SoundLoCD: An Efficient Conditional Discrete Contrastive Latent Diffusion Model for Text-to-Sound Generation

About

This is a demo for our paper 'SoundLoCD: An Efficient Conditional Discrete Contrastive Latent Diffusion Model for Text-to-Sound Generation_'. The SoundLoCD is a LoRA-based conditional discrete contrastive latent diffusion model for text-to-sound effects generation. Unlike recent large-scale audio generation models, our SoundLoCD can be efficiently trained under limited computational resources. The integration of a contrastive learning strategy enhances the connection between textual conditions and the generated audio outputs, resulting in coherent performance.

Citation

If you are interesting in our work, please cite it as below:

@INPROCEEDINGS{niu2023soundlocd,
  author={Niu, Xinlei, Jing Zhang, Christian Walder, and Charles Patrick Martin}
  title={SoundLoCD: An Efficient Conditional Discrete Contrastive Latent Diffusion Model for Text-to-Sound Generation}, 
  year={2023}
}