recognition. We investigated two approaches to adapt the RNNLMs. In the first approach the forward propagating hidden activations are scaled - learning hidden unit contributions (LHUC). In the second approach we adapt all parameters of RNNLM.We evaluated the adapted RNNLMs by showing the WERs on multi genre broadcast speech data. We observe small (on an average 0.1% absolute) but significant improvements in WER compared to a strong unadapted RNNLM model. Finally, we present the context-enhancement of RNNLMs using prosody and syntactic features. The prosody features were computed from the acoustics of the context words and the syntactic features were from the surface form of the words in the context. We trained the RNNLMs with word duration, pause duration, final phone duration, syllable duration, syllable F0, part-of-speech tag and Combinatory Categorial Grammar (CCG) supertag features. The proposed context-enhanced RNNLMs were evaluated by reporting PPL and WER on two speech recognition tasks, Switchboard and TED lectures. We observed substantial improvements in PPL (5% to 15% relative) and small but significant improvements in WER (0.1% to 0.5% absolute).

The following master thesis is devoted to the design of distribution systems of LV (Low Voltage) and HV (High Voltage) voltage levels. The aim of the thesis is to create a complete project documentation for the construction of a new distribution transformer station (22kV/0.4kV) situated directly in the city of Blansko along with a transformer capable of performance of up to 630kVA. Subsequently a design of cable routes of LV and HV is included in the project documentation. The first part of the thesis introduces the fundamental issues and solutions regarding the design of electrical equipment related to the LV and HV voltage levels. Thereafter it deals with proper legislation, technical regulations and related standards. The second part is dealing with project documentation. More precisely analyzing each step in the process and thoroughly going through individual parts of the project documentation. Other focus of the thesis is software resources and tools which are essential for drawing documentation. Then follows a selection of appropriate electrical equipment and budget evaluation. The major portion of the work is a design of distribution transformer station and cable lines of LV and HV. Including dimensioning of the LV and HV switchboards, earthing system and fuse protection system. The essence of all this work serves as a foundation for complete design documentation.

The aim of this diploma thesis is to compile a demonstration panel as an example of the functions of intelligent electrical installation KNX. One of the advantages of this system bus is the ability to use devices from multiple manufacturers. A total of 20 devices from 12 different manufacturers are used in this work. The work begins with an explanation of the principle on which the KNX system bus works, followed by a description of the equipment used and the electrical connection of the switchboard according to the wiring diagram. Next, how the whole panel was revived is described. The next chapter contains a detailed description of device programming in ETS, ie setting parameters for individual devices and assigning group addresses. The last chapter deals with visualization and remote control. It explains how a communication channel was created to connect the panel to a server from FlowBox. Using the web interface on this server, a visualization was created, which can be used to remotely control the entire panel from anywhere.

This diploma thesis deals with rebuild of manual drill-mill machine BF20L into CNC mill machine, 3D model of that and eventual control using PLC. All axes are powered by stepper motors, with encoders, connected to control cards and industrial computer from company Beckhoff. For this control is made full switchboard, supplemented by electrical wiring diagram. Program control along with visualization is made in software TwinCAT 3. Drill-mill machine can be controlled manually or automatically by program.