Modified Group Delay Based Features for Asthma and HIE Infant Cries Classification

Modified group delay features have shown promising results for automatic speech recognition (ASR) task. In this paper, features are derived from the modified group delay function. These features are then used to classify asthma and hypoxy ischemic encephalopathy (HIE) infant cries. Our experimental results show that the performance of the proposed features is better than the state-of-the-art feature set, i.e., Mel frequency cepstral coefficients (MFCC). Best classification accuracy is achieved with the proposed features is 90.38 % as opposed to 84.92 % obtained with MFCC, when applied to a SVM classifier with radial basis function kernel. The proposed feature set performs much better for classification of asthma infant cries. However, for HIE both features perform equally well. The class separability distance of the group delay feature is higher than the MFCC feature (for most of the features Bhattacharya class separation distance is higher by 0.2 units compared to MFCC), which also confirms that the proposed features are better than MFCC.