---
license: mit
pipeline_tag: video-to-audio
---

# Hear-Your-Click: Interactive Object-Specific Video-to-Audio Generation

This repository contains the official model for **Hear-Your-Click**, an interactive framework designed for object-specific video-to-audio (V2A) generation. It enables users to generate sounds for specific objects within a video simply by clicking on the frame, addressing the limitations of global video information in complex scenes.

**[📚 Paper](https://huggingface.co/papers/2507.04959)** | **[💻 GitHub Repository](https://github.com/SynapGrid/Hear-Your-Click-2024)**

<p align="center">
    <img src="https://github.com/user-attachments/assets/2ca49ab5-80ca-42c4-b9a5-9dc7959ac358">
</p>

## About Hear-Your-Click

Hear-Your-Click introduces several key innovations to improve V2A generation:
-   **Object-aware Contrastive Audio-Visual Fine-tuning (OCAV)** with a **Mask-guided Visual Encoder (MVE)** to obtain object-level visual features aligned with audio.
-   Two tailored data augmentation strategies: **Random Video Stitching (RVS)** and **Mask-guided Loudness Modulation (MLM)**, which enhance the model's sensitivity to segmented objects.
-   A new evaluation metric, the **CAV score**, designed to measure audio-visual correspondence more accurately.

This framework offers more precise control and significantly improves generation performance across various metrics.

## Installation

To set up the Hear-Your-Click environment, follow these steps:

1.  **Clone the repository**:
    ```bash
    git clone https://github.com/SynapGrid/Hear-Your-Click-2024.git
    cd Hear-Your-Click-2024
    ```

2.  **(Optional) Create a Conda environment**:
    ```bash
    conda env create -n hyc python=3.9.11
    conda activate hyc
    ```

3.  **Install dependencies**:
    ```bash
    pip install -r requirements.txt
    ```

## Model Checkpoints

1.  **Download the model weights** and place them in `./hyc_inference/inference/ckpt/`:
    *   [epoch=000059.ckpt](https://drive.google.com/file/d/1QX24gEmN-cG03NlO0zT1geK1eUgOqDtk/view?usp=drive_link)
    *   [epoch_10.pt](https://drive.google.com/file/d/15tbqXR-99QNg-Il6wxPD66q4EM4UkVvJ/view?usp=drive_link)
    *   [eval_classifier.ckpt](https://huggingface.co/SimianLuo/Diff-Foley/resolve/main/diff_foley_ckpt/eval_classifier.ckpt)
    *   [double_guidance_classifier.ckpt](https://huggingface.co/SimianLuo/Diff-Foley/resolve/main/diff_foley_ckpt/double_guidance_classifier.ckpt)

    You can use `gdown` and `wget` for convenient downloading:
    ```bash
    pip install gdown

    cd ./hyc_inference/inference/ckpt

    gdown https://drive.google.com/uc?id=1QX24gEmN-cG03NlO0zT1geK1eUgOqDtk

    gdown https://drive.google.com/uc?id=15tbqXR-99QNg-Il6wxPD66q4EM4UkVvJ

    wget https://huggingface.co/SimianLuo/Diff-Foley/resolve/main/diff_foley_ckpt/eval_classifier.ckpt

    wget https://huggingface.co/SimianLuo/Diff-Foley/resolve/main/diff_foley_ckpt/double_guidance_classifier.ckpt
    ```

2.  **Download additional model weights** and place them in `./checkpoints`:
    *   [clap_clip.pt](https://github.com/MCR-PEFT/C-MCR/blob/main/checkpoints/clap_clip.pt)
    *   [laion_clap_fullset_fusion.pt](https://huggingface.co/lukewys/laion_clap/blob/main/630k-fusion-best.pt)
    *   [clip-vit-base-patch32](https://huggingface.co/openai/clip-vit-base-patch32)

## Inference Command

Launch the inference demo using the following command:
```bash
python app.py --device cuda:0,1 --sam_model_type vit_b
```

## Citation

If you find this work useful for your research or applications, please cite our paper:

```bibtex
@misc{liang2025hearyourclickinteractivevideotoaudiogeneration,
      title={Hear-Your-Click: Interactive Video-to-Audio Generation via Object-aware Contrastive Audio-Visual Fine-tuning}, 
      author={Yingshan Liang and Keyu Fan and Zhicheng Du and Yiran Wang and Qingyang Shi and Xinyu Zhang and Jiasheng Lu and Peiwu Qin},
      year={2025},
      eprint={2507.04959},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2507.04959}, 
}
```