Music Producer: Real Recordings with MIDI Annotations (Remote)

We are looking for musicians, producers and audio engineers to supply real music recordings paired with note-level MIDI annotations, as part of an AI training dataset project.

The work:

Each deliverable is one real music recording, plus a separate annotation track for every instrument in that recording, with each note carrying instrument, pitch, onset and offset. Annotations must reflect what was actually played, with timing matching the performance to within 50 ms and ideally 10 to 20 ms.

The most practical route is capturing audio and MIDI in the same take, for example performing into Logic or Ableton via a microphone using a keyboard or piano, electric guitar, drum kit, saxophone, trumpet etc, recorded acoustically and as MIDI at the same time.

Methodology: how the data can be produced

There is more than one valid route, and we are open to whichever suits your setup:

Recording a live instrument into your DAW - Capturing audio and MIDI in the same take, for example performing into Logic or Ableton via a microphone using a keyboard or piano, electric guitar, drum kit, saxophone, trumpet etc, recorded acoustically and as MIDI at the same time. The acoustic or electric audio and the MIDI are captured in the same take, so timing lines up natively.
Existing multitrack or DAW project archives. Sessions of your own work where separate instrument stems and MIDI or note data already exist and can be exported, as long as it is from real instrument or vocal recordings.
Manual annotation or score alignment. Transcribing or aligning notes against an existing real recording, with human verification of timing and instrument labels.

Audio rendered from MIDI using virtual instruments is not suitable, and annotations must not be generated by running an automatic transcription tool over the audio.

Note: synthetic/digital instruments will not be accepted, real music/instrument recordings only.

Each recording is delivered as one lossless audio file (WAV or FLAC, 44.1 kHz or higher) plus one annotation file (MIDI, CSV or JSON). Full guidance and the agreed format will be provided, so you will know exactly what is expected before you start.

Scope and opportunity:

We begin with a small paid sample to validate timing and format, followed by a pilot batch from roughly 300 recordings upward. If it goes well, there is potential for a significantly larger project to follow, so we are looking to build ongoing working relationships rather than fill a one-off job.

Rate:

Please quote your price per recording, or per finished minute of annotated audio, stating which you are quoting. Most competitive pricing will receive the majority of the work.

What we are looking for:

Working musicians, producers, composers or audio engineers
A setup that can capture real audio and MIDI together, or an existing archive of suitable material
Genres beyond pop, rock and Western classical are especially valuable, as are recordings including singing
Comfortable working to a defined delivery format
Rights to the material you supply, including permission for AI training and commercial use

To apply:

Send us your price and availability, a short note on what you could record or already hold (genres, instrumentation, and how the audio and annotations would be produced), along with a few examples of relevant work.

For what period of time will the freelancer be needed?
Best part of a week