Pronunciation Modeling for conversational Austrian German

Project Type: Student Project, Master/Diploma Thesis
Project Status: Open

Description

One serious problem for Automatic Speech Recognition (ASR) of spontanous, conversational speech is that words may have various pronunciations as compared to read speech, where they are usually always read and pronounced the same way. The word HABEN, for instance, is mostly pronounced as hab@n  in read speech but can be pronounced as hab@n , habn, ham, hobn, hom  in spontaneous Austrian German conversations. Thus, in order to improve ASR recognition for conversational speech, pronunciation modeling is necessary.

The aim of this thesis is to incorporate the pronunciation lexicon, which  already has been developed during the project "Crosslayer pronunciation modelling for conversational speech" into the pronunciation modelling component of our KALDI speech recognizer, and to test its impact on different speech styles.

 

Your Tasks

  • Review of literature

  • Incorporate existing pronunciation lexicon into pronunciation modeling component of an ASR system

  • System tests on different speech styles

  • Documentation

Your Requirements

  • Motivation and interest in the topic

  • Speech communication background

  • Recommended: Speech Communication 2 (VO, SS 2016)

  • Recommended: Spoken language in human and human computer dialogue (VU, SS 2016 or in parallel to writing the thesis)