de-leak-rt API

Bulk leak removal for separated stems via REST. The same ML classifier (F1 = 0.993) that powers the plug-in, exposed for stem-separation pipelines processing 100+ tracks at a time.

Pricing on request Contact for access Open Dashboard →

Throughput100+ stems / batch
Speed~134× realtime per stem
PresetsAll 7 plug-in presets
Models3 character profiles

# Submit separated stems for leak removal
curl -X POST https://api.intrect.io/v1/leak-gate \
  -H "Authorization: Bearer sk-int-..." \
  -H "Content-Type: application/json" \
  -d '{
    "preset": "ai_ghost",
    "model": "balanced_v7",
    "stems": [
      {"id": "vocals_001",  "kind": "vocal",  "url": "..."},
      {"id": "drums_001",   "kind": "instr",  "url": "..."},
      {"id": "other_001",   "kind": "other",  "url": "..."}
    ]
  }'

{
  "job_id": "lkg_8b1a...",
  "status": "processing"
}

Designed for stem-separation pipelines

The plug-in cleans one stem at a time at the DAW. The API handles the other case: a service that runs Demucs / htdemucs across thousands of tracks and needs leak gating at the end of the pipeline.

Per-stem preset routing

Tag each stem with its kind (vocal / instr / other) and the API picks the right preset automatically — Vocal for voice, Instrument for drums, AI Ghost for residuals.

Same models as the plug-in

Default v3, Balanced v7, Aggressive Reverb v5 — all three available per-request. Same ONNX, same character.

134× realtime per stem

4-minute stem renders in ~1.8 s. Batch of 100 stems averages under 3 minutes wall-clock with parallel workers.

Diff output available

Optional return_diff: true emits the residual (dry − wet) as a parallel track — useful for forensic auditing.

Webhook callbacks

Don't poll. Register a webhook, receive a POST per completed batch with presigned URLs valid 24 h.

SLA & private deployment

Enterprise tier: 99.9 % SLA, dedicated workers, on-premise deployment for catalogs that can't leave the org.

Use cases

Who	What they do with it
Stem separation services	Run leak gating as the final stage of a Demucs / htdemucs pipeline before delivering stems to customers.
Sample library makers	Clean separated drum / bass / vocal stems for sample packs — no inter-stem residue.
Post-production houses	Restore vocal / dialog stems extracted from finished productions.
Music education / transcription	Process catalog of separated stems for training material — clean vocals for mimicry, clean drums for tab analysis.