Detection of shouted speech in noise: Human and machine

摘要: High vocal effort has characteristic acoustic effects on speech. This study focuses the utilization of this information by human listeners and a machine-based detection system in task detecting shouted speech presence noise. Both female male speakers read Finnish sentences using normal voice controlled conditions, with sound pressure level recorded. The material was artificially corrupted noise supplemented pure performance statistically evaluated listening test, where subjects labeled noisy samples according to whether shouting heard or not. A Bayesian constructed evaluated. Its compared against that listeners, substituting different spectrum analysis methods feature extraction stage. Using features capable taking into account spectral fine structure (i.e., fundamental frequency its harmonics), machine reached humans even noisiest conditions. In detected significantly better than especially making smaller increase for shouting.