Role Description

This role involves contributing real music recordings paired with detailed note-level MIDI annotations for an AI training dataset. You will supply either existing or newly created tracks, ensuring each recording is 3 to 4 minutes long and accompanied by a separate annotation track for every instrument. Each note must be annotated with instrument, pitch, onset, and offset, with timing accuracy within 10 to 20 ms. The preferred workflow is capturing audio and MIDI simultaneously in your DAW, but existing multitrack archives or manual annotation are also accepted. Genres beyond pop, rock, and Western classical are especially valuable, and a diverse range of instrumentation is encouraged. Full guidance on format and labeling will be provided before you begin.

Deliverables

• Provide 20 unique music recordings, each 3–4 minutes in length, split across two quality tiers (Standard and Enhanced note-off accuracy).
• For each recording, deliver one lossless audio file (WAV or FLAC, 44.1 kHz or higher) and one annotation file (MIDI/JSON) per instrument.
• Clearly indicate the tier (Standard or Enhanced) and capture method (performed, programmed, or hybrid) for each instrument track.
• Ensure all annotations are human-verified and not generated by automatic transcription tools.
• Supply a brief note on your available genres, instrumentation, and production methods, along with relevant work samples.