Narrow your search

Library

KU Leuven (2)

Hogeschool West-Vlaanderen (1)

KBR (1)

UAntwerpen (1)

UGent (1)

UNamur (1)


Resource type

book (1)

dissertation (1)


Language

English (2)


Year
From To Submit

2007 (2)

Listing 1 - 2 of 2
Sort by

Dissertation
Example based continuous speech recognition
Authors: ---
ISBN: 9789056828141 Year: 2007 Publisher: Leuven Katholieke Universiteit Leuven

Loading...
Export citation

Choose an application

Bookmark

Abstract

De voorbije 25 jaar werd het onderzoek naar automatische spraakherkenning gedomineerd door systemen gebaseerd op verborgen Markov ketens (HMMs). In deze dissertatie onderzoeken we een alternatief, waarbij de te herkennen spraak wordt vergeleken met een groot aantal referentievoorbeelden. Motivatie voor deze aanpak vinden we zowel in een analyse van de zwakheden van verborgen Markov ketens, als in een vergelijking met andere onderzoeksgebieden zoals menselijke spraakverwerking, fonologie en automatische spraaksynthese. We ontwikkelen een volledig voorbeeldgebaseerd systeem voor continue spraakherkenning, waarbij vooral aandacht wordt besteed aan voorbeeldgebaseerde akoestische modellering, aan een nieuw model voor a priori waarschijnlijkheden van de herkenningshypotheses en aan de ontwikkeling van een efficiënt zoekalgoritme. Op bepaalde taken evenaart of verbetert het nieuwe systeem de herkenningsgraad van state-of-the-art HMM systemen, hetzij met een significant grotere behoefte aan rekenkracht. Over the past 25 years, research in automatic speech recognition has been dominated by systems based on hidden Markov models (HMMs). In this dissertation, an example based alternative is investigated, where the input speech is compared with a large number of reference templates. We find ample motivation for our approach in a study of the weaknesses of hidden Markov models, and in a comparison with other research domains such as human speech recognition, phonology and automatic speech synthesis. We develop a complete example based system for continuous speech recognition, focusing on example based acoustic modelling, on a new model for the prior probability of recognition hypotheses and on the development of an efficient search algorithm. On a number of tasks, the new system performs as good as or better than state-of-the-art HMM recognisers, although using a significantly larger amount of computational resources. Systemen voor automatische spraakherkenning zoeken de best passende tekstuele transcriptie van de opgenomen spraak door verschillende kennisbronnen te combineren. Het taalmodel bevat de kennis over hoe zinnen zijn opgebouwd, zowel qua syntax als qua semantiek, terwijl het akoestische model beschrijft welke waarnemingen (via de microfoon) worden gedaan voor de verschillende klanken. Deze dissertatie behandelt een vernieuwende aanpak voor het akoestische model. De huidige standaardsystemen beschrijven het akoestische model met statistische modellen. In dit werk vergelijken we de opgenomen spraak rechtstreeks met een heel groot aantal referentievoorbeelden, waarvan zowel de correcte klankidentiteit als de contextuele informatie volledig gekend is. Op deze manier hopen we de informatie, die verloren gaat bij het bouwen van een statistisch model, nuttig te kunnen gebruiken. Deze aanpak houdt twee grote uitdagingen in. De grote conceptuele uitdaging is het vinden van manieren om alle aanwezige informatie ook nuttig te gebruiken. De grootste praktische uitdaging is het beperken van de benodigde rekenkracht. Beide uitdagingen worden uitvoerig besproken, en de aangereikte oplossingen zorgen voor een nieuw type spraakherkenner dat op beperkte taken reeds lichtjes beter presteert dan de beste bestaande systemen, zij het met een veel grotere behoefte aan rekenkracht en geheugengebruik. Automatic speech recognition systems find the most likely textual transcription of recorded speech by combining different knowledge sources. The language model contains both syntactic and semantic knowledge, while the acoustic model describes which observations (made using a microphone) correspond to the different elementary sounds. This dissertation discusses a novel method for acoustic modelling. Current state-of-the-art systems use statistical acoustic models. In this dissertation, we instead compare the input speech with a very large number of reference examples whose exact transcription and contextual information is known. This way, we hope to be able to advantageously use the information which is normally lost in statistical model building. This approach faces two main challenges. The major conceptual challenge is the search for possible ways in which all the available information can be used. The major practical challenge is the limitation of the required computational resources. Both challenges are discussed in detail, and the proposed solutions lead to a new type of speech recognition system which already slightly outperforms traditional systems on limited tasks, although needing significantly more computational resources.

