eduzhai > Applied Sciences > Engineering >

Aaltos End-to-End DNN systems for the INTERSPEECH 2020 Computational Paralinguistics Challenge

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: End-to-end neural network models (E2E) have shown significant performancebenefits on different INTERSPEECH ComParE tasks. Prior work has applied eithera single instance of an E2E model for a task or the same E2E architecture fordifferent tasks. However, applying a single model is unstable or using the samearchitecture under-utilizes task-specific information. On ComParE 2020 tasks,we investigate applying an ensemble of E2E models for robust performance anddeveloping task-specific modifications for each task. ComParE 2020 introducesthree sub-challenges: the breathing sub-challenge to predict the output of arespiratory belt worn by a patient while speaking, the elderly sub-challenge toestimate the elderly speaker s arousal and valence levels and the masksub-challenge to classify if the speaker is wearing a mask or not. On each ofthese tasks, an ensemble outperforms the single E2E model. On the breathingsub-challenge, we study the impact of multi-loss strategies on taskperformance. On the elderly sub-challenge, predicting the valence and arousallevels prompts us to investigate multi-task training and implement datasampling strategies to handle class imbalance. On the mask sub-challenge, usingan E2E system without feature engineering is competitive to feature-engineeredbaselines and provides substantial gains when combined with feature-engineeredbaselines.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×