Music Professional: Stem Source Separation Expert (Proof of Experience Required)

What you'll do

You listen to short segments of real music and name what is playing: the specific instruments and the instrument families present. Those labels become training data for source separation, the technology that pulls a finished mix apart into its separate parts.

The listing explains why. Current open-source separation models (it names HTDemucs, Open-Unmix, MDX-Net and the RS-RoFormer family) split a mix into only 4 to 6 stems, typically along the lines of vocals, drums, bass and "everything else". A real mix can hold dozens of instruments, and that coarse split limits remixing, music education, transcription and music information retrieval. The project wants models that can separate far more finely, and they need human-labelled examples to learn from.

The listing does not say how long each segment is, how many you label per hour, which genres are covered, or what labelling interface or taxonomy is used.

Who qualifies

The title makes one requirement explicit: proof of experience as a music professional. The listing does not define what counts as proof. The work itself calls for:

• A trained ear that can pick individual instruments out of a dense mix, including ones buried under others
• Knowledge of instrument families and the difference between similar-sounding instruments (a viola against a cello, a Rhodes against a Wurlitzer)
• Consistency over many short, repetitive judgments

Mixing and mastering engineers, session musicians, arrangers, orchestrators and music teachers are the natural fit. No working language or country requirement is stated.

What it pays

$17 an hour, the standard rate for Invisible's general voice, audio and evaluation work. For work that requires proof of professional experience that is low; the same rate applies to listings with no professional bar at all.

The contract terms say payment is weekly via Stripe or Wise, and that Invisible cannot support H-1B or STEM OPT contractors.

Worth knowing

Good:

• Genuinely interesting for musicians: it feeds a known limitation in separation tools you may already use
• Short segments, so the work splits into small pieces around your own schedule
• Remote, no location requirement stated

Less good:

• $17 an hour for a professional skill
• The listing does not say what counts as proof of experience
• Listening-heavy, repetitive work; ear fatigue is real over long sessions
• A resume upload (about 5 minutes) and an unpaid qualification step (about 30 minutes) before any paid work
• No minimum amount of work; independent contractor terms with no benefits

About this listing

Read on the signed-in Meridial marketplace on 26 September 2026, posted about a month before we read it. The project rationale, named models, rate and contract terms are the listing's own; the notes on who fits and the pay comparison are ours. See Invisible Technologies.