English · Français · Website

_{Free · macOS & Windows · Ableton Live 12 + Max for Live · all versions}

Your mix sounds muddy and you don't know which track to touch? MixCoach listens to every track of your Ableton Live project and tells you which one to fix, how to do it in Ableton, and why. It's made for producers who never learned mixing or music theory. You don't need to know what "masking" or "headroom" means: MixCoach explains it like a friend sitting next to you in the studio. When you're in a hurry, a ⚡ button applies the fix for you.

Most beginner mixes have the same handful of problems: a muddy low end, a kick and a bass fighting each other, a harsh lead, a bass that falls apart on a club system. Spectrum analyzers show you these problems as curves, which only helps if you already know how to read them.

MixCoach goes one step further:

• 🎯 It compares your tracks with each other, not just the master. It doesn't stop at "too much 300 Hz": it tells you "the Pad brings 60% of the mud between 250 and 500 Hz, cut it there".
• 🗣️ It explains everything in everyday words, with images ("like two people talking at the same time") and always says what you gain by fixing it.
• 🛠️ It speaks Ableton: every tip is a list of steps with real device and parameter names (EQ Eight → Bell → 120 Hz → −3 dB).
• ⚡ It can fix things for you: one click recenters a pan, raises a fader or adds an EQ Eight / Utility with the right settings. Cmd+Z / Ctrl+Z undoes it.
• 🌍 It speaks 18 languages.
• 🔓 It's free, open source, and runs offline. No account, no server, no AI service: just math on your audio.

Measured in Live on a real house track. When two tracks are loud in the same place, you only really hear the louder one. For each of the three main clashes, the EQ MixCoach suggests frees up the shared zone (hatched):

Requires Ableton Live 12 Suite (or Standard + Max for Live), on macOS or Windows.

1. Download MixCoach.zip (always the latest version) and unzip it.
2. Copy the MixCoach folder (2 devices + mixcoach.js) to
   • macOS: ~/Music/Ableton/User Library/Presets/Audio Effects/Max Audio Effect/
   • Windows: Documents\Ableton\User Library\Presets\Audio Effects\Max Audio Effect\
   • Not sure? In Live's browser, right-click User Library → Show in Finder / Explorer.
3. (Optional, but magic) copy the MixCoachAuto folder to User Library/Remote Scripts/ (same User Library as above), restart Live, then pick MixCoachAuto as a Control Surface in Settings → Link, Tempo & MIDI. Sensors will now add themselves to every track.
4. Otherwise, drag MixCoach Sensor onto each track (at the end of the chain) and MixCoach onto the Master.
5. Press play. After a few seconds, the panel fills up. Click ⤢ to open it in a big window, and pick your language in the top-right menu.

Optional, command line (Python). You don't need it to use MixCoach in Live.

git clone https://github.com/xeewi/mixcoach.git && cd mixcoach
python3 -m venv .venv && .venv/bin/pip install -r requirements.txt

.venv/bin/python mixcoach.py my_mix.wav                                  # mix before mastering
.venv/bin/python mixcoach.py my_master.wav --stage master                # master for streaming (−14 LUFS)
.venv/bin/python mixcoach.py my_master.wav --stage master --target club  # loud club master (~−8 LUFS)
.venv/bin/python mixcoach.py my_mix.wav --reference pro_track.mp3        # compare with a pro track you love

Export from Ableton with File → Export Audio/Video, Normalize off. An HTML report opens in your browser.

💡 Use --reference with a professional track in the style you're aiming for. The generic target curve is only a rough guide: even pro tracks deviate from it by ±5 dB.

In Ableton, export with Rendered Track: All Individual Tracks, then:

.venv/bin/python tracks.py ~/Music/Exports/my_project/

The master file is recognized automatically. Group tracks (Drums, etc.) are detected and skipped, so their sound isn't counted twice.

