Sound Event Detection

Participated in DCASE (Detection and Classification of Acoustic Scenes and Events) Task4 Sound Event Detection in 2023, our team applied an attention network optimized for extracting frequency information.

You can take a look at our Technical Report.

In the process of addressing this task, I learned that it is imperative to tailor methodologies to suit the specific characteristics of the dataset. Notably, audio data, although transformed into 2D images via log mel spectrograms, presents distinct features from those found in images. During the course of this experiment, a comprehensive review of papers such as Swin Transformer and Wav2Vec inspired me. Particular emphasis was placed on the extraction of frequency information from audio, focusing on implementing attention mechanisms. Looking ahead, my trajectory in research endeavors is inclined towards the exploration of domains like voice recognition and related pursuits.