Structural Analysis of Hindi Phonetics and A Method for Extraction of Phonetically Rich Sentences from a Very Large Hindi Text Corpus

摘要

Automatic speech recognition (ASR) and Text to speech (TTS) are two prominentarea of research in human computer interaction nowadays. A set of phoneticallyrich sentences is in a matter of importance in order to develop these twointeractive modules of HCI. Essentially, the set of phonetically rich sentenceshas to cover all possible phone units distributed uniformly. Selecting such aset from a big corpus with maintaining phonetic characteristic based similarityis still a challenging problem. The major objective of this paper is to devisea criteria in order to select a set of sentences encompassing all phoneticaspects of a corpus with size as minimum as possible. First, this paperpresents a statistical analysis of Hindi phonetics by observing the structuralcharacteristics. Further a two stage algorithm is proposed to extractphonetically rich sentences with a high variety of triphones from the EMILLEHindi corpus. The algorithm consists of a distance measuring criteria to selecta sentence in order to improve the triphone distribution. Moreover, a specialpreprocessing method is proposed to score each triphone in terms of inverseprobability in order to fasten the algorithm. The results show that theapproach efficiently build uniformly distributed phonetically-rich corpus withoptimum number of sentences.