Acoustic Data Based Grapheme to Phoneme Conversion
Akshay A Deshpande · OhioLink ETD Center (Ohio Library and Information Network) · 2013
Grapheme to Phoneme conversion (G2P), also called Letter to Sound mapping (LTS) is defined as the task of learning the relationship between the written word and its phonetic transcription.It is a necessary part of Text to Speech systems and plays a vital role in handling Out of Vocabulary (OOV) words in Automatic Speech Recognition systems.G2P is a complex task, because for many languages, the correspondence between the orthography (spelling) and its phonetic transcription is not completely consistent.Over time, the techniques used to tackle this problem have evolved, from earlier rule based systems to the current more sophisticated machine learning approaches.All the existing systems for G2P rely on pronunciation dictionaries as their source of training data.In this thesis we explore the G2P problem from the perspectives of using speech data to aid an existing system and possibly improve its performance.In order to do that, we will look at some of the current G2P techniques and reproduce established results as a part of the background work.We then utilize a measure of closeness between graphemes and phonemes that formulates an acoustic score.We conduct experiments by combining our score with existing systems and analyze the results.The current results do not surpass the baseline system but point the way towards future innovations.iii Dedication This document is dedicated to my family.