Survey on speech emotion recognition: Features, classification schemes, and databases

http://www.ffri.hr/~ibrdar/komunikacija/seminari/Ververidis,%202006%20-%20Emotional%20speech%20recognition.pdf

Emotional speech recognition: Resources, features, and methods

On the occasion of the centenary of the discovery of the Lombard effect, we review the literature on noise-dependent regulation of vocal amplitude in humans and other animals The article addresses the scientific and the biological history of the Lombard effect: first, it sketches the evolution of the study of the Lombard effect, and second it reflects on the biological evolution of the effect itself By comparing the findings from anurans, birds and mammals, we try to trace back the phylogenetic origins of this basic vocal mechanism for acoustic communication in noise The current evidence suggests two alternative parsimonious hypotheses: either the Lombard effect is the outcome of a convergent evolution in birds and mammals or it may be a synapomorphy of all amniotes If the latter is true, then the Lombard effect would have evolved to maintain vocal communication in the presence of noise more than 300 million years ago

The evolution of the Lombard effect: 100 years of psychoacoustic research

http://felix.syntheticspeech.de/publications/ParalingusiticChallenge2010_CSL.pdf

Paralinguistics in speech and language-State-of-the-art and the challenge

This review investigates the effects of psychological stress on the human body measured through biosignals. When a potentially threatening stimulus is perceived, a cascade of physiological processes occurs mobilizing the body and nervous system to confront the imminent threat and ensure effective adaptation. Biosignals that can be measured reliably in relation to such stressors include physiological (EEG, ECG, EDA, EMG) and physical measures (respiratory rate, speech, skin temperature, pupil size, eye activity). A fundamental objective in this area of psychophysiological research is to establish reliable biosignal indices that reveal the underlying physiological mechanisms of the stress response. Motivated by the lack of comprehensive guidelines on the relationship between the multitude of biosignal features used in the literature and their corresponding behaviour during stress, in this paper, the impact of stress to multiple bodily responses is surveyed. Emphasis is put on the efficiency, robustness and consistency of biosignal data features across the current state of knowledge in stress detection. It is also explored multimodal biosignal analysis and modelling methods for deriving accurate stress correlates. This paper aims to provide a comprehensive review on biosignal patterns caused during stress conditions and reliable practical guidelines towards more efficient detection of stress.

Review on psychological stress detection using biosignals

Robust speech recognition systems must address variations due to perceptually induced stress in order to maintain acceptable levels of performance in adverse conditions. One approach for addressing these variations is to utilize front-end stress classification to direct a stress dependent recognition algorithm which separately models each speech production domain. This study proposes a new approach which combines stress classification and speech recognition functions into one algorithm. This is accomplished by generalizing the one-dimensional (1-D) hidden Markov model to an N-channel hidden Markov model (N-channel HMM). Here, each stressed speech production style under consideration is allocated a dimension in the N-channel HMM to model each perceptually induced stress condition. It is shown that this formulation better integrates perceptually induced stress effects for stress independent recognition. This is due to the sub-phoneme (state level) stress classification that is implicitly performed by the algorithm. The proposed N-channel stress independent HMM method is compared to a previously established one-channel stress dependent isolated word recognition system yielding a 73.8% reduction in error rate. In addition, an 82.7% reduction in error rate is observed compared to the common one-channel neutral trained recognition approach.

/pdf/n-channel-hidden-markov-models-for-combined-stressed-speech-2hap5o0795.pdf

B.D. Womack

Papers

N-channel hidden Markov models for combined stressed speech classification and recognition