Employing Bottleneck and Convolutional Features for Speech-Based Physical Load Detection on Limited Data Amounts

Abstract

The detection of different levels of physical load from speech has many applications: Besides telemedicine, non-contact detection of certain heart rate ranges can be useful for sports and other leisure time devices. Available approaches mainly use a high number of spectral and prosodic features. In this setting of typically small data sets, such as the Talk & Run data set and the Munich Biovoice Corpus, the high-dimensional feature spaces are only sparsely populated. Therefore, we aim at a reduction of the feature number using modern neural net inspired features: Bottleneck layer features, obtained from standard low-level descriptors via a feed-forward neural network, and activation map features, obtained from spectrograms via a convolutional neural network. We use these features for an SVM classification of high and low physical load and compare their performance. We also discuss the possibility of hyperparameter transfer of the extracting networks between different data sets. We show that even for limited amounts of data, deep learning based methods can bring a substantial improvement over "conventional" features.

Multi-Task Learning using Frequency-Warped Spectrum for Detecting Out-of-breath Speech

“…Both the blocks are followed by maxpooling layers having kernel sizes (4, 4) VIII. Our proposed method betters the classification performance by at least 7% for all the metrics compared to the model in [15].…”

“…In literature, Egorow et al have used a convolutional neural network for classifying neutral, and OBS classes on the Munich biovoice corpus and TalkR databases [15], [34], [35]. Due to the unavailability of those databases, we have replicated the network for comparing classification performance using OBSE database.…”

“…Recently Boelders et al used a pair of speech utterances in a siamese network for dyspnea detection [14]. Egorow et al used a convolutional neural network for automatic feature extraction followed by classification by SVM [15].…”

Hybrid Handcrafted and Learnable Audio Representation for Analysis of Speech Under Cognitive and Physical Load

“…Alternatively, several deep learning frameworks have been proposed for cognitive and physical load detection [14][15][16]. Yet, training such models directly on task load corpora remains a challenging issue due to their small sizes, as large or complex networks tend to overfit and have poor generalisation capacity.…”