archana Balyan / maharaja surajmal institute of technology
Speech databases are necessary for developing speech technology applications such as automatic speech recognition (ASR), Text- to- speech (TTS) synthesis systems and in the field of computational linguistics. Research on TTS synthesis requires high quality speech database. Hindi is spoken by majority of population in India; still a very little amount of work on creating Hindi speech database (HSDB) has been carried out and reported in literature. For Hindi language, it is difficult to find well annotated high quality speech resources. Here, we explore the existing group- delay technique on continuous speech read by a male speaker .This approach produces high quality speech segments at syllable level and helps to solve the issue of unavailability of syllable-like tokens. The database is composed of 150 hand labeled tokens covering 39 phonemes. The quality of speech segments is evaluated using Mean Opinion Score which is found to be 4.033.