1. Advancements in whispered speech detection for interactive/speech systems
Chi Zhang, John H. L. Hansen · 2018
While the challenge of non-neutral speech is attracting more research attention in recent years, the specific speaking style of “whisper” is of particular interest in both speech technology and human-to-human communication, especially for hearing-impaired individuals. Whisper is a common mode of natural speech communication, which is an alternative speech production mode, used in public circumstances to protect content privacy or identity, or potentially a common speaking style for individuals with speech production limitations. The spectral structure of whispered speech is dramatically different from that of neutral speech. This profound difference between whisper and neutral speech is caused by the fundamental differences in productionmechanisms of these two speech modes. Given these differences, the performance of current speech-processing systems, which are designed to process neutral speech, degrades significantly. In this chapter, challenges of speech processing for whispered speech will be reviewed. Next, the requirement of front-end whispered speech detection for speech-processing applications will be introduced. An overview of current advancements for both unsupervised/model-free whispered speech detection and model-based distant-whispered speech detection will be presented. Finally, a brief summary and conclusion is followed by a discussion on directions for future work.