Noise Estimation for Generative Diffusion Models
This work addresses a bottleneck in deploying diffusion models efficiently for speech and image generation, offering a practical solution for faster inference with minimal overhead.
The paper tackles the problem of costly noise parameter tuning in generative diffusion models for efficient few-step denoising, presenting a learning scheme that adjusts parameters step-by-step for any step count and improves synthesis results without model weight changes at negligible computation cost.
Generative diffusion models have emerged as leading models in speech and image generation. However, in order to perform well with a small number of denoising steps, a costly tuning of the set of noise parameters is needed. In this work, we present a simple and versatile learning scheme that can step-by-step adjust those noise parameters, for any given number of steps, while the previous work needs to retune for each number separately. Furthermore, without modifying the weights of the diffusion model, we are able to significantly improve the synthesis results, for a small number of steps. Our approach comes at a negligible computation cost.