or click to browse
Deckmatics is a 100% client-side audio processing tool for personal, educational, and technical use. Ensure you have legitimate rights to processed tracks. Public distribution, commercialization, or exploitation of remixes, acapellas, and derivative works requires prior copyright authorization. Processing runs entirely in your browser. No files are uploaded to the cloud.
Deterministic batch renamer for DJ library tracks based on metadata tags and audio analysis.
Inspect metadata tags, detect messy names, and batch rename automatically or with manual inline edits.
Drop your audio tracks here or click to browse
Lossless (WAV, FLAC, AIFF) and Lossy (MP3, AAC, M4A) · Local Processing Only
Identify identical and alternate quality duplicate tracks in your collection.
or click to browse
Detect exact duplicates (SHA-256) and acoustic matches with quality recommendations
Convert Rekordbox, Traktor, VirtualDJ and Serato collections 100% locally.
or click to browse
Supports Rekordbox .xml, Traktor .nml, and VirtualDJ .xml
Transmute DJ USB drives between Pioneer, Denon Engine OS and Serato in-place without copying or duplicating audio files.
Transmutes existing DJ USB drives in-place without copying or duplicating audio files.
Connect or select your USB drive to inspect its DJ topology.
Pioneer, Engine Library and _Serato_ folders will be checked to provide cross-compatibility without touching your audio tracks.
Drag and drop audio files here or use the buttons below to import tracks.
or click to browse
High-fidelity audio recording for microphones, DJ USB sound cards or system audio with peak and RMS metering
Split any song into four stems — vocals, drums, bass and instrumental — with a multi-engine MIT-licensed local architecture: instant Deckmatics Pure DSP (0 MB download) or studio-grade MIT-licensed neural networks (SCNet INT8, SCNet-large INT8, and Mel-Band RoFormer Kim). Download every stem as 16-bit WAV at the original sample rate: processing happens 100% locally on your machine with zero server uploads.
Both Deckmatics Pure DSP and MIT-licensed neural engines return the four channels: vocals, drums, bass and instrumental, with a real-time multitrack mixer and immediate 16-bit WAV download controls.
Choose between the instant speed of Deckmatics Pure DSP (0 MB download, CPU spectral filtering) or the surgical precision of MIT-licensed neural models (SCNet INT8 4-stems or Mel-Band RoFormer 2-stems, WebGPU/WASM accelerated). All approved engines are released under permissive MIT licenses.
Pure DSP combines dual-median HPSS, Wiener masks, and a 250 Hz crossover. Neural models run modern architectures such as SCNet INT8 and RoFormer with WebGPU hardware acceleration or CPU WASM for surgical track isolation.
The results window adds transport with playback, a master volume and mute, solo and volume per stem with each channel waveform. Every download is 16-bit PCM WAV at the original sample rate of the file.
What a DJ separates stems for, before a gig or before opening a session.
Isolate the vocals of a track to build a mashup, a transition edit or a vocal chop without waiting for the label to publish an acapella. Export the vocals WAV and take it to your DAW or to your decks.
Mute and solo across the four stems and you hear the track without vocals, or the isolated voice, in seconds — to check whether a transition works before you play it live. It is a mixer preview, not an export.
Download drums, bass, instrumental and vocals as 16-bit WAVs and work them in your DAW: remix, practise basslines on headphones, cut samples or deliver the stems to a collaborator.
Load an isolated stem into the Audio Analyzer and read on the spectrogram what survived: vocal bands in the vocals stem, transients in drums, sub energy in bass. Separation and analysis share the same browser.
What the code guarantees and what this block will not promise.
Separation runs entirely inside your browser combining Web Audio API, Web Workers, WebGPU (with automatic WebAssembly fallback), and CPU DSP: no server receives your audio, there is no queue, and no telemetry is gathered. Your music never travels across the network.
Published here so nothing surprises you in the middle of a job.
The engine caps a file at 480 seconds (8 minutes) to protect browser memory; anything longer fails with a clear error. There is also no batch queue: one file is separated at a time.
Downloads are exclusively 16-bit PCM WAV at the original sample rate; nothing here exports MP3, FLAC, AAC, OGG or 24-bit. If you need another format, use the Audio Format Converter.
Pure DSP delivers instant spectral approximation with zero downloads, while MIT-licensed neural models download their open weights once locally for studio quality. Both coexist under permissive MIT licensing.
The decoded file and the four stems stay in browser RAM while you work. For Neural AI, WebGPU accelerates inference; if unavailable, the engine falls back to CPU via WebAssembly.
From how the models work to mashups and the rights behind derivative works.
Technical guide to source-separation metrics and spectral analysis algorithms (SDR, HPSS, Wiener), AI noise reduction, reverberation removal and audio restoration (Academy guide in Spanish).
How to prepare acapellas: stem separation, de-noise, de-reverb, pitch and tempo, chops and stutters in the booth (Academy guide in Spanish).
A practical route to a live mashup: acapella and instrumental, stems on stage, three or four decks and cue points ready before you play (Academy guide in Spanish).
Learn how to interpret audio spectrograms: frequency axes, harmonics, dynamic range in dBFS, and how to spot low-pass cuts before playing live.
What you need before and after separating a track.
Load an isolated stem and read on the STFT spectrogram what survived, with BPM, Camelot key and LUFS loudness.
Convert the 16-bit WAVs to MP3, FLAC or AIFF with control over sample rate, bit depth and dither.
Separate a track into four lanes inside the project, silence the original and export the mix.
Cut the acapella or slice a stem into sample-accurate loops without leaving the browser.
