Plugin Boutique, mastering chart
Held the top spot for 11 consecutive months — February through December 2024 — out of every paid and free mastering plugin on the market.
An engineering brief, not a marketing page.
11 patents. 18M tracks. 3:42 per master. Built by the DSP team your DAW was built by.
Gemix Audio is a co-pilot for mixing engineers — an adaptive mastering engine trained on professionally finished work, not a black box that promises to replace your ears. What follows is the engineering.
- 11 Granted DSP patents (USPTO, 2021–2024)
- 18M Tracks in Reference-Match training set
- 3:42 Average session, across 4.1M runs in 2024
- 9 PhDs in audio signal processing on staff
CHAPTER 01 — BRIEF
Why we built Gemix Audio: the 90% solution in 4 minutes.
What we built
A co-pilot that gets you to the last ten percent of a release-ready master before your coffee cools. It analyzes your stems, learns your sonic taste from reference tracks you supply, and locks the result to a delivery target — Spotify, Apple Music, YouTube, club, vinyl — without you touching a fader.
Why it exists
Because the bottleneck in independent production is almost never the master itself — it's the hour between "the mix is done" and "the file is on the distributor." Gemix collapses that hour into a render queue, so an engineer can spend their attention on the take, not the loudness target.
What it is not
A replacement for a senior mastering engineer on a tentpole release. A click-and-forget button that ignores arrangement, genre nuance, or a client's brief. A general-purpose AI music generator. If you need any of those, this is the wrong product — and we would rather you know now.
CHAPTER 02 — RECEIPTS
Recognized by the people who measure these things.
MusicTech Awards, NAMM 2024
"Best AI Mastering Tool 2024." Judged against nine shortlisted competitors across price, fidelity, latency, and transparency of the underlying model.
Independent label co-signs
Used on catalogue releases by Anjunadeep, Monstercat Uncaged, and mau5trap — three labels whose A&R process is famously unforgiving on master quality.
Pro studios shipping through Gemix
From project rooms in Kreuzberg to Dolby Atmos rooms in LA. 2,400 paying studio accounts and 180+ active label integrations as of Q1 2026.
- PLUGIN BOUTIQUE #1
- MUSICTECH AWARDS '24
- NAMM SHOWCASE
- ANJUNADEEP
- MONSTERCAT
- mau5trap
CHAPTER 03 — THE STACK
Three systems that do the heavy lifting.
Gemix is not one model. It is a deterministic pipeline of three production-engineered systems, each independently testable, each trainable, each with its own USPTO filings behind it.
01 / Reference-Match AI v3
18 million tracks. 41 genres. One taste function.
The training corpus is not scraped from streaming. It is curated from commercially released masters across electronic, hip-hop, pop, indie, jazz, classical, and 35 other genres — each track paired with its pre-master mix where recoverable, so the model learns the delta, not the absolute.
- 18.0M professionally finished masters in the training set
- 41 genre-conditioned heads in the model
- 4.1M production sessions measured against the corpus in 2024
02 / Stem-Aware Rebalancer
Re-balance vocals, bass, and drums — before limiting.
The only AI mastering pipeline that treats stems as a first-class input. When you hand the engine your vocal, bass, and drum stems, it identifies masking between kick and bass fundamentals, applies a psychoacoustic tilt to keep the vocal above the snare, and only then hands the summed bus to the limiter.
- 3-stem minimum, 8-stem maximum input
- Patent 11,023,991 granted 2024 — stem-aware rebalancing
- −1.4 dB average improvement in vocal-to-bus ratio A/B'd against mono-bus masters
03 / Psychoacoustic Loudness Engine
Loudness without the "loudness wars" artifact.
A short-term psychoacoustic model — not an RMS target — drives the limiter. Per-platform delivery presets are LUFS-locked, but the work happens in the model's perceptual domain, which is why a Gemix master on streaming reads as louder and cleaner than a same-LUFS bus compressed with a conventional limiter chain.
- −14 Spotify · −16 Apple Music · −14 YouTube
- −9 Club preset · DR14 Vinyl preset (context note: streaming-optimized releases may prefer platform presets)
- 6 psychoacoustic patents filed 2022–2024
CHAPTER 04 — PEDIGREE
The team: 38 people, 9 PhDs, 4 from your DAW's R&D bench.
PhDs in audio signal processing
Filter design, perceptual coding, real-time DSP, machine learning on spectrograms — every senior engineer on the audio stack has a doctorate in something you can build a plugin on.
Ex-Native Instruments · Ableton · iZotope
Four of our principal engineers shipped commercial audio products at the three companies that built the DAW, the sampler, and the spectral editor you already use. They joined because this is the brief they always wanted to work on.
Granted USPTO patents
Adaptive EQ (USPTO 11,023,991), psychoacoustic loudness modeling (11,184,402), stem-aware mixing (11,302,775), and eight more across the 2021–2024 filing window. Every patent is publicly searchable and indexed.
Berlin HQ · Los Angeles studio
Founded 2020 in Berlin-Kreuzberg; the LA office is a working mastering room with a Neve Genesys and a Bricasti M7 used to ground-truth every algorithmic release.
Gemix Audio GmbH — Kohlfurter Strasse 14, 10999 Berlin, Germany. Series A 2023 (US$14M, Index Ventures, SoundCloud founder Alexander Ljung participating). [email protected] · +49 30 5557 0428.
CHAPTER 05 — A PEER'S NOTE
“Gemix doesn't try to do your job. It does the eighty percent of mastering that is mechanical — loudness target, tonal balance against a reference, format conversion — and leaves the twenty percent where taste matters to you. As a co-pilot, that's the most honest framing an AI audio tool has shipped with.”