MMDisCo: Multi-Modal Discriminator-Guided Cooperative Diffusion for Joint Audio and Video Generation

2024-05-28

AI Summary

This work addresses the challenge of high-cost, low-alignment multimodal generation for audio-video synthesis. To this end, we propose a lightweight audio-video co-generation framework that avoids end-to-end joint training. Instead, it leverages pretrained unimodal diffusion models and introduces a novel “Denoiser-as-Discriminator” mechanism: the optimal discriminator’s gradient directly steers cross-modal denoising, enabling efficient, alignment-aware generation without joint fine-tuning. The framework incorporates only a minimal set of learnable parameters—integrating gradient-guided conditioning and a lightweight cross-modal guidance module. Experiments demonstrate substantial improvements in temporal and semantic alignment across multiple benchmarks, while preserving high unimodal fidelity. Notably, the parameter count is reduced by over an order of magnitude compared to end-to-end multimodal diffusion models.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

This study aims to construct an audio-video generative model with minimal computational cost by leveraging pre-trained single-modal generative models for audio and video. To achieve this, we propose a novel method that guides single-modal models to cooperatively generate well-aligned samples across modalities. Specifically, given two pre-trained base diffusion models, we train a lightweight joint guidance module to adjust scores separately estimated by the base models to match the score of joint distribution over audio and video. We show that this guidance can be computed using the gradient of the optimal discriminator, which distinguishes real audio-video pairs from fake ones independently generated by the base models. Based on this analysis, we construct a joint guidance module by training this discriminator. Additionally, we adopt a loss function to stabilize the discriminator's gradient and make it work as a noise estimator, as in standard diffusion models. Empirical evaluations on several benchmark datasets demonstrate that our method improves both single-modal fidelity and multimodal alignment with relatively few parameters. The code is available at: https://github.com/SonyResearch/MMDisCo.

Problem

• Minimizes computational cost
• Enhances audio-video alignment
• Leverages pre-trained models

Innovation

• Multi-Modal Discriminator-Guided Cooperative Diffusion
• Lightweight Joint Guidance Module
• Stabilized Discriminator Gradient as Noise Estimator