Mark D. Plumbley

Efficient Text-to-Audio Generation via Pruning

Arshdeep Singh, Yi Yuan, Yun Chen, Wenwu Wang, Mark D. Plumbley

Diffusion-based text-to-audio generative models such as AudioLDM achieve high perceptual quality and strong semantic consistency; however, their practical deployment is hindered by the substantial computational cost of the U-Net denoising backbone. In this work, we apply model pr…Read more