In this paper we explore the use of features derived from multiresolution analysis of speech and the Teager Energy Operator for classification of drivers' speech under stressed conditions. We apply this set of features to a database of short speech utterances to create userdependent discriminants of four stress categories. In addition we address the problem of choosing a suitable temporal scale for representing categorical differences of the data. This leads to two sets of modeling techniques. In the first approach, we model the dynamics of the feature set within the utterance with a family of dynamic classifiers. In the second approach, we model the mean value of the features across the utterance with a family of static classifiers. We report and compare classiftcation performances on the sparser and full dynamic representations for a set of four subjects.

Speech production variations due to perceptually induced stress contribute significantly to reduced speech processing performance. One approach that can improve the robustness of speech processing (e.g., recognition) algorithms against stress is to formulate an objective classification of speaker stress based upon the acoustic speech signal. In this paper, an overview of recent methods for stress classification is presented. First, we review traditional pitch-based methods for stress detection and classification. Second, neural network based stress classifiers with cepstral-based features, as well as wavelet-based classification algorithms are considered. The effect of stress on linear speech features is discussed, followed by the application of linear features and the Teager Energy Operator (TEO) based nonlinear features for effective stress classification. A new evaluation for stress classification and assessment is presented using a critical band frequency partition based TEO featur...

There are many stressful environments which deteriorate the performance of speech recognition systems. Examples include aircraft cockpits, 911 emergency telephone response, high workload task stress, or emotional situations. To address this, we investigate a number of linear and nonlinear features and processing methods for stressed speech classification. 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 critical band filters and properties of the resulting TEO autocorrelation envelope. In this study, we employ a Bayesian hypothesis testing approach and a hidden Markov model (HMM) processor as classification methods. Evaluations focused on speech under loud, angry, and the Lombard effect 1 from the SUSAS database. Results using receiver operating characteristic (ROC) curves and EER (equal ...

In this paper we explore the use of features derived from multiresolution analysis of speech and the Teager Energy Operator for classification of drivers' speech under stressed conditions. The potential stress categories are determined by driving speed and the frequency with which the driver has to solve a mental task while driving. We first use an unsupervised approach to gain some understanding as to whether the discrete stress categories form meaningful clusters in feature space, and use the clustering results to build a user-dependent recognition system which combines local discriminants of 4 discreet stress categories. Recognition results are reported for 4 subjects. 1 Introduction Much of the current effort on studying speech under stress has been aimed at detecting several stress conditions for improving the robustness of speech recognizers; typical categories of speech under stress have targeted perceptual (e.g. Lombard effect), psychological (e.g. timed tasks), as well as ph...

It is well known that the performance of speech recognition algorithms is greatly influenced by stressful conditions in which speech is produced. It is suggested that algorithms which are capable of classifying stress could be beneficial in the formulation of more effective speech recognition algorithms under stressful conditions. Recently, three nonlinear features derived from Teager Energy Operator (TEO) were proposed and shown to be effective for stress classification [1]. In this study, a new feature entitled TEO-CB-Auto-Env, is proposed which extends the work reported in [1]. The new feature is based on the normalized TEO autocorrelation envelope area employing critical band frequency analysis. The TEO-CBAuto -Env feature is shown to outperform the three previous TEO based features for stress classification. Furthermore, it also outperforms the traditional Mel-frequency cepstral coefficients (MFCC) by +4:7%, as well as being more consistent for the task of stress classification us...