"ShATR-Web"

Supervised by M.Cooke

Abstract

Speech corpora are databases consisting of audio-data along with aligned text-tags or 'transcriptions '. There have been many speech corpora constructed by many research institutes around the world,but no harmonisation of these diverse databases has taken place.As such,there are now hundreds of corpora, most of which use their own proprietory database storage representation,along with their own custom- built query tools and query languages. The project's aim was to provide an intuitive search system for the ShATR multi-simultaneous speaker corpus,and additionally to provide a framework for a generic search system applicable to any speech corpus.As such research work was conducted using the ShATR corpus,which is considered to be the most complex type of corpus the system should have to deal with. Various existing and emerging technologies were investigated as to their suitability for use in the project,along with the research of current speech corpora and their search facilities.The goal was to create a single framework,with an intuitive interface,which could used to deliver many large speech corpora to researchers all over the world. A database structure was created into which other speech corpora could be inserted,and an intuitive web-based interface was built which allows complex searching of the corpora.In addition,the interface can profile the user and retain settings between sessions.