1. Measure. Each sensor runs a bank of 28 third-octave filters (31 Hz → 16 kHz), plus stereo and peak meters. It takes the track's fader, pan and mute into account. No audio leaves the device: it only shares about 40 averaged numbers, once per second.
2. Compare. The brain (m4l/brain.js) looks at who occupies which frequency, where the mix deviates from a target curve measured on 18 electronic tracks (house, techno, minimal, Creative Commons licensed), and which pairs of tracks mask each other.
3. Explain. Each finding comes with the facts, a plain-language "Why?", what you gain by fixing it, the steps in Ableton, and, when possible, a ⚡ automatic fix.

The Python tools (mixcoach.py, tracks.py) and the Max for Live brain share the same rules and the same sentences. A test checks that both give exactly the same advice on the same audio.

English · Français · Español · Deutsch · Italiano · Português (Brasil) · Nederlands · Polski · Русский · Українська · Türkçe · العربية · हिन्दी · Bahasa Indonesia · 日本語 · 한국어 · 简体中文 · 繁體中文

• In Live: language menu in the top-right corner of the panel. The default is your system language, and the sensors follow your choice.
• Reports: --lang ja (default: system language, or the MIXCOACH_LANG environment variable).
• Ableton device and parameter names (EQ Eight, Low Cut, Threshold…) stay in English, exactly as they appear in Live.

Spotted a clumsy translation? Translations live in i18n/, one JSON file per language. See Contributing.

mixcoach.py           analysis of a whole mix / master  →  HTML report
tracks.py             track-by-track analysis of an export  →  HTML report
i18n.py, i18n/        translation system + 17 translation files (French is the source language)
m4l/
  brain.js            the advice engine used inside Ableton (same rules as tracks.py)
  mixcoach.js         Max script: measurements, Live API, ⚡ fixes
  ui.html             the panel page (rendered by Max's jweb)
  build.py            builds the .amxd devices  (--install copies them to your User Library)
  MixCoachAuto/       Remote Script that adds the sensors automatically
run_tests.py          runs every test

.venv/bin/python run_tests.py            # all tests, with a summary
.venv/bin/python m4l/build.py --install  # rebuild + install the devices (refuses if a test fails)
.venv/bin/python i18n.py                 # list translatable sentences, check every language

On Windows, use .venv\Scripts\python instead of .venv/bin/python.

Publishing a new version: git tag v0.3.0 && git push --tags. GitHub Actions runs the tests, builds the devices and publishes MixCoach.zip on the Releases page. The download links always point to the latest version.

Tests are plain Python scripts built on synthetic audio with known defects ("a stereo bass at 55 Hz must be flagged"). They also check that the JavaScript brain matches the Python one, that the Max script answers ⚡ clicks, and that all 18 languages are complete. They run on every push via GitHub Actions.

Contributions are very welcome, especially:

• 🌍 Translations. Native speakers: please review your language! Each file maps the French sentence to its translation. Keep the {placeholders} and the Ableton names, then run python i18n.py to check.
• 🪟 Windows testing. Windows support is brand new: if you use Live on Windows, tell us how it goes!
• 🎚️ Mixing knowledge. Better thresholds, better explanations, genre-specific target curves.

Please run python run_tests.py before opening a pull request.

• Live 12.3+ is needed for ⚡ fixes that add a device (EQ Eight, Utility). Fader and pan fixes work on any Live 12.
• Clashes are judged on the average spectrum: two sounds that never play at the same time may still be flagged.
• The generic target curve is the average of 18 electronic tracks. Sub and highs vary so much between pro tracks that, without a reference, MixCoach only flags outliers. For finer advice, or another genre, use --reference.
• In Live, the panel judges roughly the last minute played: loop the busiest part (the drop) before reading the advice.
• Windows: supported, and the analysis is tested on Windows by CI, but the devices haven't yet been tried inside Live on a Windows PC. Feedback welcome.

MIT. Free to use, modify and share.

Built on NumPy, SciPy, soundfile and pyloudnorm (ITU-R BS.1770 loudness). Key detection uses the Krumhansl-Kessler profiles.