de-leak-rtStem bleed removal for Demucs separations

What does de-leak-rt do?

de-leak-rt removes stem bleed from Demucs and htdemucs separations in real time. A classifier flags inter-stem bleeding frame by frame and a multiband gain pulls it down without touching the music (F1 = 0.993 on MUSDB18-HQ). It runs as a VST3 / CLAP plug-in with a 14-day free trial.

7 presets, tuned per stem type

Pick by what kind of stem you're cleaning. Each preset is validated against a real corpus of Suno, Udio, and Demucs separations. Fine-tune later with the gate sliders.

Three ML models, three character profiles

All preserve vocals; they differ in how aggressively they handle reverb tails and how much non-vocal content they let through.

Built for stem work

How it works

de-leak-rt frames each STFT bin, classifies it as content vs. leak, and applies multiband gain. Causal — no lookahead — so live monitoring works.

audio (44.1k mono) → STFT → mag → LeakGate (ONNX) → gate_t (T,)
                  → multiband gain (per-freq exponent w/ hf_comp)
                  → spec * gain → ISTFT → clean

The classifier is a 95 K-parameter MLP (~370 KB ONNX) — it makes one decision per STFT frame. Multiband gain is applied with HF compensation so the high-band content you want to preserve isn't pulled down with the leak you don't.

System requirements

de-leak-rt's classifier is a 95 K-parameter MLP (~370 KB ONNX) — light enough that even an entry-level M1 hits 134× realtime, so a single instance is comfortable on every supported configuration.

FAQ

Why do I get clicks when toggling the band solo?

Fixed — solo transitions use a one-pole ramp (~30 ms half-life). If you still hear them, please report with the host and SR.

Diff + solo together — what should I hear?

Only the dry-minus-wet of the active band. If a vocal appears in the residual you're over-gating that band; back off intensity or raise HF compensation.

Offline bounce comes out silent or short.

The host needs PDC enabled. de-leak-rt reports latency = host_chunk_samples (~93 ms @ 44.1 k). Most modern DAWs handle this automatically.

Will it work on a non-44.1 kHz session?

Yes — rubato resamples 8 k to 384 k transparently. The model always runs at its native 44.1 k internally.

How does it differ from a regular noise gate?

A noise gate triggers on RMS below threshold. de-leak-rt's classifier was trained to distinguish musical content from leak per-frame — it stays open during sub-threshold content and closes during loud leak.

What if my problem is separation residue rather than bleed?

They are different defects and want different treatment. Bleed is another instrument present in the wrong stem — what this plug-in targets. Residue is the separator's own reconstruction damage smeared through the stem. Assess them separately instead of raising intensity to suppress an instrument: the bleed-removal workflow covers the bleed case, and cleaning a separated vocal stem separates the two by ear and fixes each.

Hear it on real stems

Pick a stem and tap Original / De-leaked / Diff to A/B in place. Playback is synchronized across all three — no re-buffer, no click, no time jump. These are real source-separation outputs (Suno/Udio internal splitters, a demucs/htdemucs re-separation of the same mixes, and a human multitrack) — not synthetic. "De-leaked" uses de-leak-rt's strongest RT model (balanced_large_v8); Diff plays only what the gate removed.

Try it free for 14 days, buy when you're sold

See de-leak-rt and de-artifact on the same drum bus first — A/B against the dry source. Then download the installer for a free 14-day full trial — every preset, every parameter, no gaps. After 14 days the plug-in inserts a brief silent gap every 60 seconds until you activate.