A non-uniform subband approach to speech-based cognitive load classification

Le, Phu Ngoc; Ambikairajah, Eliathamby; Choi, Eric H. C.; Epps, Julien

doi:10.1109/icics.2009.5397501

Cited by 10 publications

(23 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The most significant band from the point of view of cognitive load classification, with regards to these two features is around 750-1000 Hz. This is consistent with the results reported by Le et al (2009) where cepstral coefficients computed in the frequency band from 0 Hz to 1000 Hz were shown to be more important than those computed from other frequency bands.…”

Section: Spectral Centroid Features -Spectral Distribution Of Cognitisupporting

confidence: 94%

“…Each experiment was performed 15 times with a different speaker used as the test speaker each time, and the results were averaged. This mode of experiment was used for the Stroop test corpus previously (Yin et al, 2008;Yap et al, 2010a;Le et al, 2009) and was adopted again for ease of comparison with this work.…”

Section: Back-endmentioning

confidence: 99%

“…Consequently, the original feature vector was concatenated with the shifted delta coefficients (SDCs) of the SCA m and SCF m and used to train the GMMs since SDCs have been shown to be effective in capturing temporal variations of spectral features for cognitive load classification (Yin et al, 2008;Le et al, 2009). The pairwise KL distance, was determined in each subband between 256-mixture GMMs corresponding to the different cognitive load levels.…”

Section: Spectral Centroid Features -Spectral Distribution Of Cognitimentioning

confidence: 99%

“…For each cognitive load level, Gaussian mixture models (GMMs), described in Section 5.2, were then trained using the SCF m and SCA m features from each subband individually, thus giving one GMM per cognitive load level per subband. The Kullback-Leibler distance (KL distance), which has been widely used to estimate the separation between two probabilistic models (Le et al, 2009;Goldberger and Aronowitz, 2005), was adopted to estimate the separation between subband specific GMMs corresponding to different cognitive load levels.…”

Section: Spectral Centroid Features -Spectral Distribution Of Cognitimentioning

confidence: 99%

“…Other features including perceptual linear prediction coefficients, spectral centre of gravity, spectral energy spread, and the vowel durations were also found to be useful for cognitive load classification systems as outlined in (Boril et al, 2010). Further analyses of speechbased features can be found in (Le et al, 2009) where the spectral distribution of information pertaining to cognitive load was studied. The results of this study suggest that the amount of cognitive load specific information in the 0-1 kHz frequency band is particularly high.…”

Section: Introductionmentioning

confidence: 97%

See 4 more Smart Citations

Investigation of spectral centroid features for cognitive load classification

Ambikairajah

Epps

et al. 2011

Speech Communication

Self Cite

View full text Add to dashboard Cite

Section: Spectral Centroid Features -Spectral Distribution Of Cognitisupporting

confidence: 94%

Section: Back-endmentioning

confidence: 99%

Section: Spectral Centroid Features -Spectral Distribution Of Cognitimentioning

confidence: 99%

Section: Spectral Centroid Features -Spectral Distribution Of Cognitimentioning

confidence: 99%

Section: Introductionmentioning

confidence: 97%

See 3 more Smart Citations

Investigation of spectral centroid features for cognitive load classification

Ambikairajah

Epps

et al. 2011

Speech Communication

Self Cite

View full text Add to dashboard Cite

Multimodal Biomarkers to Discriminate Cognitive State

Quatieri¹,

Williamson²,

Smalt³

et al. 2017

The Role of Technology in Clinical Neuropsychology

View full text Add to dashboard Cite

Multimodal biomarkers based on behavioral, neurophysiological, and cognitive measurements have recently increased in popularity for the detection of cognitive stress and neurologically based disorders. Such conditions significantly and adversely affect human performance and quality of life in a large fraction of the world’s population. Example modalities used in detection of these conditions include speech, facial expression, physiology, eye tracking, gait, and electroencephalography (EEG). Toward the goal of finding simple, noninvasive means to detect, predict, and monitor cognitive stress and neurological conditions, MIT Lincoln Laboratory is developing biomarkers that satisfy three criteria. First, we seek biomarkers that reflect core components of cognitive status, such as working memory capacity, processing speed, attention, and arousal. Second, and as importantly, we seek biomarkers that reflect timing and coordination relations both within components of each modality and across different modalities. This is based on the hypothesis that neural coordination across different parts of the brain is essential in cognition. An example of timing and coordination within a modality is the set of finely timed and synchronized physiological components of speech production, whereas an example of coordination across modalities is the timing and synchrony that occur between speech and facial expression during speaking. Third, we seek multimodal biomarkers that contribute in a complementary fashion under various channel and background conditions. In this chapter, as an illustration of the biomarker approach, we focus on cognitive stress and the particular case of detecting different cognitive load levels. We also briefly show how similar feature-extraction principles can be applied to a neurological condition through the example of major depressive disorder (MDD). MDD is one of several neuropsychiatric disorders where multimodal biomarkers based on principles of timing and coordination are important for detection (Cummins et al., 2015; Helfer et al., 2014; Quatieri & Malyska, 2012; Trevino, Quatieri, & Malyska, 2011; Williamson, Quatieri, Helfer, Ciccarelli, & Mehta, 2014; Williamson et al., 2013, 2015; Yu, Quatieri, Williamson, & Mundt, 2014).

show abstract

A study of effectiveness of speech enhancement for cognitive load classification in noisy conditions

Ambikairajah

Thiruvaran

et al. 2015

2015 International Conference on Advanced Technologies for Communications (ATC)

Self Cite

View full text Add to dashboard Cite

In the last decade, speech-features have been effectively utilized for estimating cognitive load level in ideal conditions where recorded speech is clean. However, in more realistic conditions, the recorded speech data is corrupted by noise. Hence, the employment of speech enhancement is essential to reduce the noise. In this paper, the effectiveness of three speech enhancement algorithms proposed in our previous studies are compared based on performance and processing time and the most suitable method is utilized to denoise the input noisy speech before feeding it to a cognitive load classification system in order to improve its performance. The results of this study indicate that the use of speech enhancement can reduce 3.0% of average relative error rate for the system under the effect of various noisy conditions.

show abstract

A non-uniform subband approach to speech-based cognitive load classification

Cited by 10 publications

References 15 publications

Investigation of spectral centroid features for cognitive load classification

Investigation of spectral centroid features for cognitive load classification

Multimodal Biomarkers to Discriminate Cognitive State

A study of effectiveness of speech enhancement for cognitive load classification in noisy conditions

Contact Info

Product

Resources

About