Publication

Abstract : In this work, individuals with psychological unit of depression are detected using speech samples. Spectrogram based convolutional neural networks and end to end convolutional neural network models are implemented to achieve the task. Parameter tuning has been performed by choosing different sub-parameters of convolutional neural network. Speech samples from audio visual emotion challenge (AVEC) 2016 DAIC-Woz dataset are utilized for validating the models. Experimental analysis has shown that performance of end to end model is ahead of spectrogram based model and baseline models by an efficiency of 13%. Further, the proposed model has been applied to estimate the severity level of depression using the PHQ-8 scores of speech samples which has been never attempted using speech samples.