eduzhai > Applied Sciences > Engineering >

Conditioned Time-Dilated Convolutions for Sound Event Detection

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 5 pages

Abstract: Sound event detection (SED) is the task of identifying sound events alongwith their onset and offset times. A recent, convolutional neural networksbased SED method, proposed the usage of depthwise separable (DWS) andtime-dilated convolutions. DWS and time-dilated convolutions yieldedstate-of-the-art results for SED, with considerable small amount of parameters.In this work we propose the expansion of the time-dilated convolutions, byconditioning them with jointly learned embeddings of the SED predictions by theSED classifier. We present a novel algorithm for the conditioning of thetime-dilated convolutions which functions similarly to language modelling, andenhances the performance of the these convolutions. We employ the freelyavailable TUT-SED Synthetic dataset, and we assess the performance of ourmethod using the average per-frame $ text{F} {1}$ score and average per-frameerror rate, over the 10 experiments. We achieve an increase of 2 (from 0.63to 0.65) at the average $ text{F} {1}$ score (the higher the better) and adecrease of 3 (from 0.50 to 0.47) at the error rate (the lower the better).

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×