Audio-alpaca
收藏arXiv2025-09-30 收录
下载链接：
https://huggingface.co/datasets/declare-lab/audio-alpaca
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了供扩散模型学习的偏好数据，每个提示都对应一个获胜的音频输出和几个失败的音频输出。此外，该数据集专门用于对Tango文本到音频模型进行微调。该数据集的规模为15,025个偏好配对，其任务是文本到音频的生成。
This dataset provides preference data for diffusion model training. Each prompt corresponds to one winning audio output and several losing audio outputs. Moreover, this dataset is specifically designed for fine-tuning the Tango text-to-audio model. It consists of 15,025 preference pairs, targeting the text-to-audio generation task.
