Stem Separator v6: DAW Export, Sections, Chords, Mashup, and 6 More
The Stem Separator just got its biggest upgrade yet โ v6 โ with ten new features that turn it from a separation tool into a full production workstation. Everything still runs on the local RTX 2070 (no cloud, no accounts, no per-track fees).
What's new
- ๐๏ธ DAW Session Export โ export your separation as a ready-to-open Reaper .rpp project, or a universal stem bundle (stems + tempo/key/sections metadata + cues) that imports into Ableton, FL Studio, Logic, and Reaper. Section markers are baked in.
- ๐ Song Section Detection โ auto-detects intro / verse / chorus / bridge / outro. A clickable color-coded timeline; click any section to export just that slice.
- ๐น Chord Progression โ detects the actual chord progression (not just the key) and exports it as a playable .mid file.
- ๐ Stem Mashup โ combine a stem from one track with a stem from another. Tempo and key are matched automatically so they lock together.
- โจ Per-Stem EQ + Effects โ low/mid/high EQ, reverb, compressor, and de-esser on any stem, with live preview and baked export.
- ๐ต Pitch Shift / Karaoke โ shift any stem by semitones (high-quality Rubberband, duration preserved) for karaoke or matching a singer's range.
- ๐๏ธ Mix Presets โ one-click Karaoke, Vocals-Only, Drums+Bass, and Instrumental, plus save your own custom mix recipes.
- ๐ฌ Multi-Model A/B โ re-separate the same track with a different model and compare results side by side.
- ๐ Audio Validation โ every track now reports duration, sample rate, peak, RMS, and approximate LUFS, with clipping/mono warnings.
- โK Command Palette + PWA โ a full command palette, keyboard shortcuts (space, 1โ8 to mute stems, C, T), a light/dark theme, and an installable PWA.
Under the hood
The backend grew seven new endpoints (sections, chords+MIDI, validation, pitch, FX, DAW export, mashup, re-separate) built on librosa, Krumhansl-Schmuckler key/chord profiling, and ffmpeg's Rubberband for pitch. The frontend is still a single dependency-free HTML file. All ten features passed a full server regression suite plus a headless-browser E2E โ zero console errors.
One item remains pending from last round: live MIDI transcription (piano-roll) needs a free HuggingFace token for the license-gated MuScriptor weights. The chord-progression MIDI export above works today without it.
Try it: Stem Separator