- Type
- Vocal remover: AI separation of a mixed song into vocal and music, mixed back at your levels
- Model
- Spleeter 2-stem by Deezer Research (MIT licence), run on your computer by the plugin’s own code; no internet, no upload
- Levels
- VOCAL and MUSIC faders, Off to +6 dB (0 dB at three quarters of the travel); REMOVE VOCAL, VOCAL ONLY and ORIGINAL
- Strength
- 0 to 100 % (default 50 %, the model’s own balance)
- Vocal range
- LOW 20 Hz (off) to 1 kHz, HIGH 1 kHz to 20 kHz (off); 48 dB/octave beyond each
- Output
- −24 to +12 dB
- Measured
- On test songs at STRENGTH 50 %: a sung vocal removed to about −21 dB, the band’s bleed in an a cappella about −23 dB; a spoken voice separates only 5 to 7 dB
- Latency
- 52,224 samples at 44.1 kHz, 56,925 at 48 kHz (about 1.18 s), reported to your DAW; with both faders at 0 dB and bypassed, the input delayed by the latency, bit for bit
- Metering
- SEPARATION display (scrolling spectrogram, 40 Hz to 16 kHz, the vocal and the music coloured); a level column per fader; AI LOAD, LATENCY and LATE; IN and OUT peak and RMS; short-term and integrated LUFS
- Presets
- 45 factory presets in 8 folders (including Karaoke, A Cappella, Practice and Speech), plus your own
- Channels
- Mono or stereo
- CPU (measured)
- AI load about 35 to 45 % of one core of a 2-core cloud VM; much less on Apple silicon; nothing on silence. About 80 MB of memory for the model, shared by every instance
- Formats
- VST3, AU, Standalone
- macOS
- macOS 11 or later, Apple silicon or Intel (universal). VST3, AU and Standalone.
- Windows
- Windows 10 or later, 64-bit. VST3 and Standalone.
- Pro Tools
- Pro Tools (AAX) versions are built and tested, and arrive once they’re signed for Pro Tools.
- Latest version
- 0.2.1, released October 31, 2026 · What’s new