HertzKit vocal and instrumental separator. Protected access with verified email. Model: UVR-MDX-NET-Voc_FT. Ultimate Vocal Remover (UVR), by Anjok07, aufr33 and contributors. The UVR README asks third-party apps to honor MIT and credit UVR and its developers. https://github.com/Anjok07/ultimatevocalremovergui Model mirror (matching verified weights): https://huggingface.co/Blane187/all_public_uvr_models Model SHA-256: 534b2070fcc7df514b13ef660dc8cbb328679c2374d04354a5c42bb14ecce111 web-audio-separation: Dan Kurtz, MIT, commit d412f5a45e34bf73883516b7fcc4561c95c028bf. https://github.com/incidentist/web-audio-separation ONNX Runtime Web 1.29.0: Microsoft and contributors, MIT. Third-party notices included. PFFFT: Julien Pommier, Hayati Ayguen and NCAR FFTPACK; license included. Local adaptation uses verified model bytes, original-level stem subtraction, automatic device detection and cooperative cancellation between inference batches and a reusable engine per page to avoid repeated threaded runtime initialization. Source quality is not guaranteed.