Speaker: Sergio Segovia González.
Abstract:
The answer to this question through these two articles, “WILDDESED: An LLM-POWERED dataset for wild domestic environment sound event detection system” and “Leveraging LLM and Text-Queried Separation for Noise-Robust Sound Event Detection” has been carried out:
1.- Create a new dataset called WildDESED to advance sound event detection (SED) research in noisy domestic environments .
2.- Explore the use of curriculum learning to develop noise-robust SED systems using the WildDESED dataset.
3.- Develop a noise augmentation method using large language models (LLMs) to improve the noise-robustness of sound event detection (SED) models.
4.- Use the fine-tuned SED model to generate text queries for a language-queried audio source separation (LASS) model, in order to improve SED performance in noisy environments.
5.- Explore the use of LLMs for noise-robust SED as a novel approach to handling overlapping sound events.