Fundamental Study of Automatic Gender Detection from Shout for Acoustic-Based Security System

摘要

A speech processing system for ensuring safety and security, namely, acoustic-based security system is addressed. For such a system, shouting detection is significant. Moreover, it is also significant to detect the speaker's properties such as gender and age since the security officers can take appropriate actions according to such properties. Based on the background, in this paper, fundamental study of gender detection from shouting is described. As for gender detection, we confirmed that F0 distributions of male and female tend to be close in shout. Here, we investigate a detection method without F0. Specifically, we adopt the Mel-Frequency Cepstrum Coefficient (MFCC) feature, which models spectral envelope, and train Gaussian mixture model (GMM) with shouting and natural speech of male and female. Shouting and gender detections are performed based onGMMscores. The accuracy of gender detection from shouting is 99.2% for male and 97.6% for female, which is sufficient for security system.