Linear and nonlinear speech feature analysis for stress classification
Guojun Zhou, John H. L. Hansen, James F. Kaiser · 1998
There are many stressful environments which deteriorate the performance of speech recognition systems. Exam-ples include aircraft cockpits, 911 emergency telephone response, high workload task stress, or emotional situa-tions. To address this, we investigate a number of lin-ear and nonlinear features and processing methods for stressed speech classication. The linear features include properties of pitch, duration, intensity, glottal source, and the vocal tract spectrum. Nonlinear processing is based on our newly proposed Teager Energy Operator (TEO) speech feature which incorporates frequency domain criti-cal band lters and properties of the resulting TEO auto-correlation envelope. In this study, we employ a Bayesian hypothesis testing approach and a hidden Markov model (HMM) processor as classication methods. Evaluations focused on speech under loud, angry, and the Lombard eect 1 from the SUSAS database. Results using receiver operating characteristic (ROC) curves and EER (equal er-ror rate) based detection show that pitch is the best of the ve linear features for stress classication; while the new nonlinear TEO-based feature outperforms the best linear feature by +5.2%, with a reduction in classication rate variability from 8.66 to 3.90. 1