// Product · Nomad Listen

The fingerprint engine
that hears the derivative.

Nomad Listen indexes 78 million reference tracks and matches incoming audio against them in under a second — including the derivatives that classical fingerprinting was never built to find: tempo-shifted edits, acoustic fan covers, lyric-swapped mashups, and the 9-second cuts that drive short-form catalogue revenue.

// 78M reference tracks // 4 parallel detectors // 1.6e-7 FP rate // sub-1s latency

The problem Listen was built for

Most fingerprinting in production today was designed in the Shazam era — to find an exact, noisy copy of a known reference in a few seconds of audio. That problem is essentially solved. The unsolved problem is everything that isn't an exact copy: the 130-BPM sped-up edit of a 100-BPM original, the acoustic fan cover with different instrumentation, the short-form mashup that splices nine seconds of your song under nine seconds of someone else's. Each of those represents catalogue revenue that should be flowing back to rights holders, and almost none of it does.

Nomad Listen exists to surface those derivatives — at planetary scale, with false-positive rates low enough that a working rights team will actually read the output.

// Architecture

Four detectors.
An agreement rule.

Detector 01

Tempo-invariant melodic

Maps the melodic contour onto a tempo-normalised pitch curve. Matches sped-up edits, slowed-down remixes, and key-shifted versions back to their references.

Detector 02

Harmonic-progression

Models the chord-change sequence at the bar level. Matches covers in different instrumentation, including acoustic re-arrangements that throw off timbre-based systems.

Detector 03

Lyric-aware

Runs an audio-to-phoneme encoder and matches phoneme runs against transcribed reference lyrics. Catches lyric-swapped mashups where 60% of the words are the original.

Detector 04

Timbre-only (Shazam-class)

Classical constellation-fingerprint matching — the reliable baseline that catches noisy exact copies in clubs, radio rips and live-stream archives.

// Agreement rule

A match is surfaced only when at least two of the four detectors agree above threshold.

This is what keeps false positives tractable. Any single detector run alone produces enough noise to drown the rights team — agreement between two is rare enough to read, common enough to recover meaningful revenue.

// In numbers

Production parameters.

78MReference tracks indexed
<1sMatch latency, 8-second window
1.6e-7FP rate, conservative band
94%Recall on cover-test corpus
// Working pattern

How a label actually uses Listen

01

Ingest the catalogue once.

Provide the WAV/FLAC masters (or high-bitrate AAC if masters aren't available); we fingerprint every track across all four detectors and store the index. One-time pass, takes about 4 hours for a 6,000-track catalogue.

02

Subscribe to the derivative stream.

We continuously scan major and minor platforms — short-form social, DSP catalogues, fan upload sites, broadcast archives, podcast networks — and emit a stream of candidate derivatives that match against your fingerprints.

03

Triage and claim.

The rights team reviews surfaced candidates, dismisses the ones that aren't real derivatives (typical false-discovery is < 3%), and dispatches claims through your existing rights workflow — Content ID, MLC dispute, neighbouring-rights agency, or a direct platform takedown.

04

Reconcile the recovery.

Listen feeds matched derivatives into Nomad Rights, which tracks every claim through to payment and surfaces the ones still in limbo. By month 6, most labels see derivative-driven revenue running at 2–5× whatever they were recovering before.

// Read more

Engineering writing on Listen.

All sound posts →
Audio Research

Fan covers, sped-up edits, short-form mashups: detecting derivatives at planetary scale

8 min · Audio
Royalties

Why your royalty statement is wrong, and how AI reconciliation finally finds the missing money

7 min · Rights
// 30-minute listen

Send us a small sample.
We'll tell you what's leaking.

A 1,000-track sample is enough for us to estimate your derivative coverage gap. We come back with a number and three specific tracks where derivatives are running uncovered.