Features Motivated From Uncertainty Principle for Classification of Normal vs. Pathological Infant Cry

This study investigates the significance of the pro-posed uncertainty vector (u-vector) derived from Heisenberg's uncertainty principle in signal processing framework, and related
$t$
-vector and
$\omega$
-vector for classification of normal vs. pathological infant cries. Fundamental frequency
$(F_{0})$
contours in an infant's cries are rhythmic and have patterns w.r.t. the variances in time and frequency-domains. On the other hand, pathological cries have more smeared variances. To that effect, we propose
$u$
-vector as a combination of two features, namely,
$t$
-vector, and
$\omega$
-vector, which capture variances in time and frequency domains, respectively. Furthermore, we also find that non-cepstral features are better suited for pathology detection as compared to the cepstral features. Given that early diagnosis of pathology in infants is important, the detection procedure should be efficient in terms of the time taken. To that effect, we also present latency analysis of
$u$
-vector,
$t$
-vector, and
$\omega$
-vector w.r.t. Constant
$\mathrm{Q}$
Transform (CQT) as baseline. We observe that
$\omega$
-vector achieves remarkable latency w.r.t. the remaining features, indicating the potential of the proposed feature for practical system deployment. The
$\omega$
-vector also shows the best performance of 98.50% as overall accuracy, and achieves an absolute improvement of 1.5% in accuracy.