Moises vs Open-Unmix

Moises has no write-up on Specifiction yet; its figures are above.

Open-Unmix is a PyTorch reference implementation for music source separation, aimed at researchers, audio engineers, and artists. Its ready-to-use models split pop music into vocals, drums, bass, and other instruments; a separate model, umxse, is for speech enhancement and was trained on the 28-speaker Voicebank+DEMAND corpus. Pre-trained choices include umxl, umxhq, umx, and umxse, with spectrogram variants also available. Each source model uses a three-layer bidirectional LSTM, a design the project says cannot run online or in real time. The code supports reproducing results, developing architectures, and adding data, while training is documented separately from the package. Install through pip or use the Docker container; pre-trained models can be loaded in Python or through torch.hub, and a Colab notebook supports online experimentation without local setup. The command line handles WAV, FLAC, and OGG, not MP3; stempeg can extend formats, including STEMS export. The software is MIT-licensed; umxl weights are non-commercial under CC BY-NC-SA 4.0.