Development of Punjabi Text To Speech Application for Mobile Phone

Abstract

In recent years speech synthesis and recognition has achieved lot of success in the field of Information Technology. Speech is means of communication between different people. This thesis is highly motivated to help visually impaired people, partially sighted people and to promote Punjabi language. Text to Speech Synthesis broadly works in two parts front end and back end. The front end takes input in the form of raw text (i.e input sting) and produce output a symbolic linguistic representation. Further, the back end takes the symbolic linguistic representation as input and outputs the synthesized waveform. This report discusses about the different phases in the text analysis and various techniques used for determining the symbolic linguistic representation. The accuracy for determining the linguistic representation plays important role in text to speech synthesis. Speech Synthesizers are used for synthesis of the speech. There are different types of techniques available to synthesise a speech signal. Concatenative synthesis basically selects the units (phoneme, syllable and words) and synthesize into the waveform. Formant Synthesis uses a set of parameters like frequency, amplitude and pitch to synthesize a speech signal. Articulatory synthesis uses set of articulators (like tongue, jaw, teeth) and generate speech. But concatenative synthesis synthesizes a natural speech. This thesis work is concerned with the development of a mobile text to speech synthesis application of Punjabi language. The goal of this project is to utter Punjabi speech when input is provided in English language. The research work is carried out with the aim to maximize the system intelligibility and naturalness. To achieve the desired aim some techniques and customized rules are followed to resolve the ambiguities present in input text i.e resolution of issues like Initials, numerals and titles etc. During mapping of English text to Punjabi Phonemes many issues arose and maximum of them are resolved to some extent. An efficiency of 82% is achieved while resolving the mapping. Finally, for generation of output speech, concatenative technique is used. It produces the desired output speech with some limitations.