A useful Suno fingerprint remover test records every outcome across different tracks. Build a small pilot from your catalogue, keep originals and outputs paired, and assess sound quality separately from the detector verdict before processing more songs.

Choose a pilot that resembles your releases

Do not select only the song that already gave your best result. Include a dense mix, an exposed voice, an instrumental and the formats you regularly use. These are suggested categories for your own pilot, not a statistically representative benchmark. If your releases are all piano instrumentals, select variation within that material.

Record generator versions when known. Keep older downloads separate from recent ones and original WAV files separate from MP3 conversions. Related exports of one song are not independent examples of broad generator coverage.

Keep a log that separates outcomes

Track ID | Generator/version | Source format | Original report
Output filename | Processing date | Output report | Listening decision
Analysis error, if any | Follow-up needed

Use stable track IDs to avoid confusing songs with the same title. Keep reports beside the exact exports. Record the score, verdict and analysis date where available; enter “unavailable” for an analysis failure, never zero.

Assess listening before reading the new score if possible. This helps you notice a quieter vocal or softer cymbals without the number shaping your expectations.

Define success before counting it

For example, require both the target verdict from the named detector and audio you would use. Report how many submitted tracks meet both conditions, how many fail listening review and how many cannot be analyzed. Keep retries visible.

A hypothetical pilot of twelve submissions with eight acceptable target-verdict outputs, two rejected on listening and two unavailable analyses has eight successful submissions out of twelve. This illustrates bookkeeping, not a Demorphus benchmark.

Use the pilot to plan the next batch

Evaluate Cleaner on your pilot, then compare plans and allowances for the remainder. Put exceptions in a review queue rather than applying repeated transformations blindly.

The Fourier artifacts paper studies model-associated spectral patterns. It does not establish a universal setting or disclose SH Labs’ implementation. Describe what happened on your files; follow the single-song workflow for individual comparisons.

Prepared with AI assistance. Sources below.

Sources and documentation