Generation of prosody in speech synthesis using large speech data-base