Obtaining Pitch Contour Model for Parts of Speech (such as noun, verb) for Kannada Language

The objective of this paper is to obtain pitch contour model for noun, verb in kannada language. One of the major problems in text-to-speech synthesis systems consists in the automatic generation of natural prosody. Prosody is described in terms of fundamental frequency contours (for voiced portion of speech), duration of speech segments (mainly phonemes or syllables) and the amplitude of the speech segments. Speech synthesizers having correct prosodic information can produce very high quality synthetic speech. In unit based speech synthesis we use large database of recorded speech. The recorded speech is divided into words and sentences. After doing manual correction with the help of PRAAT, extract pitch values of words or sentences. Normalize all the words to 100 points and for sentences to 1000 points. Then using cftool plot try out the best fit. In unit selection synthesis, the units selected may not have the required pitch characteristics. There needs to be pitch modification for naturalness of output speech. To modify the pitch, it is necessary to know the pitch variations for different types of sentences. Hence we are analysing the natural Kannada speech to obtain mathematical model for noun and verb in kannada language which would fit for any declarative and interrogative sentences.