Official Hugging Face Diffusers Implementation of QA-MDT
QAMDT: Quality-Aware Diffusion for Text-to-Music 🎶
QADMT brings a new approach to text-to-music generation by using quality-aware training to tackle issues like low-fidelity audio and weak labeling in datasets.
With a masked diffusion transformer (MDT), QADMT delivers SOTA results on MusicCaps and Song-Describer, enhancing both quality and musicality.
It follows from this paper by the University of Science and Technology of China, authored by @changli et al..
Usage:
!git lfs install
!git clone https://huggingface.co/jadechoghari/openmusic qa_mdt
This command will change the folder name from openmusic to qa_mdt
pip install -r qa_mdt/requirements.txt
pip install xformers==0.0.26.post1
pip install torchlibrosa==0.0.9 librosa==0.9.2
pip install -q pytorch_lightning==2.1.3 torchlibrosa==0.0.9 librosa==0.9.2 ftfy==6.1.1 braceexpand
pip install torch==2.3.0+cu121 torchvision==0.18.0+cu121 torchaudio==2.3.0 --index-url https://download.pytorch.org/whl/cu121
from qa_mdt.pipeline import MOSDiffusionPipeline
pipe = MOSDiffusionPipeline()
pipe("A modern synthesizer creating futuristic soundscapes.")