search

RoyChao19477 / SEMamba

This is the official implementation of the SEMamba paper. (Accepted to IEEE SLT 2024)
Other
143 stars 14 forks source link

Evaluation on DNS dataset #11

Open Ludvig-Joborn opened 3 months ago

Ludvig-Joborn commented 3 months ago

Hi!

Thank you for your very interesting research regarding the application of Mamba to Speech Enhancement! Mamba models are very exciting, and their application to Sound-related tasks seem to be a very fruitful line of research.

Now, from my experience, models trained and evaluated on the DNS (Deep Noise Suppression) dataset are more robust to various conditions of audio recordings, and so I would find it very interesting to see how well SEMamba performs on that dataset.

In short, are you planning on evaluating on the DNS dataset? And if so, released model weights would be tremendously helpful. 👍

RoyChao19477 commented 4 days ago

Hi Ludvig-Joborn,

Thank you for your kind words and for showing interest in our research!

Currently, we have trained our model on the DNS Challenge 2022 dataset, and we plan to release the results in the future.

Additionally, we’ve also developed a larger model trained on an extensive dataset (approximately 1.3 TB). We intend to release both the model weights and further results once our corresponding publication is finalized.

We truly appreciate your enthusiasm and will keep the community updated on our progress!

Best regards, Roy Chao