Matchering matches four measurements and ships its own limiter
🎚️ Open Source Audio Matching and Mastering
At a glance
- What is it?
- Matchering is a GPL-3.0 Python library, container image and ComfyUI node that masters a target track to match a reference. Its mechanics are narrow: four measurements are matched, a brickwall limiter written for the project does the final stage, and the last release was in 2022.
- Who is it for?
- Matchering fits a producer who wants album tracks to sound consistent with each other and has a reference track they trust, and who is willing to treat a reference as a starting point rather than a target. Three things to know before you commit.
- Can I use it commercially?
- Yes, with conditions. GPL-3.0 is a copyleft licence: if you distribute software that includes it, you must release that software's source code under the same licence. Running it internally without distributing it does not trigger that obligation.
- Is it still maintained?
- Yes. The repository last received commits 39 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.
Star history
- sergree/matchering2,713 ★ · +61
| Repository | 2026-10-04 | 2026-11-15 |
|---|---|---|
| sergree/matchering | 2.7k | 2.7k |
61 Stars in the last 42 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
Two files in, and four measurements are matched
The whole idea fits in two inputs. The target is the track you want to master, the one you want to sound like something else. The reference is a track you like, described in the readme as a wet popular song. The algorithm matches them and hands back the mastered target, matched to the reference on four specific measurements rather than on an aesthetic judgement: root mean square loudness, frequency response, peak amplitude, and stereo width. Each of those four is linked in the readme to a definition, which is a small sign the project treats them as precise quantities. Everything else the tool does, including a preview option and per-result formatting, is arranged around that single comparison, and the examples directory shows five ways of calling it: a basic script, one that edits the configuration, one that produces advanced results, one that writes custom text output, and one that renders a preview.
A 4 GB machine, Python 3.8, and a system audio library
The library requirements are stated in one line and they are not trivial. A four gigabyte machine with Python 3.8 or higher is required, which tells you the resampling and analysis stage is memory hungry rather than streaming. The audio dependency is a two-layer chain: the Python library for sound files sits on top of a system library that has to be present underneath. On Windows and macOS the system library installs itself, and on Linux you are told to use the distribution's package manager, with an Ubuntu example given:
sudo apt update && sudo apt -y install libsndfile1The readme then handles two more gaps that Linux users hit: the pip module is not present by default on some distributions, and pip itself has to be installed. FFmpeg is listed as optional and scoped to one thing only, loading MP3 files.
Five dependencies, floor versions only, no ceilings
The requirements file is five lines long and every one is a lower bound with no upper bound, which is an unusual amount of freedom for audio software. Two are the numerical stack. One is the sound file library. One is a separate resampling library rather than the one inside the numerical stack, which is a deliberate choice if you care about resampling quality. The fifth is a statistics package, which is the interesting entry: nothing in the four-measurement description obviously needs regression, so its presence says something about how the matching is fitted. The packaging metadata is also old-school. There is no pyproject file, the install requirements are read out of that text file by the setup script at build time, and the readme is passed in as the long description with its content type declared as markdown. The declared audience is three different groups at once, from developers to desktop end users to another unnamed audience, and the topic list runs from artistic software through to analysis and conversion, which is a fair description of a tool that both masters and inspects.
The limiter was written for this, and its quality test is published
The second major version shipped its own limiter rather than borrowing one, and the readme says so directly: an open source brickwall limiter of the author's own implementation replaced whatever the previous version used. The limiter lives in a subpackage of the library. Two things make this unusual for a project of this size. First, there is a document in the repository dedicated to the limiter's quality test, so the claim that it works is inspectable rather than asserted. Second, a third party has published a JavaScript reimplementation of the same limiter design in a plugin format, which is the sort of thing that only happens when a limiter's character is distinctive enough to be worth copying.
Four audiences, four entry points, one routing paragraph
The installation section is a routing table rather than a procedure, and it is the clearest piece of writing in the readme. If you are a music producer or an audio engineer, use the desktop application or the ComfyUI node. If you are an AI mastering startup, use the container image, and there are four separate documents for it, one per platform plus one for updating an installed image. If you are a developer, use the Python library. The ComfyUI node is a separate repository by a different author, and the desktop route runs through a third-party vocal removal application where the tool appears in a nested menu. There is also a command line application with its own enhanced fork.
The review is quoted with the author's caveat attached
The readme quotes a third-party reviewer and the quoting matters. The claim is that the tool beat every other automated system tried, ranking third out of twelve judged entries behind two professional mastering engineers, on what the reviewer describes as a study with several hundred entries. The author's note immediately under it says the result depends on carefully selecting a proper song as the reference. That is an unusual thing to print next to a favourable quote, and it is also the load-bearing caveat for anyone using the tool: the reference selection is doing work. The same page links three commercial services where you can try the approach without installing anything, which is a sensible way to find out whether your reference choice works before you commit to the pipeline.
No release since 2022, and the repository is not archived
Version history is where this project stops being current. The three most recent releases are two patches published on the same day in April 2022 and a third in October 2022. The last push to the default branch is dated 8 July 2026, so the branch has moved a long way past the newest tag, and the version string in the packaging metadata still reads the 2022 release. The repository is not archived. What that combination means practically is that installing from the package index and cloning the branch give you different code, and only one of them has a version number you can cite. The repository also keeps a press sheet, a supporters list, an overview of the log codes the tool emits, and a document of the author's own thoughts, which is more documentation than most projects of this size keep outside the readme.
Editorial conclusion
Matchering fits a producer who wants album tracks to sound consistent with each other and has a reference track they trust, and who is willing to treat a reference as a starting point rather than a target. Three things to know before you commit. Read the reference as a choice, because the one external review quoted in the readme makes its point about beating other systems conditional on picking a good reference song. Check the format floor, since the library needs Python 3.8 or newer, a four gigabyte machine and a system audio library that Linux users install separately. And note that the newest release is from 2022 while the branch has moved since, so a pinned install and a clone are not the same thing. If you need a mastering engineer, nothing here replaces one.
Frequently asked questions
what is matchering
An open source audio matching and mastering tool, published as a Python library, a containerised web application and a ComfyUI node. You give it two files: a target track you want to master and a reference track whose character you want it to take. It returns the mastered target matched to the reference's loudness, frequency response, peak amplitude and stereo width.
how to install matchering 2.0
As a library, install the package with python3 -m pip install -U matchering, or with python -m pip on Windows. You need Python 3.8 or newer and a 4 GB machine. On Linux you must also install the system sound library yourself, and some distributions need the pip module installed separately. FFmpeg is optional and only enables MP3 loading.
What does Matchering actually match?
Four measurements: root mean square loudness, frequency response, peak amplitude, and stereo width. The mastered target comes out with the same values as the reference you supplied, and each of the four is defined and linked in the readme rather than left as a term of art.
Which Matchering interface should I use?
The readme routes by audience: music producers and audio engineers to the desktop application or the ComfyUI node, an AI mastering startup to the container image, and developers to the Python library. There is also a separate command line application and an enhanced fork of it, and the tool is bundled inside a third-party vocal removal application.
How current is the Matchering project?
The most recent release is 2.0.6, published on 19 October 2022, and the last push to the default branch is dated 8 July 2026, so the branch has moved well past the last tag. The repository is not archived, and the version recorded in the packaging metadata still reads the 2022 release.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/sergree-matchering)Star history chart for your README
[](https://hysenlabs.com/projects/sergree-matchering)Was this useful?