π arXiv | π¬ MMAE Demo Video | π οΈ GitHub Code | π HuggingFace Audio Download
We introduce MMAE, a Massive Multitask Audio Editing benchmark, serving as the first comprehensive evaluation testbed designed for general-purpose instruction-based audio editing.
MMAE extends to a broad spectrum of real-world scenarios, encompassing 7 distinct audio modalities, including sound, speech, music, and⦠See the full description on the dataset page:
https://huggingface.co/datasets/BoJack/MMAE.