Pitch shifting changes the audio, but it does not establish that AI fingerprints have been removed. A detector may respond differently to the transformed file. To attribute a change to pitch, compare a pitch-only treatment with the original and an unchanged export control.

Pitch shifting is not the same as changing speed

Audacity’s Change Pitch changes pitch while retaining tempo. Changing playback speed is a different operation: it can change pitch and duration together. Name the actual operation when recording a detection experiment.

Whole-mix transposition affects the song’s key as well as its voice and instruments. It is not equivalent to removing a file tag. If keeping the original key matters, set that requirement before evaluating any score.

A controlled pitch-only experiment

The following is a proposed protocol, not a table of completed test results.

  1. Choose known-origin source tracks and retain the original WAV files.
  2. Export an unchanged control through the same software and format.
  3. Create pitch-only variants from each original; record semitones, algorithm, quality mode and whether formants are preserved.
  4. Keep tempo, gain, EQ and export settings fixed.
  5. Analyze the original, control and variants using the same detector, recording date, verdict and score.
  6. Check whether the resulting key, voice and transients still suit the track.

A control that changes the score already shows that export or analysis differences need investigation. Do not attribute that difference to pitch.

What would count as useful evidence?

For every row, retain the exact input and output, their hashes, processing settings and full reports. Include every tested setting, not just the lowest score. Keep different exports of the same composition grouped rather than counting them as independent songs.

Repeat unchanged files to establish whether the observed change exceeds run-to-run variation. A lower number on one song supports an observation on that file; it does not establish a transferable recipe for Suno, Udio or another generator.

Why combined treatments cannot isolate pitch

If a process changes pitch, level, bandwidth and stereo timing together, its output score measures the combination. It cannot tell you how much pitch contributed. Compare a treatment with and without the pitch step before making that attribution.

No matched pitch-only files and reports accompany this article, so no numerical benefit is claimed. The tempo experiment tests a different variable. For processing with an output report today, see Demorphus Cleaner and plans.

Prepared with AI assistance. Sources below.

Sources and documentation