Bi-SCNet / README.md
richa-namballa's picture
Update README.md
cb64af1 verified
|
Raw History Blame Contribute Delete
1.09 kB
---
license: mit
---
# Bi-SCNet
**Binaural Sparse Compression Network for Music Source Separation**
See: [Spacing Out: On the Reliability of Binaural Music Source Separation Metrics](https://github.com/richa-namballa/binaural-mss/)
**Binaural Model Checkpoint**
* `binaural_scnet.tar.gz`: Checkpoint for the Bi-SCNet model trained on [Binaural-MUSDB](https://github.com/richa-namballa/binaural-mss/tree/main/data) with only RMSE loss.
* `binaural_scnet_spat_loss.tar.gz`: is the checkpoint for the Bi-SCNet model trained on Binaural-MUSDB with the ITD-based spatial loss equally weighted with the RMSE loss.
The [spatial loss](https://github.com/richa-namballa/Bi-SCNet/blob/d75678e19f969bd4a1c981f794724d4aefb6ef86/scnet/loss.py#L142) is based on ["Interaural Time Difference Loss for Binaural Target Sound Extraction"](https://ieeexplore.ieee.org/abstract/document/10694612) by Hernandez-Olivan et al. (2024)
This model is an adaptation of the [SCNet: Sparse Compression Network for Music Source Separation](https://github.com/starrytong/SCNet) repository by Tong, et al. (2024).