Classification of speech under stress using target driven features
Speech Communication - Special issue on speech under stress
Speech Communication - Special issue on speech under stress
Methods for stress classification: nonlinear TEO and linear speech based features
ICASSP '99 Proceedings of the Acoustics, Speech, and Signal Processing, 1999. on 1999 IEEE International Conference - Volume 04
| Hi-index | 0.00 |
Emotion and stress/neutral detection based on an input audio stream has been a topic of interest in the literature with various applications. This paper reports on a preliminary study of stress/neutral detection based on naturalistic home environment recordings of children. One major motivation of the work is to add stress/neutral detection functionality into the LENA™ System [10]. The study started with an acted emotion database, and tested the acoustic feature of Mel-frequency cepstral coefficients and the Gaussian Mixture Model (GMM) for stress/neutral detection on this relatively simple database. The method was then applied to the adult speech segments automatically extracted from home recordings of children with the LENA System, achieving 72% accuracy for adult stress/neutral detection. The application of this new functionality to a large number of naturalistic home environment recordings of children reveals interesting and meaningful statistical differences among the families of typically developing children, language-delayed children, and children with Autism Spectrum Disorders (ASD). The result suggests the potential for stress/neutral detection, along with the LENA System, as an integrated solution for (i) quality assessment of the child language environment, (ii) monitoring language interventions for disordered children, or (iii) general psychological and behavioral research.