Fundamental Study of Radiation Characteristics of Shouted Speech for Shouted Speech Detection Towards Acoustic-Based Security System
A speech processing system for ensuring safety and security, namely, acoustic-based security system is addressed. For such system, shouted speech detection is indispensable technique, because the humans will spontaneously shout under emergency conditions. We thus recently focus on shouted speech detection for acoustic-based security system. For study of shouted speech detection, enormous shouted speech is necessary. However, such speech database had not been published and designed in all over the world. To overcome this problem, we firstly design the shouted speech database consists of 6,000 words including shouted/natural speech. And, we then investigate the radiation characteristics of shouted speech for shouted speech detection and detect the shouted/normal speech with the radiation characteristics for shout. For designing shouted speech database, we employed 20 females and 20 males, and their speech were recorded with 50 shouted and 50 natural words per one subject using three microphones located at mouse-closed, 1.0m distance back and forth in soundproof room. With this database, we firstly investigated the radiation characteristics of shouted speech. As a result, we could confirm that the vowel of shouted speech were 20 dB higher than one of natural speech on sound pressure difference between two microphones located at 1.0m distance back and forth. Thus, we tried to detect the shouted/natural speech with sound pressure difference between two microphones located at back and forth. As a result, we achieved average 89% detection rate without conventional speech features such as MFCC (Mel Frequency Cepstrum Coefficient).
Hiroshi Kawano Masanori Morise Takanobu Nishiura Hiroaki Nanjo
Graduate School of Science and Engineering,Ritsumeikan University,Japan. Faculty of Science and Technology,Ryukoku University,Japan.
国际会议
The 10th Western Pacific Acoustics Conference(第十届西太平洋声学会议WESPAC X)
北京
英文
1-5
2009-09-21（万方平台首次上网日期，不代表论文的发表时间）