Prominent word detection using pitch patterns

Prosody is an important aspect of spoken communication. Prominence is an important property of prosody. An algorithm for automatic prominent word detection is reported in this paper. We address the automatic detection of word prominence by examining the variation in pitch. Pitch is extracted from a correlation based approach, as they are robust to noise and therefore to telephone speech. From the experiments, it is shown that the pitch patterns can detect the prominent word in telephone speech also. The proposed automatic prominence detection can be used in automatic speech recognizers or as a tool to highlight prominent words in the speech corpus for unit selection in text-to-speech synthesis.