Vocal Isolation & Stem Separation

Demucs, Spleeter, Ultimate Vocal Remover, Moises, iZotope RX — pull vocals / drums / bass / other out of a mix.

AI stem separation matured dramatically through 2023-26. Free local tools (Demucs, UVR) now produce results that rival paid cloud services. Use cases: karaoke / acapella creation, sampling from finished tracks, podcast bleed reduction, lyric transcription pipelines (pair with transcription for vocals → Whisper). For audio editors see audio-editors. For voice / podcast specifics see podcast-voice-tools.

Free / OSS local separation

★ Demucs (Meta / Facebook AI Research) — MIT, Python. The reference open-source separation model. htdemucs and htdemucs_ft are the current best models (4-stem: vocals / drums / bass / other). Runs on CPU or GPU; an M-series Mac or modest GPU can separate a 4-min song in under a minute. The best free quality in 2024-26.
★ Ultimate Vocal Remover (UVR / UVR5) — MIT GUI front-end; bundles MDX-Net / VR Architecture / Demucs models. Cross-platform. Drag-drop interface, model picker, 2 / 4 / 5 / 6-stem options, denoising / dereverb-specific models. The default GUI for most users in 2026.
Spleeter (Deezer) — MIT, Python. The original 2019 release that started the wave; Demucs has surpassed it on quality. Still useful — fast, well-documented, easy CLI.
MVSEP-MDX23 / Bytedance KitS — newer model checkpoints; community ports drop into UVR.
Open-Unmix — MIT, PyTorch; older but actively used in research.

Paid / cloud stem separation

Moises.ai — paid freemium; ~5 free tracks/mo, paid for unlimited; web + mobile. Practice-app-leaning UX with chord/key detection bundled.
LALAL.AI — paid pay-per-minute; very high quality, especially older catalog masters.
Music.AI (formerly AudioShake) — paid B2B; broadcast / sync licensing tier.
Music Source Separation Pro — paid web.
iZotope RX Music Rebalance — included with RX 10/11 (paid suite); good quality, oriented at restoration workflows.
Adobe Podcast / Premiere Enhance Speech — bundled with Creative Cloud.

Restoration / repair (broader)

★ iZotope RX 11 — paid (~$400 standard / ~$1200 Advanced; deep discounts common). The industry standard restoration suite. De-noise, de-clip, de-reverb, mouth-de-click, spectral repair brush. Used everywhere in film / TV / podcasting post.
CrumplePop — paid Mac plugin suite; per-effect AI plugins (RustleRemover, EchoRemover).
Accusonus ERA Bundle — paid simple-knob restoration plugins (acquired by Meta 2022; aging).
Audacity Noise Reduction / Noise Gate — free; built-in but functional.
Auphonic — see podcast-voice-tools; free tier with loudness norm + dialog leveler.

Moises.ai — chord detection + transposition + loop sections.
Chordify — paid web; chord recognition.
AnthemScore — paid; AI-powered audio-to-MIDI / sheet music.
MelodyScanner — paid app.

Vocal-to-MIDI / pitch tools

Melodyne (Celemony) — paid (~$100-700); the gold-standard pitch / time editor.
Auto-Tune Pro X (Antares) — paid; the eponymous pitch-corrector.
Waves Tune Real-Time — paid; live real-time autotune.
GSnap — free Win VST autotune; classic.
MAutoPitch (Melda) — free MFreeFXBundle pitch corrector.
ReaTune (Reaper) — free with Reaper.

Pick this if…

Best free local stem separation: Demucs (CLI) or UVR (GUI).
Cloud, low-friction: Moises (free tier) or LALAL.AI.
Audio restoration / repair: iZotope RX (paid industry standard) or Audacity (free).
Free autotune: GSnap or MAutoPitch (Melda).
Pitch / time pro tool: Melodyne.
Lyrics / vocal-to-text: UVR vocals → Whisper (see transcription).