Wen Wang

Huadai Liu, Kaicheng Luo, Jialei Wang et al.

While end-to-end video-to-audio generation has greatly improved, producing high-fidelity audio that authentically captures the nuances of visual content remains challenging. Like professionals in the creative industries, this generation requires sophisticated reasoning about items such as visual dynamics, acoustic environments, and temporal relationships. We present ThinkSound, a novel framework that leverages Chain-of-Thought (CoT) reasoning to enable stepwise, interactive audio generation and editing for videos. Our approach decomposes the process into three complementary stages: foundational foley generation that creates semantically coherent soundscapes, interactive object-centric refinement through precise user interactions, and targeted editing guided by natural language instructions. At each stage, a multimodal large language model generates contextually aligned CoT reasoning that guides a unified audio foundation model. Furthermore, we introduce AudioCoT, a comprehensive dataset with structured reasoning annotations that establishes connections between visual content, textual descriptions, and sound synthesis. Experiments demonstrate that ThinkSound achieves state-of-the-art performance in video-to-audio generation across both audio metrics and CoT metrics, and excels in the out-of-distribution Movie Gen Audio benchmark. The project page is available at https://ThinkSound-Project.github.io.

Hao Lang, Wen Wang

We propose a learning approach for turn-level spoken language understanding, which facilitates a user to speak one or more utterances compositionally in a turn for completing a task (e.g., voice ordering). A typical pipelined approach for these understanding tasks requires non-trivial annotation effort for developing its multiple components. Also, the pipeline is difficult to port to a new domain or scale up. To address these problems, we propose an end-to-end statistical model with weak supervision. We employ randomized beam search with memory augmentation (RBSMA) to solve complicated problems for which long promising trajectories are usually difficult to explore. Furthermore, considering the diversity of problem complexity, we explore automated curriculum learning (CL) for weak supervision to accelerate exploration and learning. We evaluate the proposed approach on real-world user logs of a commercial voice ordering system. Results demonstrate that when trained on a small number of end-to-end annotated sessions collected with low cost, our model performs comparably to the deployed pipelined system, saving the development labor over an order of magnitude. The RBSMA algorithm improves the test set accuracy by 7.8% relative compared to the standard beam search. Automated CL leads to better generalization and further improves the test set accuracy by 5% relative.