Campaigns AI Training Multi-track audio — stems & session …
AI Training Open On-demand request

Multi-track audio — stems & session recordings

for AI data buyers via Troveo

The brief

Buyers are sourcing multi-track audio — the separated stems behind a finished recording. Source separation, mixing assistants and music models all need real stems, and almost nobody licenses them. If you record music, podcasts or voice sessions, your project folders are the product.

What a good submission looks like:

  • All stems exported from time zero so they're sample-aligned and equal length
  • Clear names: 01_kick.wav, 07_lead-vocal.wav, 09_room-mic.wav
  • A stereo reference mix of the same session included in the zip
  • Dry-ish stems preferred — light processing is fine, but avoid heavy master-bus limiting

Rights: every sound must be yours — your performance, your recording, your composition (or one you hold full rights to). No loops, sample packs or session players without written clearance.

Requirements

Multi-track audio
  • WAV (or FLAC) stems at 44.1/48kHz, 24-bit, all exported from time zero and equal length
  • Descriptive stem names + a stereo reference mix in every zip
  • One zip per song/session, with a short README (instrumentation, BPM, key if known)
  • 100% your own recordings — no third-party loops, samples or uncleared players
  • No master-bus heavy limiting on stems

What your footage trains

Aligned stems are the ground truth for source-separation and music-generation models — the model learns what a mix is made of by seeing the mix and its parts together.

Campaign FAQ

More like this

Similar campaigns