Deterioration of intelligibility in whispered Japanese speech

It is still not fully revealed how people convey and understand whispered speech without critical deterioration of intelligibility. Typical questions are how the height of sound is produced and how voiced consonants are produced without vibration of vocal folds. In this paper, we have investigated the deterioration of intelligibility of whispered speech and tried to reveal the mechanism to convey voiced consonants by whispering. Through an experiment using 110 monosyllables whispered by a speaker and ten evaluators of the intelligibility of those syllables, we have obtained 80% of intelligibility of normally uttered ones. In addition, a clear difference was observed in the spectrograms between whispered syllables and vocoded ones made from normally phonated speech by replacing the sound source with a Gaussian noise. It suggests that a larger energy balance in high-frequency emphases the nature of unvoiced consonants and, reversely, a larger energy balance in low-frequency contributes to make unvoiced consonants sound like voiced ones.