Intensity feature for speech stress detection

Abstract

Suprasegmental features are fundamental properties of speech. They can improve not only the naturalness of synthesized speech, but the performance of machine speech recognition in voice controlled logistic systems. In linguistics, stress is the relative emphasis that may be given to certain syllables in a word, or to certain words in a phrase or sentence. The term is also used for similar patterns of phonetic prominence inside syllables. The stress placed on syllables within words is called word stress or lexical stress. The stress placed on words within sentences is called sentence stress or prosodic stress. Sentence and word stress are crucial prosodic features. One of the features usually used for stress detection is the energy of syllables, but the average energies of vowels are various. Energy of a stressed weak vowel can be lower than that of an unstressed strong vowel. We compare the amplitude of the actual vowel to that of its average to show the stressed or unstressed nature of the syllable. Average energies of vowels are obtained from a speech recognizer trained with voices of hundreds of speakers.