eduzhai > Applied Sciences > Engineering >

Exploiting Fully Convolutional Network and Visualization Techniques on Spontaneous Speech for Dementia Detection

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 6 pages

Abstract: In this paper, we exploit a Fully Convolutional Network (FCN) to analyze theaudio data of spontaneous speech for dementia detection. A fully convolutionalnetwork accommodates speech samples with varying lengths, thus enabling us toanalyze the speech sample without manual segmentation. Specifically, we firstobtain the Mel Frequency Cepstral Coefficient (MFCC) feature map from eachparticipant s audio data and convert the speech classification task on audiodata to an image classification task on MFCC feature maps. Then, to solve thedata insufficiency problem, we apply transfer learning by adopting apre-trained backbone Convolutional Neural Network (CNN) model from theMobileNet architecture and the ImageNet dataset. We further build aconvolutional layer to produce a heatmap using Otsu s method for visualization,enabling us to understand the impact of the time-series audio segments on theclassification results. We demonstrate that our classification model achieves66.7 over the testing dataset, 62.5 of the baseline model provided in theADReSS challenge. Through the visualization technique, we can evaluate theimpact of audio segments, such as filled pauses from the participants andrepeated questions from the investigator, on the classification results.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×