The Development of Infant Cry Identifier Based on Time-Frequency Domain Features

• Main Authors: Ming-Feng Chang, 張銘豐
• Other Authors: Chuang-Chien Chiu
• Format: Others
• Language: en_US
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/84976332596409892741

碩士 === 逢甲大學 === 自動控制工程所 === 92 === The main purpose of this study is to design an automatic recognition system that can detect the crying voice from babies as a control signal of the intelligent cradle. When the infant cry is identified by this system, the control signal can be sent to swing the cradle or generate some alert messages. How to identify the infant cry accurately is the key issue in this study. We extracted voice parameters including fundamental frequency, zero-crossing rate, short-time energy, and mel-frequency cepstrum coefficients (MFCC), then using pattern projection and template matching methods to identify the input voices. The test set of infant cries was compared with many non-crying voices using the proposed identification processes. The recognition rate of our proposed system is up to 97% by applying pattern projection method and 100% by template matching method. However, only 70% of recognition rate was obtained using pattern projection technique to distinguish the cries caused by hungry and injection pain.