DAW project file datasets: what exists, what a session carries

Dataset	Holds	Licence	Commercial training
Mix Evaluation Dataset	10 songs and 180 mixes made by students in Pro Tools 10 (one group in Logic Pro X), with the DAW sessions and parameter settings for most mixes and listening-test ratings	Raw tracks of six of the ten songs CC BY 4.0; the other songs are copyrighted and not online	Per track six songs CC BY 4.0
MixologyDB (MixParams)	Effect parameter values per track as JSON, read from the Mix Evaluation sessions; no session files, no automation	MIT on the repository	Yes parameter values only
MixAssist	7 recorded mixing sessions between expert and amateur producers, 431 conversational turns with the DAW playback; audio and text, no project files	CC BY 4.0	Yes no project files
Cambridge-MT (Mixing Secrets)	Raw multitrack audio of hundreds of songs; no project files	Educational purposes as agreed with the contributors; no formal licence	No ask each contributor
Slakh2100	2,100 MIDI songs rendered with sample libraries: MIDI plus stems, no project files	CC BY 4.0	Yes synthetic renders
CoT Music Production DAW	7,000 question and answer pairs about FL Studio and Ableton Live; text only, no project files	MIT	Yes text only

Each name links to the page the line was read from. Licences change and pages move. Check the source before you rely on a line, and tell us at mikas@ximaudio.com if one is out of date.

What exists

There is no open corpus of producers' sessions in Ableton Live, Logic Pro, FL Studio or Cubase. What can be downloaded is one of three things: a few dozen sessions made for a mixing study, with raw tracks of six songs under CC BY 4.0 and the rest copyrighted; parameter values or conversations extracted from such sessions, without the project; or audio without any project at all. The one set that ships thousands of aligned MIDI and stems, Slakh, is rendered from Lakh MIDI with sample libraries and has no project behind it. The parsers that exist for the native formats were written against files their authors collected themselves.

What a session carries

Tracks with the names the producer gave them, grouped into buses, with volume, pan, sends and sidechain routing.
The plugin chain on every track, with its parameters: the instrument, the preset, the EQ, the compressor, the reverb send.
The notes the instruments played, as MIDI clips tied to the sound that played them.
Clips on the timeline and the sections of the arrangement, so the structure of the song comes from the arrangement itself.
Automation: every parameter that moves over time, and when.
Tempo, key, time signature and the DAW version that saved the file.

Formats

An Ableton Live set (.als) is gzipped XML and can be read with a script. An FL Studio project (.flp) is a binary chunk format. A Logic Pro project (.logicx) is a macOS package around a binary project file. A Cubase project (.cpr) is binary too. A session dataset is only usable with a parser per format, or with a parsed description next to every file.

What XIM Audio holds

9,468 DAW sessions sold by their producers under terms that name machine learning: 4,343 Ableton, 3,111 Logic, 1,531 FL Studio, 366 Cubase, measured 4 Nov 2026. Every session is parsed into one JSON schema across DAWs: tracks, plugin chains, volume, pan, sends, buses, sidechain, clips, notes, automation and sections. 3,571 songs come as paired sets, the session next to every lane printed from it, the MIDI it played and the masters. 378 tutorial sessions add the producer building the track on camera, with the finished Logic project. Modern music made in the DAW, pop to trance. Compositions and lyrics stay out of scope, and no delivery carries a title or an artist name. See the sessions set.