The definitive ANTLR reference : building domain-specific languages
Author:
ISBN: 9780978739256 0978739256 9780978739249 Year: 2007 Publisher: Lewisville, Tex.: Pragmatic,

Loading...
Export citation

Choose an application

Bookmark

Abstract

ANTLR is a parser generator: a program that generates code to translate a specified input language into a nice, tidy data structure. You might think that parser generators are only used to build compilers. But in fact, programmers usually use parser generators to build translators and interpreters for domain-specific languages such as proprietary data formats, common network protocols, text processing languages, and domain-specific programming languages. Domain-specific languages are important to software development because they represent a more natural, high fidelity, robust, and maintainable means of encoding a problem than simply writing software in a general-purpose language. For example, NASA uses domain-specific command languages for space missions to improve reliability, reduce risk, reduce cost, and increase the speed of development. Even the first Apollo guidance control computer from the 1960s used a domain-specific language that supported vector computations. This book is the definitive guide to using the completely rebuilt ANTLR v3 and describes all features in detail, including the amazing new LL(*) parsing technology, tree construction facilities, StringTemplate code generation template engine, and sophisticated ANTLRWorks GUI development environment. You'll learn all about ANTLR grammar syntax, resolving grammar ambiguities, parser fault tolerance and error reporting, embedding actions to interpret or translate languages, building intermediate-form trees, extracting information from trees, generating source code, and how to use the ANTLR Java API.

Keywords

Programming --- Parsing (Computer grammar) --- Programming languages (Electronic computers) --- Syntax --- Project management. --- Syntax. --- Parsing (Computer grammar). --- 681.3*D34 --- 681.3*D32 --- 681.3*F42 --- 681.3*I27 --- Processors: code generation; compilers; interpreters; optimization; parsing; preprocessors; run-time environments; translator writing systems and compilergenerators (Programming languages) --- language classifications: applicative languages; data-flow languages; design languages; extensible languages; macro and assembly languages; nonprocedural languages; specialized application and very high-level languages (Programminglanguages) --- Grammars and other rewriting systems: decision problems; grammar types; parallel rewriting systems; parsing; thue systems (Mathematical logic and formal languages)--See also {681.3*D31} --- Natural language processing: language generation; language models; language parsing and understanding; machine translation; speech recognition and under-standing; text analysis (Artificial intelligence) --- 681.3*I27 Natural language processing: language generation; language models; language parsing and understanding; machine translation; speech recognition and under-standing; text analysis (Artificial intelligence) --- 681.3*F42 Grammars and other rewriting systems: decision problems; grammar types; parallel rewriting systems; parsing; thue systems (Mathematical logic and formal languages)--See also {681.3*D31} --- 681.3*D32 language classifications: applicative languages; data-flow languages; design languages; extensible languages; macro and assembly languages; nonprocedural languages; specialized application and very high-level languages (Programminglanguages) --- 681.3*D34 Processors: code generation; compilers; interpreters; optimization; parsing; preprocessors; run-time environments; translator writing systems and compilergenerators (Programming languages) --- Parsers (Computer grammar) --- Computational linguistics --- Formal languages --- Generative grammar --- Grammar, Comparative and general --- Programming languages (Electronic computers) - Syntax --- Programmeren. --- ANTLR.

Listing 1 - 2 of 2
Sort by