Automation of Pitch Accent Markings in Punjabi Speech

Abstract

Speech has been one of the major forms of human communication and a unique characteristic of the human species. Speech is produced by vibrations of the vocal cords. The use of prosodic knowledge in automatic speech recognition is a widely researched topic in recent years. To incorporate prosody knowledge into speech recognition by marking the variations in pitch is the main objective of this thesis. The rate of vibration of vocal cords is called fundamental frequency or pitch. Pitch marking is related to the instances of glottal closure. An accurate pitch marking directly influences the quality of speech signal. For pitch marking, voiced and unvoiced regions are separated because pitch marking is applicable only to the voiced parts of speech, no pitch can be detected in case of unvoiced speech. A GUI has also been developed for automatic prosody labeling of pitch with variation of pitch labels. This thesis is divided into five chapters. A brief outline of these chapters is given below. Chapter 1 includes the basic terminology and tools which one needs to be familiar with for prosody labeling of pitch. Chapter 2 includes literature survey. This chapter is divided into three parts. First part deals with the review of literature of basic fundamentals of speech. Second part deals with the review of literature of prosody labeling systems and the last part deals with the review of literature of pitch marking techniques in voiced speech segments. Chapter 3 discusses the problem statement and also includes the data collection for pitch marking. Total duration of data considered in this work is 32 minute and 13 seconds. A total of 40 files are taken from 10 different speakers. Each speaker has contributed 4 files. Chapter 4 focuses on the algorithm used for pitch marking. This chapter also includes the outputs of the algorithm and results are presented in different figures. It shows the output of algorithm on 40 files, manual pitch marking on 40 files by 4 users and the comparison between them. Chapter 5 presents the conclusion of the work done and the further scope of this work for increasing the accuracy.