Drum Kit Separation (DrumSep) · TelkNet

Separation & stem splitGPU workload

Drum Kit Separation (DrumSep)

Q: Who is Drum Kit Separation (DrumSep) for?

Separation & stem split Choose five WAV stems—kick, snare, toms, hi-hat and cymbals—or six WAV stems—kick, snare, toms, hi-hat, ride and crash.

Q: What tasks can Drum Kit Separation (DrumSep) help with?

Separation & stem split; Choose five WAV stems—kick, snare, toms, hi-hat and cymbals—or six WAV stems—kick, snare, toms, hi-hat, ride and crash.; Drum remixing, replacement, transient editing, practice tracks, and drum-focused analysis.; A full song or an already isolated drum stem. Full-song mode first extracts drums with AudioSep-Hive and forwards only that stem to MDX23C DrumSep.; Extract the official five- or six-part drum layout from a full song.; Split an isolated drum recording without rerunning six-stem separation

Q: Which models and related terms are associated with Drum Kit Separation (DrumSep)?

MDX23C DrumSep · 5 stems, MDX23C DrumSep · 6 stems, AudioSep-Hive (full-mix preprocessing), Official 5-stem checkpoint, Official 5-stem config, Pinned immutable mirror used by TelkNet, Official 6-stem checkpoint, Official 6-stem config

35 credits/useFile inputMax 50 MBEstimated 180-900 seconds

Input

A full song or an already isolated drum stem. Full-song mode first extracts drums with AudioSep-Hive and forwards only that stem to MDX23C DrumSep.

Supported formats

mp3wavflacoggm4aaacaifaiffmp4movmkv

Output

Choose five WAV stems—kick, snare, toms, hi-hat and cymbals—or six WAV stems—kick, snare, toms, hi-hat, ride and crash.

Use case

Drum remixing, replacement, transient editing, practice tracks, and drum-focused analysis.

Under Maintenance

This tool is under maintenance. Please try again later.

Models, parameters, and sources

MDX23C DrumSep is an open, locally deployable drum-element separator. Full-mix input first uses AudioSep-Hive; drum-only input goes directly to the selected official five- or six-stem checkpoint.

MDX23C DrumSep · 5 stemsMDX23C DrumSep · 6 stemsAudioSep-Hive (full-mix preprocessing)Official 5-stem checkpointOfficial 5-stem configPinned immutable mirror used by TelkNetOfficial 6-stem checkpointOfficial 6-stem configPinned checkpoint bytes used by TelkNetAudioSep-Hive (full-mix drum preprocessing)Pinned AudioSep-Hive checkpointPinned Hive source

MDX23C DrumSep · 5 stems

The production route uses the pinned official MDX23C DrumSep five- and six-stem checkpoints through the MSST loader at 44.1 kHz.

Official 5-stem checkpoint Official 5-stem config Pinned immutable mirror used by TelkNet

MDX23C DrumSep · 6 stems

The production route uses the pinned official MDX23C DrumSep five- and six-stem checkpoints through the MSST loader at 44.1 kHz.

Official 6-stem checkpoint Official 6-stem config Pinned checkpoint bytes used by TelkNet

AudioSep-Hive (full-mix drum preprocessing)

Pinned AudioSep-Hive checkpoint Pinned Hive source Hive

Scores apply only to the cited dataset, metric and model version. Leading this table does not mean a current world record or a result from this service. Capabilities and tests can justify prioritizing models without comparable scores.

Benchmark chart: Paper and benchmark notes

Paper and benchmark notes

Pinned MSST pretrained-model table

Values come from the linked public source. Row labels identify the test set or checkpoint; models without a matching result are omitted.

Test setting	Model	SDR (dB) ↑	Detailed metrics
DrumSep test · official five-stem checkpoint	MDX23C DrumSep · 5 stemsTelkNetMSST table · kick SDR is the headline value	16.66	Snare11.53Toms12.33Hi-hat4.04Cymbals6.36
DrumSep test · official six-stem checkpoint	MDX23C DrumSep · 6 stemsTelkNetMSST table · kick SDR is the headline value; ride and crash are reported jointly	14.54	Snare9.79Toms10.63Hi-hat3.19Ride + crash6.08

Paper and benchmark notes

A Semantically Consistent Dataset for Data-Efficient Query-Based Universal Sound Separation

ICML - 2026

Extract an open-vocabulary sound with AudioSep-Hive, FlowSep-Hive or CLAPSep, or isolate a source that matches a 10-second reference with Banquet.

arXiv:2601.22599

Official / repositories

Official 5-stem checkpoint Official 5-stem config Pinned immutable mirror used by TelkNet Official 6-stem checkpoint Official 6-stem config Pinned checkpoint bytes used by TelkNet Pinned AudioSep-Hive checkpoint Pinned Hive source

Papers / technical notes

Hive

Parameter guide

Model profile: selects the checkpoint and inference path used for this task.
Input route: Full mix first extracts drums with AudioSep-Hive; Drums only sends an isolated drum stem directly to MDX23C DrumSep.
Official MSST test-shift count. Higher values cost more time and may improve stability.
Official test-time augmentation. It increases processing time and is disabled by default.
Full-mix pipeline: AudioSep-Hive extracts drums, then the selected pinned MDX23C DrumSep checkpoint returns five or six drum parts. The intermediate drum stem is not delivered.
Outputs: five-stem mode returns kick, snare, toms, hi-hat and cymbals; six-stem mode returns kick, snare, toms, hi-hat, ride and crash.

Use cases

Extract the official five- or six-part drum layout from a full song.
Split an isolated drum recording without rerunning six-stem separation
Prepare editable drum parts for a DAW or practice mix

Workflow

1Choose Full mix for a song, or Drums only for an isolated drum stem
2Full mix uses AudioSep-Hive to extract drums; Drums only skips this stage.
3MDX23C DrumSep runs locally and returns the selected five or six final drum stems.

Before you submit

Choose Drums only only when the upload contains isolated drums
The AudioSep-Hive preprocessing output is temporary and is not returned.
Expect some bleed when cymbals, room sound, and other instruments overlap

Related use-case guides

Working on a similar task? These guides list the matching tools, model boundaries, sources, and FAQs for each scenario.

Drum Kit Separation (DrumSep) · online tool and model guide

Quick feedback

Include the tool name, steps and the problem in your feedback. Screenshots can help us locate it.